Risk Journal 34

Journal
The Capco Institute Journal of Financial Transformation
Recipient of the Apex Awards for Publication Excellence 2002-2011
#34
Cass-Capco Institute Paper Series on Risk

03.2012
Journal
Editor
Shahin Shojai, Global Head of Strategic Research, Capco
Advisory Editors
Cornel Bender, Partner, Capco Christopher Hamilton, Partner, Capco Nick Jackson, Partner, Capco
Editorial Board
Franklin Allen, Nippon Life Professor of Finance, The Wharton School, University of Pennsylvania Joe Anastasio, Partner, Capco Philippe dArvisenet, Group Chief Economist, BNP Paribas Rudi Bogni, former Chief Executive Officer, UBS Private Banking Bruno Bonati, Strategic Consultant, Bruno Bonati Consulting David Clark, NED on the board of financial institutions and a former senior advisor to the FSA Gry Daeninck, former CEO, Robeco Stephen C. Daffron, Global Head, Operations, Institutional Trading & Investment Banking, Morgan Stanley Douglas W. Diamond, Merton H. Miller Distinguished Service Professor of Finance, Graduate School of Business, University of Chicago Elroy Dimson, BGI Professor of Investment Management, London Business School Nicholas Economides, Professor of Economics, Leonard N. Stern School of Business, New York University Michael Enthoven, Former Chief Executive Officer, NIBC Bank N.V. Jos Luis Escriv, Group Chief Economist, Grupo BBVA George Feiger, Executive Vice President and Head of Wealth Management, Zions Bancorporation Gregorio de Felice, Group Chief Economist, Banca Intesa Hans Geiger, Professor of Banking, Swiss Banking Institute, University of Zurich Peter Gomber, Full Professor, Chair of e-Finance, Goethe University Frankfurt Wilfried Hauck, Chief Executive Officer, Allianz Dresdner Asset Management International GmbH Michael D. Hayford, Corporate Executive Vice President, Chief Financial Officer, FIS Pierre Hillion, de Picciotto Chaired Professor of Alternative Investments and Shell Professor of Finance, INSEAD Thomas Kloet, Chief Executive Officer, TMX Group Inc. Mitchel Lenson, former Group Head of IT and Operations, Deutsche Bank Group Donald A. Marchand, Professor of Strategy and Information Management, IMD and Chairman and President of enterpriseIQ Colin Mayer, Peter Moores Dean, Sad Business School, Oxford University John Owen, Chief Operating Officer, Matrix Group Steve Perry, Executive Vice President, Visa Europe Derek Sach, Managing Director, Specialized Lending Services, The Royal Bank of Scotland ManMohan S. Sodhi, Professor in Operations & Supply Chain Management, Cass Business School, City University London John Taysom, Founder & Joint CEO, The Reuters Greenhouse Fund Graham Vickery, Head of Information Economy Unit, OECD Norbert Walter, Managing Director, Walter & Daughters Consult
Cass-Capco Institute Paper Series on Risk

Ideas at Work
6 IT Complexity: Model, Measure, Manage, and Master Peter Leukert, Andreas Vollmer, Mat Small, Peter McEvoy IT Complexity Metrics How Do You Measure Up? Peter Leukert, Andreas Vollmer, Bart Alliet, Mark Reeves
Part 2
103 Markets for CCPs and Regulation: Considering Unintended Consequences Serge Wibaut, D Sykes Wilford 119 What Have We Learned from the 2007-08 Liquidity Crisis? A Survey Hamid Mohtadi, Stefan Ruediger 129 Making Sense of Asset Prices: A Guide to Required Yield Theory, Part 1 Valuing the Stock Market Christophe Faugre 149 Our Understanding of Next Generations Target Operating Models Andreas Andersen, Nicolas Faulbecker 155 The First Line of Defense in Operational Risk Management The Perspective of the Business Line Udo Milkau, Frank Neumann 165 Optimal Bank Planning Under Basel III Regulations Sebastian Pokutta, Christian Schmaltz 175 A Risk Measure for S-Shaped Assets and Prediction of Investment Performance Qi Tang, Haidar Haidar, Bernard Minsky, Rishi Thapar 183 ILLIX A New Index for Quantifying Illiquidity Tim Friederich, Carolin Kraus, Rudi Zagst 195 How Homogeneous Diversification in Balanced Investment Funds Affects Portfolio and Systemic Risk Rocco Ciciretti, Raffaele Corvino 211 Breaking Through Risk Management, a Derivative for the Leasing Industry Sylvain M. Prado, Ram Ananth 219 A Financial Stress Index for the Analysis of XBRL Data Amira Dridi, Silvia Figini, Paolo Giudici, Mohamed Limam
11
Part 1
19 Toward a Bottom-Up Approach to Assessing Sovereign Default Risk: An Update Edward I. Altman, Herbert Rijken International Liquidity Provision and CurrencySpecific Liquidity Shortages Richhild Moessner, William A. Allen The Failure of Financial Econometrics: Stir-Fry Regressions as a Con Job Imad Moosa Lehman A Case of Strategic Risk Patrick McConnell Explaining Credit Default Swaps Pricing for Large Banks nci tker-Robe, Jiri Podpiera Investing in Private Equity Capital Commitment Considerations Sameer Jain Moving the OTC Derivatives Market to CCPs Manmohan Singh Cultural, Political, and Economic Antecedents of Country Risk in Sixty-Two Countries Moshe Banai
31
43
51 63
77
83 89
Ideas at Work
Ideas at Work
IT Complexity: Model, Measure, Manage, and Master

Peter Leukert Chief Information Officer, Commerzbank AG Andreas Vollmer Chief Architect, Commerzbank AG Mat Small Partner, Capco Peter McEvoy Partner, Capco
IT complexity: a new metric to evaluate the impact of change

What if complexity was a discreet, measurable metric rather than a discretionary, ambiguous term? Could a complexity metric reshape the decisions and activities of a CIO? The information technology (IT) field is filled with quotes and anecdotes about organizations trying to contain, explain, and manage complexity. Often these words reflect an inclination to assume that if complexity exists, we should try to reduce it. As computer science pioneer Alan Perlis once said, Fools ignore complexity. Pragmatists suffer it. Some can avoid it. Geniuses remove it. Pure genius, however, rarely exists. And with our world in a state of constant change, we need to master, not remove complexity. The reduction of complexity can reach diminishing returns, and the existence of complexity can allow greater business flexibility. Without a doubt, an appropriate level of complexity is necessary to maximize return. Leading 6 financial institutions recognize the need to
acknowledge, embrace, and harness complexity. They understand that the many layers and connected dependencies of todays IT environments cannot be managed with simple heuristic, or experience-based, approaches alone, but instead require objective, quantifiable measurement of their origins and effects. This paper describes a groundbreaking initiative underway by Germany-based Commerzbank and consulting firm Capco to develop a model for measuring IT complexity in the financial services industry. Importantly, individual financial institutions will be able to derive value from this new model through immediate application to their existing environment, while reaping long-term benefits through contribution to an industry database of results.
for complexity was available to complement experience and the typical financial metrics, how might those decisions change? Historical parallels from other industries help illuminate the possibilities. A mid-20th century auto assembly line manager could not have imagined the possibility of Six Sigma and Lean techniques delivering five-nines production quality. The pharmaceutical industry has made stunning advancements in statistical analysis to prove drug effectiveness. Securities trading offers an illuminating example in financial services. Trading has come a long way since a century ago when curbstone brokers traded small and speculative stocks outside the New York Stock Exchange. Back then, brokers did not stand on the street with a computer in hand. They had no notion of mathematically valuing an instrument and quantitatively evaluating the risk associated with price movements. Today, algorithmic trading using Greeks is an instantaneous activity that leverages risk metrics as a core component of calculation. What was viewed as hard to measure
The power of quantitative metrics

Today, CIO decisions are tied almost exclusively to experience and straightforward financial analyses, such as project costs and estimations. Some companies also apply scenario analysis to technology decisions for risk management purposes. But if a quantitative metric

10 years ago has been made routine by the availability of models and data.
issue, such as reengineering the system. CIOs with a lot of experience in complex deci-
The cost impact Cost containment has been one of the most pressing issues for CIOs in recent years. IT is continuously expected to increase efficiency. CIOs of large financial institutions preside over significant IT budgets often U.S.$1-2 billion and, by extension, substantial complexity. The decisions they make can influence direction for many years to come and have potentially large financial consequences. Clearly, complexity drives cost. One or both of
Similarly, IT leaders need a transparent, objective framework to augment the experience they bring to decision making. Such a framework can aid in articulating and communicating decisions both across and outside the organization, an increasingly important requirement as corporate directors and regulators expand their scrutiny of technology decisions.
sions are arguably in a better position, as they can subjectively associate future change with historical observations. However, if CIOs supplement their experience, native intuition, and traditional financial analyses with an IT complexity metric, they will have all of the pieces needed to make critical decisions that benefit both their IT department and the enterprise.
How decisions related to complexity are made today and could be made tomorrow
Complexity has become an overworked word in the IT field. It is used with increasing frequency in publications and elsewhere to describe how management views decision-making broadly, as well as in relation to specific functions across and within industries, such as IT and risk management. All CIOs have biases based on experience. Their understanding of the familiar forms the basis for the gut decision they inevitably must make. The introduction of a complexity metric allows for an exploration of alternatives that experience may not shed light on.
How a complexity metric fits into the CIO agenda

Whether thought about consciously or not, every decision a CIO makes influences overall IT complexity. Complexity in turn drives up cost, makes quality harder to achieve and affects flexibility. Thus the core dimensions that dominate a CIOs agenda cost, quality and flexibility are all influenced by complexity, and IT leaders have to make daily trade-offs between them. Consciously taking complexity into account, via a complexity metric, will improve the effectiveness of CIO decisions and make them easier to explain and defend.
the following examples will be familiar to any CIO:
Technological complexity a new software package offers great functionality to the business. Unfortunately, it runs on technology that requires different skills from those of the existing IT employees. Consequently, people need to be retrained, additional upgrades need to be made for the new technology components, and external expertise has to be contracted for.
Interface complexity a major new development is well underway. Then integration testing starts. Suddenly a number
Consider IT complexity to make more informed decisions
When faced with decisions related to delivering IT change, CIOs must categorize the situation, usually resulting in a subjective measure of complexity: We will replace system X because the economics indicate lower operational cost and a positive payback. Would that exercise not be more productive, and might that scenario unfold differently, with the introduction of a model-driven complexity measure? Based on the IT complexity metric in our business case, we have determined that if we replace system X, the resulting system will likely be just as complex because of underlying process and integration issues, and we will never realize a positive return. This type of thought innovation then allows for an examination of other methods to address the
Figure 1 Complexity framework
Decisions regarding IT change
i.e., projects, architecture, transformation
IT complexity
Cost Dimensions
Function Interface Data Technology
Quality Flexibility
Ideas at Work
of interfaces are affected and need to be tested that were not part of the original scope. There are unexpected effects on the downstream systems, and testing expense climbs exponentially. Complexity drives cost, and is also a good indicator of unexpected cost increases. Modeling how complexity increases in the two examples above would have alerted IT managers to the cost impact. Based on this exercise, managers might mitigate cost increases or at least make more informed choices. The quality impact The issue of quality has also become more
important on the CIO agenda. There are clear correlations between quality and long-term operational costs. And the maturation of technologies has led IT end-users to be less forgiving of quality issues. Complexity correlates to quality intuitively, as the following examples highlight:
example. In more complex environments it is harder to identify the root cause of an error and more risky to fix without introducing other errors. With a complexity metric, IT executives can manage quality more effectively. The flexibility impact The relationship between complexity and flexibility often forms a vicious circle. Flexibility gained by adding new functionality and new technology quickly drives up complexity. The more complex the system landscape, the harder and more risky changes become. As a result, flexibility decreases, as these examples show:
Technology and interface complexity production stability is one aspect of system quality. A more complex application landscape is more prone to suffer from incidents after deploying changes because there can be unexpected, and thus untested, interdependencies with systems that were not supposed to be affected by the change.
Functional and interface complexity error analysis and resolution is another
Data complexity when implementing
2010_Q3
Stand: 11.10.2010
Complexity metric - quarterly aggregated report

Percental coverage of QFB1 and QFI1 Implementation Scope (QFI1): Degree of functional requirements and "as is" solution architecture
Number of weighted interfaces (QII1)
Infrastructure Requirements (QT1)
Percental coverage of QT1
Number of IT products per functional group/ cluster
Ratio QFI1 / QFB1: Implementation Quotient in percent
Ratio internal interfaces / total number of interfaces per functional group/ cluster (II2)
Architecture Cluster Functional Group
# IT products 334
PCov QFB1 & QFI1 3.6% 33.8% 36.4% 40.0% 30.0% 42.9% 0.0% 38.1% 32.8% 23.5%
16.7%
QFI1 27.8 72.5 7.3 2.6 8.3 8.5 0.0 11.1 34.7 70.8
2.7
QFI1 Av. 2.3 1.5 1.8 1.3 1.4 1.4 n/a 1.4 1.7 1.8
1.4
QFI1 Std. Dev. 0.8 0.5 0.8 0.0 0.1 0.0 #DIV/0! 0.1 0.6 0.7
0.1
Impl.Quotient 25.7% 20.5% 22.1% 20.0% 23.7% 16.7% n/a 26.4% 19.4% 26.2%
18.0%
PCov QII1 15.3% 86.3% 63.6% 100.0% 85.0% 100.0% 100.0% 85.7% 85.9% 59.0%
83.3%
QII1 110.9 842.5 90.7 19.0 184.6 95.4 28.4 105.4 319.1 723.0
88.7
QII1 Av. 2.2 7.0 13.0 3.8 10.9 6.8 7.1 5.9 5.8 7.4
8.9
QII1 Std. Dev. 1.9 7.6 16.3 2.1 9.6 5.2 12.7 7.6 5.0 9.7
6.5
QT1 322.3 299.2 19.0 14.0 45.2 45.7 11.3 34.7 129.3 279.0
27.3
PCov QT1 44.6% 93.5% 81.8% 100.0% 95.0% 100.0% 100.0% 95.2% 92.2% 68.1%
83.3%
II2 n/a 25.9% 0.0% 0.0% 7.0% 15.2% 0.0% 15.2% 7.7% 14.0%
3.9%
II2 Av. n/a 6.4% n/a n/a n/a n/a n/a n/a n/a 10.3%
n/a
II2 Std. Dev. n/a 6.8% n/a n/a n/a n/a n/a n/a n/a 11.1%
n/a
Top-level view on aggregated complexity metrics indicates areas for further analysis: High implementation quotient values for investment banking related groups i.e., implementation of investment banking applications is more complex than comparable commercial banking applications (assuming an equal number of functions are covered)
Sales Customer Relationship Management (CRM) Documents & Archiving Services Sales Internet Banking & Intranet Services Sales Basis Services Sales Management Sales Support (internal & external) Business Operations Custody Service / Corporate Actions Custodian Bank Services Fund Management Trading Product & Services Pricing Trading Front End Trading Front Office Loans Management Treasury / Liquidity Planning Rating Transaction Processing Accounting Order Processing Service Billing Entry Routing Channeling Trading Back Office Loans Processing Customer Information Provision Provision Processing Treasury Processing Securities Processing Payments Processing Group Steering Compliance / Revision Reg. Reporting External Accounting Cost / Profit Risk Controlling /-Management Treasury / Liquidity Planning Data Warehouse Reporting Services Enrichment / Consolidation Metadata Management Staging
139 11 5 20 14 4 21 64 166
12
10 0 11 8 50 42 4 18 280 42 8 22 17 10 31 4 12 78 56 219 35 34 0 23 56 10 23 4 9 2 8
20.0% #DIV/0! 18.2% 50.0% 16.0% 31.0% 50.0% 22.2% 21.4% 31.0% 25.0% 22.7% 17.6% 70.0% 12.9% 0.0% 33.3% 11.5% 23.2% 33.3% 45.7% 41.2% #DIV/0! 21.7% 32.1% 30.0% 30.4% 0.0% 44.4% 50.0% 25.0%
2.8 0.0 4.4 7.1 19.1 20.8 5.8 5.3 103.8 20.2 2.7 7.7 8.8 12.9 5.4 0.0 10.1 15.1 20.9 131.9 30.2 20.2 0.0 8.5 37.2 4.2 9.5 0.0 5.4 1.4 2.7
1.4 #DIV/0! 2.2 1.8 2.4 1.6 2.9 1.3 1.7 1.6 1.4 1.5 2.9 1.8 1.4 n/a 2.5 1.7 1.6 1.8 1.9 1.4 #DIV/0! 1.7 2.1 1.4 1.4 n/a 1.4 1.4 1.4
0.0 #DIV/0! 1.1 0.8 0.8 0.6 0.1 0.2 0.7 0.6 0.1 0.6 0.1 0.8 0.1 #DIV/0! 0.8 0.7 0.6 0.7 0.8 0.5 #DIV/0! 0.7 0.8 0.0 0.1 #DIV/0! 0.1 #DIV/0! 0.1
31.1% n/a 73.3% 33.8% 36.7% 18.4% 27.6% 25.2% 18.8% 18.2% 14.2% 15.4% 28.4% 25.3% 6.1% n/a 40.4% 29.6% 16.7% 24.6% 29.0% 21.0% n/a 20.2% 22.4% 20.0% 20.7% n/a 25.7% 23.3% 14.2%
60.0% #DIV/0! 36.4% 62.5% 30.0% 78.6% 100.0% 61.1% 86.4% 83.3% 87.5% 90.9% 64.7% 100.0% 93.5% 100.0% 91.7% 79.5% 94.6% 64.4% 68.6% 76.5% #DIV/0! 47.8% 64.3% 90.0% 78.3% 75.0% 66.7% 50.0% 100.0%
11.8 0.0 6.8 13.0 120.5 324.0 27.0 46.9 2633.6 506.8 105.8 179.6 121.7 139.6 319.7 32.1 88.7 672.0 467.6 723.8 91.0 135.5 0.0 57.7 157.9 27.7 533.3 89.4 145.0 3.9 295.1
2.0 #DIV/0! 1.7 2.6 8.0 9.8 6.8 4.3 10.9 14.5 15.1 9.0 11.1 14.0 11.0 8.0 8.1 10.8 8.8 5.1 3.8 5.2 #DIV/0! 5.2 4.4 3.1 29.6 29.8 24.2 3.9 36.9
1.8 #DIV/0! 0.9 0.9 8.8 13.4 4.4 8.2 13.3 16.0 12.6 12.3 14.0 16.8 10.3 6.2 15.2 13.4 12.4 8.6 3.9 6.9 #DIV/0! 4.5 3.9 2.0 38.3 49.9 21.3 #DIV/0! 48.4
12.0 0.0 15.8 20.7 85.8 73.0 11.3 17.7 637.2 87.7 16.7 43.8 40.3 24.7 84.5 7.2 32.2 160.5 139.7 359.0 66.0 50.7 0.0 42.3 86.0 24.2 41.5 3.5 16.0 4.2 17.8
60.0% #DIV/0! 54.5% 100.0% 62.0% 76.2% 100.0% 38.9% 82.9% 76.2% 87.5% 63.6% 82.4% 100.0% 96.8% 100.0% 100.0% 74.4% 91.1% 74.9% 91.4% 70.6% #DIV/0! 87.0% 69.6% 90.0% 69.6% 25.0% 66.7% 100.0% 87.5%
21.4% 13.3% 0.0% 6.7% 5.2% 8.1% 0.0% 33.8% 43.3% 9.9% 0.0% 21.4% 5.2% 1.1% 9.9% 2.0% 6.3% 34.9% 16.3% 17.4% 19.8% 6.2% 7.8% 14.3% 11.9% 2.9% 7.5% 0.0% 6.4% 0.0% 2.1%
n/a n/a n/a n/a n/a n/a n/a n/a 10.7% n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 10.5% n/a n/a n/a n/a n/a n/a 2.1% n/a n/a n/a n/a
n/a n/a n/a n/a n/a n/a n/a n/a 10.9% n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 6.1% n/a n/a n/a n/a n/a n/a 3.0% n/a n/a n/a n/a
Functional group, securities processing, is characterized by a high number of applications and interfaces with other functional groups
Data warehouse staging with only eight applications shows highest value for quantity of interfaces which reflects the number of source systems
Rise against the quarter before Descent against the quarter before
Figure 2 Complexity dashboard, a sample view

compliance regulations, many banks are hindered by the fact that they do not have all customer data stored consistently in one place in the enterprise.
significant correlation between the complexity indicators and both cost indicators, such as maintenance spend, and quality indicators, such as the number of incidents occurring during production. The complexity indicators of the model cover four dimensions: function, interfaces, data, and technology.
the complexity of legacy architecture over time. For other problems and decisions, the individual dimensions need to be considered for example, the trade-off between functional scope and interface intensity in defining application domains. To date, we have measured time series of statistically validated complexity metrics for Commerzbank. We believe that, ultimately, a single company cannot develop a comprehensive approach to complexity modeling on its own. Models evolve when multiple organizations contribute data and experience. For example, think about how risk measures improve when a financial institution has more years of defaulted bonds to model. Bottom line: the more companies that use the complexity model and contribute to a standard database over time, the higher the benchmarking quality.
Functional and data complexity product introduction, considered to be a hallmark of flexibility, is more difficult in a highly complex environment because of the interdependencies and effects of systemic changes.
Function one driver of functional complexity is the functional scope of an application cluster. The sum of weighted use cases serves as the figure to measure this. In each use case, a weight factor is assigned, from one for simple use cases, such as changing one data field, like address, to four for use cases with involved logic. Use cases have been classified by expert judgment. Other indicators measure the functional redundancy and standard conformity of the solution architecture.
In summary, a complexity metric will increase transparency and provide insights that help in managing cost, quality, and flexibility. A metric will also help uncover intelligence associated with every strategic decision, improving appreciation of the delicate intricacies associated with the CIO agenda.
How do you measure complexity?

Measurement is the first step that leads to control and eventually to improvement. If you cannot measure something, you cannot understand it. If you cannot understand it, you cannot control it. If you cannot control it, you cannot improve it. The observation offered above by the leading performance and quality expert H. James Harrington highlights the central role that measurement plays in addressing complexity. In its current state, the Commerzbank/Capco complexity model applies to the application landscape: single applications, application clusters, application domains, and the entire application landscape. The model consists of several complexity indicators covering the relevant dimensions of application complexity. We have statistically validated these complexity indicators through quantitative research using Commerzbank data on approximately 1,000 applications over three years. A tacit inverse correlation exists between flexibility and complexity that is difficult to quantify. However, we have established a statistically

Interfaces interface complexity is determined through the sum of weighted interfaces. The weights are calculated by type of interface, such as API, file exchange, database view, and broker web service. This indicator shows the interface intensity. The ratio of internal to external interfaces is another relevant indicator. Data data complexity is measured by the number of database objects in an application cluster. Technology technology complexity is monitored through a number of drivers, including business criticality and prescribed time for recovery of applications under consideration. Another technology indicator is the variety of operating systems employed in the application cluster.
Immediate application and benefits

The complexity model, as it stands today, can be applied to important IT decisions at any financial institution after some calibration and data gathering. The model can also help educate IT leaders. And it should foster dialog between the IT department and the business units it serves.
Decision and trending analysis IT executives will immediately be able to use the complexity model to analyze the impact of decisions and perform trending analyses. This will help them understand how their decisions might affect the future cost, quality, and flexibility of IT projects. To achieve these results, a phase of data gathering and calibrating the model is necessary. Depending on the availability of relevant information, such as data on the interfaces of applications under consideration, this process will take from a few weeks to a couple of months. Again, this type of analysis will become more precise and robust over time as collected data is enriched. 9
Each indicator measures a relevant aspect of complexity. In our experience, these measures become most powerful if considered in conjunction. Sometimes it is useful to aggregate them into one overall complexity score to simplify matters, such as tracking
Ideas at Work
Educating IT leaders the numbers produced through the use of the complexity model will vary across organizations. Also, similar to other analytics-driven metrics, the definition of thresholds for the model will be set over time as both IT leaders and business units gain better appreciation of its merits. Each scenario modeled, coupled with the passing of time to observe the outcomes of a metric-driven decision, will enable IT executives to test their own hypotheses and build a stronger foundation of knowledge to support future decisions.
partnerships and create new opportunities to achieve measurable results.
metric could serve as a process measurement or a heuristic to measure the efficacy of business process changes. These applications could allow for the quantification of business complexity and provide a more holistic view of changes needed to drive the CIO agenda. Emboldened by insights into IT, organizations might consider measuring the complexity of business architecture as well. IT decisionmaking on the business side would likely benefit from more explicit and quantitative consideration of complexity.
Expansions and possibilities

While the complexity model at present is focused primarily on IT applications, it is not difficult to envision how it might evolve and be extended to a broader set of technology decisions. Similar to the evolution of other analytics-driven metrics as they were applied to different solution spaces, the complexity model will offer a number of possibilities in the not-too-distant future, including:
Bridging the gap between business and technology in addition to informing decision-making within the IT organization, a complexity measure should help IT executives to better explain and defend their decisions to the business. An objective measurement of complexity will be a monumental step toward addressing the concerns of business unit managers and other executives regarding IT systems development. CIOs will no longer need to rely on soft, subjective explanations supported by experience. Instead, they will be able to produce a quantifiable metric that provides a new level of transparency.
IT architecture perhaps most interesting to IT leaders is the notion of looking for complexity across not just the application stack but the overall architecture. This will support conclusions regarding the impact of change on the entire infrastructure, including networks and other aspects, and is perhaps one of the most natural extensions of the complexity model. While this advancement will require expansion of inputs to the model, such as indicators on infrastructure and middleware components, the output will provide IT executives with a more rounded perspective.
Conclusion
The notion that you cannot manage something that you cannot measure has been engrained in business for many years. The creation of a complexity model presents an extraordinary new opportunity to understand the levers that drive cost, quality, and flexibility. Importantly, such a model becomes stronger as more data is added to it. We hope this article has helped stimulate your interest in exploring complexity and how you can harness it to improve your organization. We also invite you to join the conversation as we continue to explore complexity issues. Among topics we plan to address:
IT processes extensive work has been done over time on IT supply chain modeling. In the future, incorporating this thought leadership could produce a complexity metric that is oriented to the process dimension of IT decisions, which could aid in decisions relating to the software development life cycle for example, determining the right allocation of time and resources to each leg of the SDLC. This concept could be extended further to include complexity aspects of sourcing decisions.
To date, the complexity model leverages data from a single firm: Commerzbank. Ultimately, the model will improve over time as data from more institutions allows for more accurate benchmarking. Making the model available to others will enable benchmarking across multiple organizations, increasing the credibility of the metric based on comparative data points from across the financial services industry. This increased reliability will be invaluable to IT executives. A complexity metric will help CIOs articulate short- and long-term implications of initiatives, costs over time, impacts on existing operations and the business as a whole, and strategic implications and justifications of high-risk projects. By articulating complexity in this way, 10 CIOs will be able to strengthen intercompany
How is complexity calculated, and how would an organization embark on a complexity measurement program?
How does the complexity metric influence the strategic decision process? What is envisioned for the executive dashboard for complexity measurement and scenario modeling, and how would the complexity metric be used in concert with other IT metrics?
Business processes it is easy to intuit a relationship between a business process and the technical complexity needed to support it. In many instances, the technology cannot be simplified without corresponding simplification of the business process. Business process simplification is evident in many initiatives, from reengineering to Six Sigma. In the future, a complexity
How is complexity modeling different across various corporate paradigms (midversus large-size, new versus old firm, single line versus multiline business, institutional versus retail)?
IT Complexity Metrics How Do You Measure Up?

Peter Leukert Chief Information Officer, Commerzbank AG Andreas Vollmer Chief Architect, Commerzbank AG Bart Alliet Partner, Capco Mark Reeves Partner, Capco
IT complexity is so important that learning through failing is no longer satisfactory given that this form of complexity so clearly and harshly impacts the cost, delivery quality, and flexibility of the IT landscape. We now need a robust and functioning model for complexity measurement and mitigation. This paper describes how Commerzbank, in partnership with Capco, is deriving such a model. We also examine how it can be developed over time to provide a cross-industry standard. Specifically, we will be looking at:
Introduction and background

IT complexity can and should become a leading indicator for IT performance management Typically, the CIO scorecard contains the following key performance indicators: costs, flexibility, and quality. Sadly, these are lagging indicators whose current values typically result from choices made many years previously. Most experts agree that IT complexity is a key hidden variable. As such, it is a leading indicator supporting the prediction of costs, flexibility, and quality that should (must) be used in todays IT management decisions. IT complexity is, however, very difficult to master, since it creeps up, growing incrementally (but can only be reduced by a substantial management intervention), is not purely local (it occurs locally in many different applications and interfaces, but the impact is only perceptible on the aggregate level), and cannot until now with our model be measured (a valid measure for IT complexity does not exist elsewhere, either in academia or in business). And, as we all know: what cannot be measured, cannot be managed.
Given these issues, Commerzbank and Capco have worked out a practical solution, by creating an IT complexity model and management decision tool. The tool is based on a number of specific and measurable IT complexity indicators, organized in a structured IT complexity analysis framework. Story so far origins and progress In the previous article in the ideas at work section of this edition of the Journal, entitled IT Complexity: Model, Measure, Manage, and Master, Leukert et al. outlined the importance of understanding IT complexity for technology leaders in the financial services market. The key conclusions reached include the following:
The
shape
of
the
model
how
Commerzbank, with Capcos input, models complexity at the moment.
The key components a deconstruction of the key components of a rigorous understanding of complexity the indicators covering applications, interfaces, data, and infrastructure.
Leading financial institutions recognize the need to acknowledge, embrace, and harness complexity. They understand that todays IT environments cannot be managed with experience and intuition-based approaches alone. Instead, true insight into complexity requires objective, quantifiable measurement of its origins and effects.
A pathway to live implementation overview of the key steps required for an organization to start using rigorous, predictive complexity insight as a management lever.
Technology leadership within financial services needs to be able to get specific, with reliable knowledge-based statements: 11
Ideas at Work
Based on the IT complexity metric in our business case, we have determined that we should replace system X, as overall it will reduce our complexity, thus reducing our costs, in parallel with a reduction in service incidents Commerzbank believe their initial complexity model, while useful, would be much richer when evolved towards a fully developed industry standard. To drive this process, the bank chose to work with Capco. Overcoming the issues through a rigorous metrics-based approach and leaving complexity unplugged is one of the foundations of Capcos forming the future of finance platform.
complexity of IT. We start at the level of individual modules/applications and then go on all the way to the total system landscape of a financial institution. Four dimensions of complexity indicators Within our IT complexity model, there are four dimensions:
Capco IT specialists, which in turn is founded on their experience and on leading IT principles used in the industry. Subsequent dialog (based on conversations with experienced IT leaders) has confirmed that these principles are to be found all across our world. The relevance of each indicator has been validated through statistical analysis of Commerzbank data. This analysis has mathematically confirmed the relationship between the IT complexity indicators and IT KPIs, such as costs and incidents. The indicators can be implemented in various ways, accommodating specific data availability issues in different organizations. Note that it is not the absolute value of the indicator that is most significant. What is actually more revealing is the insight gained by combining and comparing indicators. When working
Function relates to functionality or the business and process logic supported by the IT asset.
Interfaces relates to interoperability between the IT assets. Data relates to logical and physical data objects. Technology relates to the underlying technology infrastructure (hardware, middleware, systemware).
What have we achieved so far? Significant progress towards a fully-functional industry tool The Commerzbank-Capco partnership has defined a path to broader industry participation. The key steps have been as follows:
Each indicator is defined on the basis of a hypothesis advanced by Commerzbank and
Commerzbanks
own
modeling
has
Indicator name Function Functional Coverage Functional Occurrence Location of Usage Number of User Number of User Departments Functional Implementation Scope Functional Implementation Overhead (= Fis/Fc) Functional Standard Conformity Functional Domain Diversity Interfaces Information Flow Intensity Interface Intensity Interface Implementation Overhead (= II/IIF) Interface Implementation Type diversity Interface External Isnformation Flow Ratios (external logical dependency) Interface External Ratio (external implementation dependency) Data Number of Information Objects Number of Database Objects Data Implementation Overhead (= DBO/DIO) Technology Number of Infrastructure Requirements Technology Infrastructure Products Number of Infrastructure Services
Indicator abbreviation
improved, becoming more generally reusable within the industry.
Accessible dashboards for IT complexity and analytics tools have been built to ensure that theoretical understanding can be applied to the technology/business management process.
FC FO FLU FU FUD FIS FIO FSC FDD
An open tool for complexity data capture, display, and analysis, reflecting the IT complexity model, is being built with the appropriate confidentiality options to allow for multi-bank user groups.
IIF II IIO IIT IEFR IER
Many sessions with potential users have been held, to confirm the need and refine the methods to reflect the realities of the industry.
DIO DBO DIO
The Complexity Model in detail

In this section we explore the overall framework of the model, which is based on complexity indicators across functions, interfaces, data, and technology (infrastructure). Our ap12 proach focuses on exploring the structural
TIR TIP TIS
Table 1 Overview of complexity indicators

with the complexity indicators, the following points should be kept in mind: (1) not all indicators are necessary all the time a pragmatic approach should prevail; (2) different indicators are relevant for different and distinct questions/ management decisions; (3) relativity is an issue i.e. the indicators relation to other IT assets within the financial institution, or within its business domain (i.e., in a threshold analysis, using average values and standard deviations); (4) benchmarking is an issue consider the relation to the (normalized) IT complexity indicator within the same business domain in similar institutions; and (5) the technology/business relationship is an issue the ratio between the IT implementation-driven indicators and the related business-driven indicators (plotting the IT design overhead and indicating the level of reducibility of IT complexity).
is measured by the number of functional blocks in an overall business domain, or enterprise architecture component. 3. Location of usage (FLU) covering the higher functionality usually required of an application used in more than one location. 4. Number of users (FU) representing the additional design complexity related to the more stringent, non-functional requirements in the areas of security, user-friendliness, performance, and availability. 5. Number of user departments (FUD) relates to the additional complexity due to greater and more disparate business requirements (the result is more complex governance when multiple business departments specify the requirements). The functional implementation scope, FIS, is
intuition and experience. Interface complexity is driven by business needs on the one hand and by IT design and implementation choices on the other. Understanding each of these aspects independently provides compelling insights. But when examination of the influences of business need and IT choice is combined, we have found that their impact on complexity (and therefore cost and service quality) is brutal. This relationship between business requirements and technical implementation is very interesting. Our research has identified three indicators that capture the relationship. And our modeling has confirmed our intuition, that these are indeed useful: 1. Interface information flows (IIF) counts the logical incoming and outgoing information flows of a business process/area (we use the expression application domain) weighted by the required data latency. 2. Interface intensity (II) counts all incoming and outgoing interfaces of an application, weighted by the implementation type (i.e., file transfer, web service, direct database access, application specific API, etc.) and the level of data transformation quantified by the different data structures supported in the interface. 3. Interface implementation type (IIT) is our mechanism to capture technical diversity. The idea is that the greater the diversity of implementation types (file, view, message brokers, etc.) employed in an application, the more complex the application will be.
For analysis and reporting purposes, the IT assets may be clustered in a functional domain model. Alternatively, you can use any form of IT asset bundling, carried out either logically or physically, that is relevant in decision-making around IT complexity (in an IT transformation or project portfolio management process). Functions Many complexity indicators are functionality related (almost half of them, in fact). This is in line with our expectation that IT complexity correlates with the level and diversity of functionality; confirmed by the Commerzbank data analysis. Our model utilizes five business-driven quantitative indicators, indicating the level of functionality that is driven by business requirements and business model choices: 1. Functional coverage (FC) for example, the number of use cases, business process activities, or function points of an application. 2. Functional occurrence (FO) its importance within the overall IT landscape of the enterprise. The functional occurrence
expressed in the number of functional components that are needed to support the required business functionality and its related software engineering needs. The ratio between this indicator and the functional coverage implies the functional implementation overhead FIO, instigated through the application design, and indicates the level of reducibility of IT complexity. This is an important indicator for IT management in reviewing the architecture of an application(s) during IT change. Next to functionality level, we also know that functional diversity drives IT complexity. We have embodied this in two indicators: 1. Functional standard conformity (FSC) of an application is measured on its deviation from enterprise-wide established IT standards and principles. 2. Functional domain diversity (FDD) is measured by the number of applications that are required to support that domain. Interfaces Very early in developing the model, we noticed a statistically significant correlation between interface complexity and cost in line with our
The external information flow dependency, or Interface external information flow ratio (IEFR), measures the relative dependency of an application domain. It is evaluated on its internal structure, based on the ratio between the number of external and internal data flows. The greater this ratio, the greater is the dependency of the domain on its environment i.e., it is not independent! Interface external ratio (IEIR) is the number of interfaces divided by the number of internal interfaces. A possible further 13
Ideas at Work
indicator could be the loop factor counting the loops within an application domain.
the source of real, holistic insight). Technology
to enable comparison of different indicators; aggregated to provide overview insight on specific business domains, such as payments or corporate risk management; and calibrated to allow for comparison between different business domains and benchmarking with other financial institutions. In addition to this information, the level of data completeness and quality (and historic trending) is disclosed in the standard reporting (Figure 1). The information illustrated in Figure 1 provides
Data We have chosen two indicators again one business-driven and one IT implementationdriven. These are statistically proven to not only be relevant for the measurement of complexity in data, they also contribute to an overall understanding. The indicators are: 1. Data information objects (DIO) this indicator looks at the business component, measuring the number of information objects and their relationships within a defined business area (functional domain). To do this, we capture the raw data/ numbers on information objects (person, product, service agreement) in a domain and entities, attributes, and relationships between entities. When combined, the information creates a clear view of the business-related data complexity. 2. Database objects (DBO) this indicator addresses the technical aspects of data, by looking at the number of database objects in an application domain (same definition as for data measure above), based on a physical count of databases, tablespaces, tables, views, constrains, and indices. Added to other views, such as the number of applications in the domain, this indicator allows for useful insights, such as profiling how data intensive these applications are. We believe the two indicators described above are the mandatory data indicators. We recognize, of course, that there could be more but, for benchmarking purposes, these are sufficient. Even working with them in isolation, they are useful. But it is much more important and productive to take the following steps: (1) link the technical to the business-related data indicator, thus gaining an indication of the IT implementation overhead and potential reducibility of design complexity; and (2) aggregate the data complexity with the application, in14 terface, and infrastructure complexity (this is
Another and crucial dimension of IT complexity is clearly technology. This overall description relates to the underlying technology itself (hardware, middleware, systemware, etc.) as well as any drivers of complexity derived from operating requirements (operating model, business criticality, prescribed time for recovery, etc.). When analyzing technology complexity, we are aiming to understand the impact of employed infrastructure for an application or an application bundle on complexity. One can also look at infrastructure bundles, for example in an operations center. We propose the following technology indicators:
a dashboard. But we find that the true value of IT complexity management lies in its use in proactive support of IT portfolio management, architecture, and transformation decisions. To explore and realize this value, we run simulations and scenario analyses to model the changes in complexity for different options. In this way, we can finally take IT complexity the hidden variable into account to make better decisions.
Technology infrastructure products (TIP) measures the number of different infrastructure products/components, including, among others, databases, operating system, and applications server.
The model works, now how can you make it work for you?
If you want to take part in this exciting development what do you need to do and how long does it take? Based on experience to date, and knowledge of the likely data available in most financial services technology shops, we envisage that three key streams of activity will be required:
Technology infrastructure services (TIS) counts the number of different infrastructure services.
Technology infrastructure requirements (TIR) measures the number of different infrastructure service level requirements to be supported. To this end, we assign numerical values to business criticality and other non-functional infrastructure service requirements.
Data availability modeling complexity firstly requires data on complexity, as we have explained. The initial step is to assess what data you have, align it with the models, and think through the data capture and transfer process. Completeness is not necessary, and one does not need every indicator, nor all IT assets covered. Follow a pragmatic approach by starting, for example, with one business domain.
Complexity model applied for reporting and management decisions The basic information used in IT complexity reporting and decision-making comprises: IT complexity indicators as described above; the impact of the indicators on flexibility, costs, and quality; and benchmark data to link the above measures to comparable organizations. In the standard tool that we have built to measure complexity, this base data is: normalized
Technical load once the data has been sourced and made available, we would recommend a first technical load into the analytical tools which can be used as a pilot and enables data refinement based on the results. This is followed by a first for-

mal data load, which would enable you to mobilize and use it as a management tool.
the analysis, load, and trial period depending on data availability. It will subsequently require 0.5 FTE to maintain the data and manage the processes to achieve greatest benefit. Initially, we considered this was a high investment level. But we have since observed that the preparation process has significant ancillary benefits, including: improved data understanding you will start to understand your IT assets better; improved data quality you will improve the quality of information on your IT assets; and putting in foundations for managing complexity this is all part of embedding a sustainable culture of ongoing, rigorous, and predictive complexity management
Conclusion
This paper highlighted the need for a robust and functioning model for complexity measurement and mitigation, described how Commerzbank, in partnership with Capco, is deriving such a model, and examined how it can be developed over time to provide a crossindustry standard. We hope that we have achieved our objective of demonstrating the importance of understanding complexity, adding to your existing awareness of the drivers of complexity, proved that IT complexity can be measured and therefore mastered and persuaded you that now is the time to act.
Sharing if you are comfortable and your organization allows it, we will work with you to ensure that a sanitized version of your data is included in the benchmarking dataset. You can participate in the user community process, which will allow you to learn along with others, benchmark against them, and also to progress the model.
It is a three- to six-month journey, worst case, to get to the point of realizing the benefits of understanding complexity. Early indications are that this activity will absorb approximately 1.5 full-time equivalent (FTE) of time through
Data quality
Functions
Indicator History
Threshold analysis
8806
5%
19 0
IT complexity Indicators IT performance Indicators Data
Thre DQ s Indicator Hist. Thre DQ s
Domain Cluster Functional domain Core Business Objects Sales Business Operations & Services
Loans Deposits / Accounts Transaction Services Securities Trading Services
#IT pro DQ ducts 73 231 138 44 8 21 25 40 162 243 21 124 992
Functions
Indicator Hist. Thre DQ s
Interfaces
Indicator Hist.
Technology
Indicator Hist. Thre DQ s Change Hst. Thr
Costs (kEUR)
DQ App. Mnt Hst. Thr DQ IT Ops Hist. Thr DQ
Incidents
Incidents Hist. Tha
8806 8806 23009 12314 3838 1611 2441 1716 2708 18257 20305 3394 14645 100730
5% 5% -6% -6% -10% 0 -6% 6% -20% 2% 6% -4% 7% 1%
19 0 75 1 30 1 8 1 7 6 4 5 37 55 21 10 3 35 2 261 28
11734 11734 15047 11349 2888 895 2368 2472 2726 14695 21472 21486 4800 100583
6% -4% -4% -6% 0.01 0.01 0.01 -12% 6% 0.03 5% 0 -2%
23 12 44 33 36 24 5 12 6 13 2 9 3 3 7 51 30 38 60 18 22 19 232 178
10753 22576 10503 2312 924 2086 2640 2541 10425 21770 17150 7268 100445
-6% -4% -8% -6% 0.01 -0.11 0.01 -12% 3% 0.03 4% 0 -2%
16 8 55 23 24 12 8 6 2 6 10 2 6 3 7 3 15 45 76 23 16 22 19 224 130
7186 20709 14575 3579 1383 2463 3465 3686 22542 14146 1822 9983 90964
6% -6% -3% -2% 0.01 0.01 0.1 -1% 6% 0.03 5% 0 -2%
20 4 23 14 55 10 9 4 4 7 3 9 3 3 7 67 20 38 60 5 22 19 230 127
398,361 26% 93,248 -10% 27,003 -7% 12,701 -15% _ 8,316 -4% 1,559 -10% 4,427 1% 83,517 6% 31,139 2% 11,759 645028 0 -0.07
1 4 2 0 0
11,206 8% 32,802 4% 0 7% 0 0
20,646 8% 53,021 -1% 0 0 23,113 8% 6,579 12% 6,181 4% 2,091 8% 2,102 0
3 2
Transaction Processing Group Steering Data Warehouse Corporate services Total
4 2
8,420 3% 10,532 -6% 19,592 0.08

1
2 0
1
6,160 7% 48,301 0.01 26,522 -6% 19,046 0
1 6 2 3
8,706 -9% 0 0 199356 -8% 18 0
9 4
82551
9%
1 0
Functions complexity
Interfaces complexity
Data complexity
Technology complexity
14000 12000 10000 8000
14000 12000 10000 8000
Interfaces complexity for business objects
IER IEFR IIT II IIF
IER IEFR IIT II IIF
6000 6000
Core business objects Sales Business operations & service Data warehouse Transaction processing Group steering Corporate services
4000
4000
2000 2000
00
Quantity Quantity
Diversity Interdependency Diversity Interdependency
Figure 1 Commerzbank IT complexity landscape
15
Part 1
Toward a Bottom-Up Approach to Assessing Sovereign Default Risk: An Update International Liquidity Provision and Currency-Specific Liquidity Shortages The Failure of Financial Econometrics: Stir-Fry Regressions as a Con Job Lehman A Case of Strategic Risk Explaining Credit Default Swaps Pricing for Large Banks Investing in Private Equity Capital Commitment Considerations Moving the OTC Derivatives Market to CCPs Cultural, Political, and Economic Antecedents of Country Risk in Sixty-Two Countries
PART 1
Toward a Bottom-Up Approach to Assessing Sovereign Default Risk: An Update

1
Edward I. Altman Max L. Heine Professor of Finance, NYU Stern School of Business Herbert Rijken Professor of Finance, Vrije Universiteit Amsterdam2
Abstract
We propose a totally new approach toward assessing sovereign risk by examining rigorously the health and aggregate default risk of a nations private corporate sector. Models such as our new Z-Metrics approach can be utilized to measure the median probability of default of the non-financial sector cumulatively for five years, both as an absolute measure of corporate risk vulnerability and a relative measure compared to other sovereigns and to the markets assessment via the now liquid credit-default-swap market. Specifically, we measure the default probabilities of listed corporate entities in 11 European countries and the U.S., as of 2008-2010. These periods coincide with the significant rise in concern with sovereign default risk in the euro country sphere. We conclude that our corporate health index of
1 2 This is an updated version of the article originally published in The Journal of Applied Corporate Finance, Vol. 23, No. 3, Winter, 2011. The authors would like to thank Dan Balan and Matthew Watt of RiskMetrics Group, a subsidiary of MSCI, Inc., for computational assistance, and Brenda Kuehne of the NYU Salomon Center for her research assistance.
the private sector measured at periods prior to the explicit recognition by most credit professionals not only gave an effective early warning indicator, but provided a mostly appropriate hierarchy of relative sovereign risk. Policy officials should, we believe, nurture, not penalize, the tax revenue paying and jobs generating private sector when considering austerity measures of distressed sovereigns.
19
During the past four years, bank executives, government officials, and many others have been sharply criticized for failing to anticipate the global financial crisis. The speed and depth of the market declines shocked the public. And no one seemed more surprised than the credit rating agencies that assess the default risk of sovereign governments as well as corporate issuers operating within their borders. Although the developed world had suffered numerous recessions in the past 150 years, this most recent international crisis raised grave doubts about the ability of major banks and even sovereign governments to honor their obligations. Several large financial institutions in the U.S. and Europe required massive state assistance to remain solvent, and venerable financial institutions like Lehman Brothers even went bankrupt. The cost to the U.S. and other sovereign governments of rescuing financial institutions believed to pose systemic risk was so great as to result in a dramatic increase in their own borrowings.
and budget deficits as gauges of a countrys economic strength and well-being. But, as the recent euro debt crisis has made clear, such macro approaches, while useful in some settings and circumstances, have clear limitations. In this paper, we present a totally new method for assessing sovereign risk, a type of bottom-up approach that focuses on the financial condition and profitability of an economys private sector. The assumption underlying this approach is that the fundamental source of national wealth, and of the financial health of sovereigns, is the economic output and productivity of their companies. To the extent we are correct, such an approach could provide financial professionals and policymakers with a more effective means of anticipating financial trouble, thereby enabling them to understand the sources of problems before they become unmanageable. In the pages that follow, we introduce Z-Metrics as a practical and ef-
The general public in the U.S. and Europe found these events particularly troubling because they had assumed that elected officials and regulators were well-informed about financial risks and capable of limiting serious threats to their investments, savings, and pensions. High-ranking officials, central bankers, financial regulators, ratings agencies, and senior bank executives all seemed to fail to sense the looming financial danger. This failure seemed even more puzzling because it occurred years after the widespread adoption of advanced risk management tools. Banks and portfolio managers had long been using quantitative risk management tools such as Value at Risk (VaR), and should have also benefited from the additional information about credit risk made publicly available by the new market for credit default swaps (CDS). But, as financial market observers have pointed out, VaR calculations are no more reliable than the assumptions underlying them. Although such assumptions tend to be informed by statistical histories, critical variables such as price volatilities and correlations are far from constant and thus difficult to capture in a model. The market prices of options or of CDS contracts, which have options embedded within them can provide useful market estimates of volatility and risk. And economists have found that CDS prices on certain kinds of debt securities increase substantially before financial crises become full-blown. But because there is so little time between the sharp increase in CDS prices and the subsequent crisis, policymakers and financial managers typically have little opportunity to change course.3 The most popular tools for assessing sovereign risk are effectively forms of top-down analyses. For example, in evaluating particular sovereigns, most academic and professional analysts use macroeconomic 20 indicators such as GDP growth, national debt-to-GDP ratios, and trade
fective tool for estimating sovereign risk. Developed in collaboration with the Risk Metrics Group, now a subsidiary of MSCI, Inc., Z-Metrics is a logical extension of the Altman Z-Score technique that was introduced in 1968 and has since achieved considerable scholarly and commercial success. Of course, no method is infallible, or represents the best fit for all circumstances. But by focusing on the financial health of private enterprises in different countries, our system promises, at the very least, to provide a valuable complement to, or reality check on, standard macro approaches. But before we delve into the details of Z-Metrics, we will briefly review the record of financial crises to provide some historical perspective. Next, we attempt to summarize the main findings of the extensive academic and practitioner literature on sovereign risk, particularly those studies designed to test the predictability of sovereign defaults and crises. With that as background, we then present our new Z-Metrics system for estimating the probability of default for individual (non-financial) companies and show how that system might have been used to anticipate many developments during the current E.U. debt crisis. In so doing, we make use of the most recent (2009 and 2010) publicly available corporate data for nine European countries, both to illustrate our models promise for assessing sovereign risk and to identify scope of reforms that troubled governments must consider not only to qualify for bailouts and subsidies from other countries and international bodies, but to stimulate growth in their economies.
See, for example, Neziri (2009), who found that CDS prices had real predictive power for equity markets, but that the lead time was generally on the order of one month.

Austria Brazil Canada Czechoslovakia China Denmark Germany GBR Greece Italy Japan Netherlands Norway Russia Spain Sweden U.S.
1893, 1989 1898, 1902,1914, 1931, 1939 1873, 1906, 1923, 1983 1870, 1910, 1931, 2008 1921, 1939 1877, 1885, 1902, 1907, 1921, 1931, 1987 1880, 1891, 1901, 1931, 2008 1890, 1974, 1984, 1991, 2007 1870, 1894, 1932, 2009 1887, 1891, 1907, 1931, 1930, 1935, 1990 1942 1897, 1921, 1939 1899, 1921, 1931, 1988 1918, 1998 1920, 1924, 1931, 1978, 2008 1876, 1897, 1907, 1922, 1931, 1991 1873, 1884, 1893, 1907, 1929, 1984, 2008
pricing indicators such as CDS spreads (while avoiding the well-known volatility of the latter). Our aim here is not to present a beauty contest of different methods for assessing sovereign risk in which one method emerges as the clear winner. What we are suggesting is that a novel, bottom-up approach that emphasizes the financial condition and profitability of a nations private sector can be effectively combined with standard analytical techniques and market pricing to better understand and predict sovereign health. And our analysis has one clear implication for policymakers: that the reforms now being contemplated should be designed, as far as possible, to preserve the efficiency and value of a nations private enterprises.
Modern history sovereign crises

When thinking about the most recent financial crisis, it is important to keep in mind how common sovereign debt crises have been during the last 150 years and how frequently such debacles have afflicted developed economies as well as emerging market countries. Table 1 shows a partial list of financial crises (identified by the first year of the crisis) that have occurred in advanced countries. Overall, Latin America seems to have had more recent bond and loan defaults than any other region of the world (as can be seen in Figure 1). But if we had included a number of now developed Asian countries among the advanced countries, the
Source: IMF Global Financial Stability Report (2010), Reinhart and Rogoff (2010), and various other sources, such as S&Ps economic reports.
Table 1 Financial crises, advanced countries 1870-2010 crisis events (first year)
140 120 100 80 60 40 20 0
Syndicated bank loans Sovereign bonds All
1997-1999 period would be much more prominent. The clear lesson from Table 1 and Figure 1 is that sovereign economic conditions appear to spiral out of control with almost predictable regularity and then require massive debt restructurings and/or bailouts accompanied by painful austerity programs. Recent examples include several Latin American countries in the 1980s, Southeast Asian nations in the late 1990s, Russia in 1998, and Argentina in 2000. In most of those cases, major problems originating in individual countries not only imposed hardships on their own people and markets, but had major financial consequences well beyond their borders. We are seeing such effects now as financial problems in Greece and other southern European countries not only affect their neighbors, but threaten the very existence of the Euro-
Asia
Western Europe
Eastern Europe
Africa
Latin America
Figure 1 Number of sovereign defaults (1824-2004)
More specifically, we examine the effectiveness of calculating the median company five-year probability of default of the sovereigns non-financial corporate sector, both as an absolute measure of corporate risk vulnerability and a relative health index comparison among a number of European sovereigns, and including the U.S. as well. Our analysis shows that this health index, measured at periods prior to the explicit recognition of the crisis by market professionals, not only gave a distinct early warning of impending sovereign default in some cases, but also provided a sensible hierarchy of relative sovereign risk. We also show that, during the current European crisis, our measures not only compared favorably to standard sovereign risk measures, notably credit ratings, but performed well even when compared to the implied default rates built into market
pean Union. Such financial crises have generally come as a surprise to most people, including even those specialists charged with rating the default risk of sovereigns and the enterprises operating in these suddenly threatened nations. For example, it was not long ago that Greek debt was investment grade, and Spain was rated Aaa as recently as June 2010.4 And this
On April 27, 2010, Standard & Poors Ratings Services lowered its long- and short-term credit ratings on the Hellenic Republic (Greece) to non-investment grade BB+; and on June14, 2010, Moodys downgraded Greece debt to Ba1 from A2 (4 notches), while Spain was still Aaa and Portugal was A1. Both of the latter were recently downgraded. S&P gave similar ratings.
21
pattern has been seen many times before. To cite just one more case, South Korea was viewed in 1996 as an Asian Tiger with a decade-long record of remarkable growth and an AA- rating. Within a year, however, the country was downgraded to BB-, a junk rating, and the countys government avoided default only through a U.S.$50 billion bailout by the IMF. It was not only the rating agencies that were fooled; most of the economists at the brokerage houses also failed to see the problems looming in Korea.
indicated that Greece and Spain and others now recognized as highrisk countries were still classified as investment grade.10 What is more, although almost all of the studies cited above have been fairly optimistic about the ability of their concepts to provide early warnings of major financial problems, their findings have either been ignored or proven ineffective in forecasting most economic and financial crises. In addition to these studies, a handful or researchers have taken a somewhat different bottom-up approach by emphasizing the health of the private sectors supporting the sovereigns. For example, a 1998 World Bank study of the 1997 East Asian crisis11 used the average Z-Score of listed (non-financial) companies to assess the financial fragility of eight Asian countries and, for comparison purposes, three developed countries, and Latin America. Surprising many observers, the average ZScore for South Korea at the end of 1996 suggested that it was the most
What do we know about predicting sovereign defaults?

There is a large and growing body of studies on the default probability of sovereigns, by practitioners as well as academics. A large number
5
of studies, starting with Frank and Clines 1971 classic, have attempted to predict sovereign defaults or rescheduling using statistical classification and predicting methods like discriminant analysis as well as similar econometric techniques.6 In a more recent development, some credit analysts have begun using the contingent claim approach [Gray et al. (2006, 2007)] to measure, analyze, and manage sovereign risk based on Robert Mertons classic structural approach [Merton (1974)]. But because of its heavy reliance on market indicators, this approach to predicting sovereign risk and credit spreads has the drawback of producing large and potentially self-fulfilling swings in assessed risk that are attributable solely to market volatility. A number of recent studies have sought to identify global or regional common risk factors that largely determine the level of sovereign risk in the world, or in a region such as Europe. Some studies have shown that changes in both the risk factor of individual sovereigns and in a common time-varying global factor affect the markets repricing of sovereign risk.7 Other studies, however, suggest that sovereign credit spreads are more related to global aggregate market indexes, including U.S. stock and high-yield bond market indexes, and global capital flows than to their own local economic measures [Longstaff et al. (2007)]. Such evidence has been used to justify an approach to quantifying sovereign risk that uses the local stock market index as a proxy for the equity value of the country [Oshiro and Saruwatari (2005)]. Finally, several very recent papers focus on the importance of macro variables such as debt service relative to tax receipts and the volatility of trade deficits in explaining sovereign risk premiums and spreads.8 A number of studies have also attempted to evaluate the effectiveness of published credit ratings in predicting defaults and expected losses, with most concluding that sovereign ratings, especially in emerging markets, provide an improved understanding of country risks for investment analytics.9 Nevertheless, the recent E.U. debt crisis would appear to contradict such findings by taking place at a time when all the rating agencies 22 and, it would seem, all available models for estimating sovereign risk,
One excellent primer on sovereign risk is Babbels (1996) study, which includes an excellent annotated bibliography by S. Bertozzi on external debt capacity that describes many of these studies. Babbel lists 69 potentially helpful explanatory factors for assessing sovereign risk, all dealing with economic, financial, political, or social variables. Except for the political and social variables, all others are macroeconomic data and this has been the standard until the last few years. Other work worth citing include two practitioner reports [Chambers (1997) and Beers et al. (2002)] and two academic studies [Smith and Walter (2003), and Frenkel et al. (2004)]. Full citations of all studies can be found in references section at the end of the article. 6 Including Grinols (1976), Sargen (1977), Feder and Just (1977), Feder et al. (1981), Cline (1983), Schmidt (1984), and Morgan (1986). 7 See Baek et al. (2005). Gerlach et al. (2010) observe that aggregate risk factors drive banking and sovereign market risk spreads in the Euro Area. In a related finding, Sgherri and Zoli (2009) suggest that Euro Area sovereign risk premium differentials tend to move together over time and are driven mainly by a common time-varying factor. 8 These include Haugh et al.s (2009) discussion of debt service relative to tax receipts in the Euro Area; Hilscher and Nobusch (2010) emphasis on the volatility of terms of trade; and Segoviano et al.s (2010) analysis of debt sustainability and the management of a sovereigns balance sheet. 9 For example, Remolona et al. (2008) reach this conclusion after using sovereign credit ratings and historical default rates provided by rating agencies to construct a measure of ratings implied expected loss. 10 To be fair, S&P in a Reuters article dated January 14, 2009 warned Greece, Spain, and Ireland that their ratings could be downgraded further as economic conditions deteriorated. At that time, Greece was rated A1 by Moodys and A- by S&P. Interestingly, it was almost a full year later on December 22, 2009 that Greece was actually downgraded by Moodys to A2 (still highly rated), followed by further downgrades on April 23, 2010 (to A3) and finally to junk status (Ba1) on June 14, 2010. As noted earlier, S&P downgraded Greece to junk status about three months earlier. 11 See Pomerleano (1998), which is based on a longer article by the author (1997). Taking a somewhat similar approach, many policymakers and theorists have recently focused on the so-called shadow banking system. For example, Gennaioli et al. (2010) argued that the financial strength of governments depends on private financial markets and their ability to attract foreign capital. They concluded that strong financial institutions not only attract more capital but their presence also helps encourage their governments to repay their debt. Chambers (1997) of S&P also mentions the idea of a bottom-up approach but not to assess sovereign risk, but corporate issuers located in a particular country. He advocates first an evaluation of an issuers underlying creditworthiness to arrive at its credit rating and then considers the economic, business, and social environment in which the entity operates. These latter factors, such as the size and growth and the volatility of the economy, exchange rates, inflation, regulatory environment, taxation, infrastructure, and labor market conditions are factored in on top of the micro variables to arrive at a final rating of the issuer.

financially vulnerable Asian country, followed by Thailand, Japan, and Indonesia. As noted earlier, Koreas sovereign bond rating in 1996 was AA- (S&P), but within a year, its rating dropped to BB-; and if not for the IMF bailout of $50 billion, the sovereign would almost certainly have defaulted on its external, non-local currency debt. A traditional macroeconomic measure like GDP growth would not have predicted such trouble since, at the end of 1996, South Korea had been growing at double-digit rates for nearly a decade.12
Type I error rate 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 AE rating: Z" score Agency rating AE rating: Z-metrics public one year
BB -
tially, of creating a new and better way of assessing the credit risk of companies. The result was our new Z-Metrics approach. This methodology might be called a new generation of the original Z-Score model of 1968. Our objective was to develop up-to-date credit scoring and probability of default metrics for both large and small, public and private, enterprises on a global basis. In building our models, we used multivariate logistic regressions and data from a large sample of both public and private U.S. and Canadian non-financial sector companies during the 20-year period 1989-2008.13 We analyzed over 50 fundamental financial statement variables, including measures (with trends as well as point estimates) of solvency, leverage, size, profitability, interest coverage, liquidity, asset quality, investment, dividend payout, and financing results. In addition to such operating (or fundamental) variables, we also included equity market price and return variables and their patterns of volatility. Such market variables have typically been used in the structural distance-to-default measures that are at the core of the KMV model14 now owned by Moodys.
Figure 2 Type I error for Agency ratings, Z-score, and Z-Metrics agency equivalent [AE ratings (1989-2008): one year prediction horizon for publicly owned firms]
model is based on data from financial statements and market data approximately one year prior to the credit event, and the five-year model includes up to five annual financial statements prior to the event. To test the predictive power of the model and the resulting PDs, we segregated all the companies in our sample into cohorts according to whether they experience credit events that include either formal default or bankruptcy, whichever comes first. All companies that experienced a credit event within either one year or five years were assigned to the distressed or credit event group; with all others assigned to the nondistressed group. Our test results show considerable success in predicting defaults across
In addition to these firm-specific, or micro, variables, we also tested a number of macro-economic variables that are often used to estimate sovereign default probabilities, including GDP growth, unemployment, credit spreads, and inflation. Since most companies have a higher probability of default during periods of economic stress for example, at the end of 2008 we wanted to use such macro variables to capture the heightened or lower probabilities associated with general economic conditions.15 The final model, which consists of 13 fundamental, market value, and macroeconomic variables, is used to produce a credit score for each public company. (And as discussed later, although our primary emphasis was on applying Z-Metrics to publicly traded companies, we also created a private firm model by using data from public companies and replacing market value with book value of equity.) The next step was to use a logit specification of the model (described in the Appendix) that we used to convert the credit scores into probabilities of default (PDs) over both one-year and five-year horizons. The one-year
the entire credit spectrum from the lowest to the highest default risk categories. Where possible, we compared our output with that of publicly available credit ratings and existing models. The so-called accuracy ratio measures how well our model predicts which companies do or do not go bankrupt on the basis of data available before bankruptcy. The objective can be framed in two ways: (1) maximizing correct predictions of defaulting
12 Afterwards, the World Bank and other economists such as Paul Krugman concluded that crony capitalism and the associated implicit public guarantees for politically influential enterprises coupled with poor banking regulation were responsible for the crisis. The excesses of corporate leverage and permissive banking were addressed successfully in the case of Korea and its economy was effectively restructured after the bailout. 13 Our first models original sample consisted of over 1,000 U.S. or Canadian non-financial firms that suffered a credit event and a control sample of thousands of firms that did not suffer a credit event, roughly a ratio of 1:15. After removing those firms with insufficient data, the credit event sample was reduced to 638 firms for our public firm sample and 802 observations for our private firm sample. 14 Developed by Crosbie in 1998 and adapted for sovereigns by Gray in 2007. 15 In all cases, we carefully examined the complete distribution of variable values, especially in the credit-event sample. This enabled us to devise transformations on the variables to either capture the nature of their distributions or to reduce the influence of outliers. These transformations included logarithmic functions, first differences and dummy variables if the trends or levels of the absolute measures were positive/negative.
In 2009, we partnered with RiskMetrics Group with the aim, at least ini-
/C
rating class (cutoff score = score at upper boundary of rating class N)
BB B-
BB +
The Z-Metrics approach [Altman et al. (2010)]
B-
B+
BB
/C
23
Z-Metrics PD estimates: five-year public model Country Netherlands U.S. Sweden Ireland Belgium U.K. France Germany Italy Spain Portugal Greece Listed companies 85 2226 245 29 69 507 351 348 174 91 33 93 Y/E 2010 median PD 3.56% 3.65% 3.71% 3.72% 3.85% 4.28% 4.36% 4.63% 7.29% 7.39% 10.67% 15.28% Y/E 2009 median PD 3.33% 3.93% 5.31% 6.45% 5.90% 3.62% 5.51% 5.54% 7.99% 6.44% 9.36% 10.60% Y/E 2008 median PD 5.62% 6.97% 6.74% 7.46% 5.89% 5.75% 7.22% 7.34% 10.51% 7.39% 12.07% 11.57%
Five-year implied PD from CDS spread* 2010 2.03% 3.79% 2.25% 41.44% 11.12% 4.73% 4.51% 2.50% 9.16% 14.80% 41.00% 70.66% 2009 2.83% 3.28% 4.60% 12.20% 4.58% 6.52% 3.75% 2.67% 8.69% 9.39% 10.90% 24.10% 2008 6.06% 4.47% 6.33% 17.00% 5.53% 8.13% 4.05% 3.66% 11.20% 8.07% 7.39% 13.22%
*Assuming a 40% recovery rate (R); based on the median CDS spread (s). PD computed as 1-e(-5*s/(1-R)). Sources: RiskMetrics Group (MSCI), Markit, Compustat.
Table 2 Financial health of the corporate, non-financial sector: selected European countries and U.S.A. in 2008-2010
and non-defaulting companies (which statisticians refer to as Type I accuracy) and (2) minimizing wrong predictions (Type II accuracy). As can be seen in Figure 2, our results, which include tests on actual defaults during the period 1989-2009, show much higher Type I accuracy levels for the Z-Metrics model than for either the bond rating agencies or established models (including an older version of Z-Scores). At the same time, our tests show equivalent Type II accuracies at all cutoff levels of scores.16 Perhaps the most reliable test of credit scoring models is how well they predict critical events based on samples of companies that were not used to build the model, particularly if the events took place after the period during which the model was built (after 2008, in this case). With that in mind, we tested the model against actual bankruptcies occurring in 2009, or what we refer to as our out-of-sample data. As with the full test sample results shown in Figure 2, our Z-Metrics results for the out of sample bankruptcies of 2009 outperformed the agency ratings and the 1968 Z-score and 1995 Z-score models using both one-year and five-year horizons.
probabilities for individual companies and then estimating both a median default probability and credit rating for different countries. In conducting this experiment, we examined nine key European countries over three time periods, end of 2008, 2009, and 2010 (Table 2) and again at the end of 2010 (Table 3), when the crisis was well known. People clearly recognized the crisis and concern for the viability of the European Union in June 2010, when Greeces debt was downgraded to noninvestment grade and both Spain and Portugal were also downgraded. Credit markets, particularly CDS markets, had already recognized the Greek and Irish problems before June 2010. Market prices during the first half of 2010 reflected high implied probabilities of default for Greece and Ireland, but were considerably less pessimistic in 2009. By contrast, as can be seen in Table 2, which shows our Z-Metric median PD estimates alongside sovereign CDS spreads over both periods,17 our PD estimates were uniformly higher (more risky) in 2009 than in early 2010, even if the world was more focused on Europes problems in the latter year. In this sense, our Z metrics PD might be viewed as providing a leading indicator of possible distress. It should be noted that the statistics in Table 2 report only on the non-financial private sector, while those in Table 3 include results from our banking credit risk model, as well.
A bottom-up approach for sovereign risk assessment

Having established the predictive power of our updated Z-score methodology, our next step was to use that model, which, again, was created using large publicly traded U.S. companies, to evaluate the default risk of European companies. After assuring ourselves that the model was transferable in that sense, we then attempted to assess the overall creditwor24 thiness of sovereign governments by aggregating our Z-Metrics default
16 We assessed the stability of the Z-Metrics models by observing the accuracy ratios for our tests in the in-sample and out-of-sample periods and also by observing the size, signs, and significance of the coefficients for individual variables. The accuracy ratios were very similar between the two sample periods and the coefficients and significance tests were extremely close. 17 The median CDS spread is based on the daily observations in the six-/four-month periods. The median Z-Metrics PD is based on the median company PDs each day and then we calculated the median for the period. The results are very similar to simply averaging the median PDs as of the beginning and ending of each sample period.

For the first four months of 2010, our Z-Metrics five-year PDs for European corporate default risk placed Greece (10.60 percent) and Portugal (9.36 percent) in the highest risk categories (ZC-ratings), followed by Italy (7.99 percent), Ireland (6.45 percent), and Spain (6.44 percent), all in the ZC category. Then came Germany and France (both about 5.5 percent ZC+), with the U.K. (3.62 percent), and the Netherlands (3.33 percent) at the lowest risk levels (ZB and ZB). The U.S. looked comparatively strong, at 3.93 percent (ZB-). For the most part, these results are consistent with how traditional analysts now rank sovereign risks. Nevertheless, there were a few surprises. The U.K. had a fairly healthy private sector, and Germany and France were perhaps not as healthy as one might have thought. The U.K.s relatively strong showing might have resulted from the fact that our risk measure at this time did not include financial sector firms, which comprised about 35 percent of the market values of listed U.K. corporates and were in poor financial condition. And several very large, healthy multinational entities in the U.K. index might have skewed results a bit. The CDS/fiveyear markets assessment of U.K. risk was harsher than that of our ZMetrics index in 2010, with the median of the daily CDS spreads during the first four months implying a 6.52 percent probability of default, about double our Z-Metrics median level. Greece also had a much higher CDS implied PD at 24.10 percent, as compared to 10.60 percent for Z-Metrics. (And, of course, our choice of the median Z-Metrics PD is arbitrary, implying as it does that 50 percent of the listed companies have PDs higher than 10.60 percent.) We also observed that several countries had relatively high standard deviations of Z-Metrics PDs, indicating a longer tail of very risky companies.
These countries included Ireland, Greece and, surprisingly, Germany, based on 2010 data. So, while almost everyone considers Germany to be the benchmark low-risk country in Europe [for example, its five-year CDS spread was just 2.67 percent in 2010, even lower than the Netherlands (2.83 percent)], we are more cautious based on our broad measure of private sector corporate health.
2010 results
Table 3 shows the weighted-average median PDs for 11 (including now Sweden and Belgium) European countries and the U.S. as of the end of 2010. Note that we now are able to include PDs for the banking sectors (listed firms only) for these countries, an important addition, especially for countries like Greece, Ireland, and the U.K. The results show the large difference between Greece (16.45 percent) and all the rest, but also that the big-five PIIGS stand out as the clear higher risk domains. Indeed, we feel that Italy could be the fulcrum country to decide the ultimate fate of the euro (see our Insight article in the Financial Times, June 21, 2011).
CDS implied PDs

Figure 3 shows the implied PDs for the big five European high-risk countries from the start of 2009 to mid-July 2011, just after the European Unions comprehensive rescue plan was announced (July 21, 2011) for Greece and a contingent plan for other countries. Note that while the PDs, based on CDS spreads and assuming a 40 percent recovery rate, all came down from their highs, all still imply a considerable default risk. Indeed, as of mid-January 2012, the Greek CDS implied that the probability of default increased to almost 95 percent, and Italy, the subject of our fulcrum risk country Insight piece, increased from 19 percent in July 2011 to 35 percent in January 2012.
Non-financial firms Country Netherlands Sweden Belgium France U.K. Germany U.S.A. Spain Italy Ireland Portugal Greece PD (%) 3.56 3.71 3.85 4.36 4.28 4.63 3.65 7.39 7.29 3.72 10.67 15.28 Weight 0.977 0.984 0.972 0.986 0.977 0.983 0.837 0.948 0.906 0.906 0.971 0.921
Banking firms PD (%) 11.1 17.3 12.4 14.0 15.5 13.1 13.8 10.9 20.0 77.6 12.1 30.1 Weight 0.023 0.016 0.028 0.014 0.023 0.017 0.163 0.052 0.094 0.094 0.029 0.079 Weighted average (%) 3.73 3.93 4.21 4.49 4.54 4.77 5.30 7.57 8.48 10.65 10.71 16.45 Rank 1 2 3 4 5 6 7 8 9 10 11 12 CDS spread PD (%)*** 2.03 2.25 11.12 4.51 4.73 2.50 3.79 14.80 9.16 41.44 41.00 70.66 Rank 1 2 8 5 6 3 4 9 7 11 10 12
*Based on the Z-Metrics Probability Model. **Based on Altman-Rijken Model (Preliminary). ***PD based on the CDS Spread as of 4/26/11.
Table 3 Weighted average median five-year (PD) for listed non-financial* and banking firms** (Europe and U.S.), 2010
25
of 2010. As can be seen in the table, most countries enjoyed increases

Default Probability (as %) 100 90 80 70 60 50 40 30 20 10 4-Jun-09 4-Feb-11 4-Jun-10 4-Jan-11 4-Apr-11 4-May-11 4-Jun-11 4-Nov-09 4-Nov-10 4-Aug-09 4-Sep-09 4-Dec-09 4-Aug-10 4-Sep-10 4-May-09 4-May-10 4-Dec-10 4-Mar-09 4-Mar-10 4-Feb-09 4-Feb-10 4-Mar-11 4-Jan-09 4-Oct-09 4-Jan-10 4-Oct-10 4-Apr-09 4-Apr-10 4-Jul-09 4-Jul-10 0 Italy 19.01 4-Jul-11 Portugal 53.38 Greece 74.52
of greater than 20 percent. Only Greece had a relatively low increase (5.5 percent), consistent with its modest improvement in its Z-Metrics PD (-8.4 percent). Figure 3 shows the percentage improvement (lower risk) in sovereigns PDs in 2010, which are largely consistent with the increases in stock market index values. Note that Ireland stands out in that while its stock market index value increased by 26.2 percent, its corporate sector experienced only a modest improvement (-7.4 percent) in its Z-Metrics median PD. This may be attributable to the earlier austerity measures taken in Ireland, as compared to those in other distressed European nations. But likely more important were changes in the many other variables in the Z-Metrics model that are not affected by stock prices, particularly the fundamental measures of corporate health.
Ireland 51.57 Spain 22.70
Spain
Italy
Greece
Portugal
Ireland
* Assumes 40% recovery rate. PD computed as 1-e(-5*s/(1-R)). ** On July 19, 2011, PDs for all countries peaked as follows: Greece 88.22, Portugal 64.74, Ireland 64.23, Spain 27.54, and Italy 23.74. These peaks were exceeded in late 2011 and early 2012. Sources: Bloomberg and NYU Salomon Center.
Figure 3 Five-year implied probabilities of default (PD)* from capital market CDS spreads, January 2009 July 22, 2011**
Comparing PD results based on privately owned versus publicly owned firm models
As shown in Tables 2 and 3, the improvement (reduction) in Z-Metrics PDs for most countries in 2010 a period in which most E.U. sovereigns appeared to be getting riskier looks attributable in large part to the stock market increases in almost all countries. But to the extent such increases could conceal a deterioration of a sovereigns credit condition, some credit analysts might prefer to have PD estimates that do not make use of stock market data. With this in mind, we applied our private firm Z-Metrics model to evaluate the same nine European countries and the U.S. The private and public firm models are the same except for the substitution of equity book values (and volatility of book values) for market values. This adjustment is expected to remove the capital market influence from our credit risk measure. Table 5 summarizes the results of our public versus private firm Z-Metrics
2010 versus 2009

As noted earlier from Table 2, our PD estimates for 2009 were uniformly higher (more risky) than those for early 2010. One important reason for the higher PDs in 2009 is the significant impact of the stock market, which is a powerful variable in the Z-Metrics model and in many other default probability models (notably, Moodys KMV). Recall that the stock markets were at very low levels at the end of 2008 and into the early months of 2009, while there was a major recovery later in 2009 and in early 2010. Table 4 shows, for each of our nine European countries and the U.S., the percentage increases in median stock market index levels and sovereign PD levels between the first six months of 2009 and the first six months
Median percent change (2010 versus 2009)* 24.1% 31.8% 5.5% 26.2% 18.2% 34.4% 17.8% 20.9% 27.8% 31.9% Median Z-Metrics percent change (2010 versus 2009) -23.6% -24.5% -8.4% -7.4% -24.0% -25.3% -22.4% -12.9% -37.6% -43.6%
models comparative PD (delta) results for 2010 and 2009. For eight of the ten countries, use of the private firm model showed smaller reductions in PDs when moving from 2009 to 2010 than use of the public model. Whereas the overall average improvement in PDs for the public firm model was a drop of 1.91 percentage points, the drop was 0.79 percent for our private firm model. These results are largely the effect of the positive stock market performance of late 2009 and into 2010. But improvements in general macro conditions, along with their effects on traditional corporate performance measures, also helped improve (reduce) the PDs. Moreover, in two of these eight countries the U.K. and France not only did the public firm model show an improved (lower) PD, but the private firm models PD actually got worse (increased) in 2010 (as indicated by the positive delta in the last column of Table 5).
Country France Germany Greece Ireland Italy Netherlands Portugal Spain U.K. U.S.
Index CAC40 DAX ASE ISEQ FTSEMIB AEX PSI-20 IBEX35 FTSE100 S&P500
*Median of the various trading day stock index values and PDs, first six months of 2009 versus first six months of 2010. Sources: Z-Metrics Model calculations from RiskMetrics (MSCI) Group, Bloomberg for stock index values.
26
Table 4 Median percentage change in various country stock market index values and Z-Metrics PDs between the first six months of 2010 versus 2009
Correlation of sovereign PDs: recent evidence on Z-metrics versus implied CDS PDs
As a final test of the predictive of our approach, we compared our ZMetrics five-year median PDs for our sample of nine European countries

Number of listed companies PDs Country Netherlands U.K. U.S. France Germany Spain Ireland Italy Portugal Greece Average *Negative sign means improved credit risk. Sources: Table 2 and Riskmetrics (MSCI). 2010 61 442 2226 297 289 82 28 155 30 79 2009 60 433 2171 294 286 78 26 154 30 77 2010 3.33% 3.62% 3.93% 5.51% 5.54% 6.44% 6.45% 7.99% 9.36% 10.60% 6.28%
Public-firm Z-Metrics model PDs 2009 5.62% 5.75% 6.97% 7.22% 7.34% 7.39% 7.46% 10.51% 12.07% 11.57% 8.19% Delta* -2.29% -2.13% -3.04% -1.71% -1.80% -0.95% -1.01% -2.52% 2.71% -0.97% -1.91% PDs 2010 5.25% 6.48% 4.28% 7.33% 6.29% 8.06% 6.31% 8.14% 8.73% 11.03% 7.19%
Private-firm Z-Metrics model PDs 2009 6.00% 5.97% 4.80% 7.19% 7.56% 9.32% 6.36% 9.07% 9.62% 13.93% 7.98% Delta* -0.75% 0.49% -0.52% 0.14% -1.27% -1.26% -0.05% -0.89% -0.89% -2.90% -0.79%
Table 5 Private versus public firm model PDs in 2010 and 2099
(both on a contemporary basis and for 2009) with the PDs implied by CDS spreads in 2010. The contemporary PD correlation during the first third of 2010 was remarkably high, with an R2 of 0.82. This was a period when it was becoming quite evident that certain European countries were in serious financial trouble and the likelihood of default was not trivial. But if we go back to the first half of 2009, the correlation drops to an R2 of 0.36 (although it would be considerably higher, at 0.62, if we excluded the case of Ireland). Irelands CDS implied PD was considerably higher in 2009 than 2010 (17.0 percent versus 12.0 percent), while the Z-Metrics PD was relatively stable in the two years (7.5 percent and 6.5 percent respectively).18 In 2010, whether we calculate the correlation with or without Ireland, the results are essentially the same (0.82 and 0.83). Given the predictive success of Z-metrics in the tests already described, we were curious to find out whether it could be used to predict capital
market (i.e., CDS) prices. So, we regressed our public firm models 2008 Z-Metrics median, non-financial sector PDs against implied CDS PDs one year later in 2009. Admittedly, this sample was quite small (10 countries) and the analysis is for only a single time-series comparison (2008 versus 2009). Nevertheless, these two years spanned a crucial and highly visible sovereign debt crisis, whereas the PDs implied by prior years ZMetrics and CDS showed remarkably little volatility.19 As can be seen in Figure 4, the correlation between our Z-Metrics PDs and those implied by CDS one year later proved to be remarkably strong, with an r of 0.69 and R-square of 0.48. In sum, the corporate health index for our European countries (plus the U.S.) in 2008 explained roughly half of the variation in the CDS results one year later.20 A potential shortcoming of our approach is that we are limited in our private sector corporate health assessments to data from listed, publicly held firms. This is especially true for relatively small countries like
30,00%
Ireland (with just 28 listed companies), Portugal (with 30), Greece (79),
y = 1.9367x - 0.0743 R-square = 48%
Greece
25,00%
20,00%
15,00% Ireland 10,00% U.K. 5,00% Belgium Netherlands 0,00% 0,00% 2,00% 4,00% 6,00% Spain Sweden France U.S. Germany 8,00% 10,00% 12,00% 14,00% Italy Portugal
Source: Table 2
Figure 4 2008 Z-Metrics PD versus 2009 CDS implied PD
18 No doubt the CDS market was reacting quite strongly to the severe problems in the Irish banking sector in 2009, while Z-Metrics PDs were not impacted by the banks. This implies a potential strength of the CDS measure, although the lower CDS implied PD in early 2010 was not impressive in predicting the renewed problems of Irish banks and its economy in the fall of 2010. 19 The last time an entire region and its many countries had a sovereign debt crisis was in Asia in 1997-1998. Unfortunately, CDS prices were not prominent and the CDS market was illiquid at that time. 20 Several other non-linear structures (i.e., power and exponential functions) for our 2009 Z-Metrics versus 2010 CDS implied PDs showed similar results. In all cases, we are assuming a recovery rate of 40% on defaults in calculation of implied sovereign PDs.
27
Netherlands (61), and Spain (82). Since the private, non-listed segment is much larger in all of the countries, we are not clearly assessing the health of the vast majority of its firms and our sovereign health index measure is incomplete.21 But if the size of the listed firm population is clearly a limitation in our calculations, there does not seem to be a systematic bias in our results. To be sure, the very small listings in Ireland, Portugal, and Greece appear heavily correlated with their high PDs, but the country with the lowest PD (the Netherlands) also has a very small listed population. Another potentially important factor is that the listed population in countries like the U.K. and the Netherlands is represented quite heavily by multinational corporations that derive most of their income from outside their borders.22
very least, as a useful complement to existing methods and market indicators one that is not subject to government manipulation of publicly released statistics. Using our approach, the credit and regulatory communities could track the performance of publicly held companies and the economies in which they reside and by making some adjustments, unlisted entities as well. And if sovereigns were also willing to provide independently audited statistics on a regular basis, so much the better.
APPENDIX: Logit Model Estimation of Default Probabilities

We estimated our credit scoring model based on a standard logit-regression functional form whereby: CSi,t = + BjXi,t + i,t CSi,t = Z-Metrics credit score of company i at time t Bj = variable parameters (or weights) Xi,t = set of fundamental, market-based and macroeconomic variables for firm i quarter observations i,t = error terms (assumed to be identically and independently distributed) CSi,t is transformed into a probability of default by PDi,t = 1/[1 + exp(CSi,t)]
(1)
Conclusion and implications

As the price for bailing out distressed sovereigns increases, todays foreign creditors, especially the stronger European nations, are demanding a heavy dose of austerity. Several governments, including those of Greece, Ireland, Spain, Portugal, Italy, and the U.K., have already enacted some painful measures. Others, such as France and Hungary, have either resisted austerity measures or faced significant social unrest when austerity measures have been proposed. These measures typically involve substantial cuts in cash benefits paid to public workers, increases in retirement age, and other reduced infrastructure costs, as well as increased taxes for companies and individuals. The objective is to reduce deficits relative to GDP and enhance the sovereigns ability to repay their foreign debt and balance their budgets. While recognizing the necessity of requiring difficult changes for governments to qualify for bailouts and subsidies, we caution that such measures should be designed to inflict as little damage as possible on the health and productivity of the private enterprises that ultimately fund the sovereign. The goal should be to enable all private enterprises with clear going concern value to pay their bills, expand (or at least maintain) their workforces, and return value to their shareholders and creditors (while those businesses that show no promise of ever making a profit should be either reorganized or liquidated). For this reason, raising taxes and imposing other burdens on corporate entities is likely to weaken the longrun financial condition of sovereigns. To better estimate sovereigns risk of default, we propose that traditional measures of macroeconomic performance be combined with more modern techniques, such as the contingent claims analysis pioneered by Robert Merton and the bottom-up approach presented in these pages. Along with the intuitive appeal of such an approach and our encouraging empirical results, the probabilities of sovereign default provided by ag28 gregating our Z-Metrics across a national economy can be seen, at the
We compare Z-Metrics results with issuer ratings. To ensure a fair comparison, credit scores are converted to agency equivalent (AE) ratings by ranking credit scores and by matching exactly the actual agency rating distribution with the AE rating distribution at any point in time.
We also compare our Z-Metrics results to the well established Altman Z-score (1995) model.23
21 We suggest that complete firm financial statement repositories, such as those that usually are available in the sovereigns central bank be used to monitor the performance of the entire private sector. 22 Results showing the percentage of home-grown revenues for listed firms across our European country sample were inclusive, however, as to their influence on relative PDs. 23 Altmans original Z-score model (1968) is well-known to practitioners and scholars alike. It was built, however, over 40 years ago and is primarily applicable to publicly-held manufacturing firms. A more generally applicable Z-score variation was popularized later [Altman et al. (1995)] as a means to assess the default risk of non-manufacturers as well as manufacturers, and was first applied to emerging market credits. Both models are discussed in Altman and Hotchkiss (2006) and will be compared in several tests to our new Z-Metrics model. Further, the Altman Z-score models do not translate easily into a probability of default rating system, as does the Z-Metrics system. Of course, entities that do not have access to the newer Z-Metrics system can still use the classic Z-score frameworks, although accuracy levels will not be as high and firm PDs not as readily available.

References

Abassi, B. and R. J. Taffler, 1982, Country risk: a model of economic performance related to debt servicing capacity, WP #36, City University Business School, London Altman, E. I., 1968, Financial ratios discriminant analysis and the prediction of corporate bankruptcy, Journal of Finance, v. 23:4, 589-609 Altman, E. I., 2001, Italy: the hero or villain of the Euro, Insight, Financial Times, June 21 Altman, E. I. and E. Hotchkiss, 2006, Corporate financial distress and bankruptcy, 3rd edition, John Wiley & Sons, NY and NJ Altman, E. I., H. Rijken, M. Watt, D. Balan, J. Forero, J. Mina, 2010, The Z-Metrics Methodology for Estimating Company Credit Ratings and Default Risk Probabilities, RiskMetrics Group, NY, June Babbel, D. F., 1996, Insuring sovereign debt against default, World Bank Discussion Papers, #328 Baek, I., A. Bandopadhyaya, and C. Du, 2005, Determinants of market-assessed sovereign risk: economic fundamentals or market risk appetite? Journal of International Money and Finance, 24:4, 533-48 Beers, D., M. Cavanaugh, and O. Takahira, 2002, Sovereign credit ratings: a primer, Standard & Poors Corp., NY, April Bertozi, S., 1996, An annotated bibliography on external debt capacity, in Babbel, D., 1996, Insuring sovereign debt against default, World Bank Discussion Papers #328 Caouette, J., E. Altman, P. Narayanan, and R. Nimmo, 2008, Managing credit risk, 2nd edition, John Wiley & Sons, NY Chambers, W. J., 1997, Understanding sovereign risk, Credit Week, Standard & Poors January 1 Cline, W., 1983, A logit model of debt restructuring, 1963-1982, Institute for International Economics, WP, June Feder, G. and R. E. Just, 1977, A study of debt servicing capacity applying logit analysis, Journal of Development Economics, 4:1, 25-38 Feder, G. R., E. Just, and K. Ross, 1981, Projecting debt capacity of developing countries, Journal of Financial and Qualitative Analysis, 16:5, 651-669 Flynn, D., 2009, S&P cuts Greek debt rating as global crisis bites, Reuters, January 14 Frank, C. R. and W. R. Cline, 1971, Measurement of debt servicing capacity: an application of discriminant analysis, Journal of International Economics, 1. 32744 Frenkel, M., A. Karmann and B. Scholtens (eds.), 2004, Sovereign risk and financial crises, Heidelberg and New York, Springer Gennaioli, N., A. Martin and S. Rossi, 2010, Sovereign default, domestic banks and financial institutions, Working Paper, Imperial College, London Gerlach, S., A. Schultz, and G. Wolff, 2010, Banking and sovereign risk in the Euro Area, Deutsche Bundesbank, Research Centre, Discussion Paper Series 1: Economic Studies Gray, D. F., R. Merton, and Z. Bodie, 2006, A new framework for analyzing and managing macrofinancial risk of an economy, IMF Working Paper, October Gray, D. F., R. Merton and Z. Bodie, 2007, Contingent claims approach to measuring and managing sovereign credit risk, Journal of Investment Management, 5:4, 5-28 Grinols, E., 1976, International debt rescheduling and discrimination using financial variables, U.S. Treasury Dept., Washington, D.C. Haugh, D., P. Ollivaud, and D. Turner, 2009, What drives sovereign risk premiums? An analysis of recent evidence from the Euro Areas, OECD, Economics Department, Working Paper, 718 Hilscher, J. and Y. Nosbusch, 2010, Determinants of sovereign risk: macroeconomic fundamentals and the pricing of sovereign debt, Review of Finance, 14:2, 235-62 IMF, 2010, Global financial stability report, Washington, D.C. KMV Corporation, 1999, Modeling default risk, KMV Corporation Krugman, P., 1988, Financing vs. forgiving a debt overhang: some analytical notes, Journal of Development Economics, 29, 253-68 Longstaff, F., J. Pan, L. Pedersen and K. Singleton, 2007, How sovereign is sovereign credit risk? National Bureau of Economic Research, Inc., NBER Working Paper: 13658 Merton, R. C., 1974, On the pricing of corporate debt: the risk structure of interest rates, Journal of Finance, 29:2, 449-70 Neziri, H., 2009, Can credit default swaps predict financial crises? Journal of Applied Economic Sciences, 1:7, 125-136 Oshiro, N. and Y. Saruwatari, 2005, Quantification of sovereign risk: using the information in equity market prices, Emerging Markets Review, 6:4, 346-62 Pomerleano, M., 1998, Corporate finance lessons from the East Asian Crisis, Viewpoint, The World Bank Group, Note #155, October Pomerleano, M., 1999, The East-Asia Crisis and corporate finance the untold micro study, Emerging Markets Quarterly Reinhart, M. and K. Rogoff, 2010, This time is different, Princeton University Press, Princeton, NJ.
Remolona, E., M. Scatigna, and E. Wu, 2008, A ratings-based approach to measuring sovereign risk, International Journal of Finance and Economics, 13:1, 26-39 Saini, K. and P. Bates, 1978, Statistical techniques for determining debt servicing capacity for developing countries: analytical review of the literature and further empirical results, Federal Reserve Bank of New York Research Paper, #7818 Sargen, H., 1977, Economics indicators and country risk appraisal, Federal Reserve Bank of San Francisco, Economic Review, Fall Schmidt, R., 1984, Early warning of debt rescheduling, Journal of Banking and Finance, 8, 357-370 Segoviano, B., A. Miguel, C. Caceres, and V. Guzzo, 2010, Sovereign spreads: global risk aversion, contagion or fundamentals? IMF Working Paper: 10/120 Sgherri, S. and E. Zoli, 2009, Euro Area sovereign risk during the crisis, International Monetary Fund, IMF Working Papers: 09/222 Smith, R. and I. Walter, 2003, Global banking, Oxford University Press, London Trebesch, C., U. Das, and M. Papaioannou, 2010, Sovereign default risk and private sector access to capital in emerging markets, IMP Working Paper: October
29
PART 1
International Liquidity Provision and CurrencySpecific Liquidity Shortages

1
Richhild Moessner Senior Economist, Bank for International Settlements and

Cass Business School
William A. Allen Honorary Visiting Senior Fellow, Cass Business School
Abstract
In this paper we discuss the main innovation in central bank cooperation during the financial crisis of 2008-09, namely the emergency provision of international liquidity through the establishment of bilateral central bank swap facilities, which have evolved to form interconnected swap networks. Based on the BIS international locational banking statistics, we present a measure of currency-specific liquidity shortages for the U.S. dollar, the euro, the yen, the pound sterling, and the Swiss franc for a large number of advanced and emerging economies. We discuss the reasons for establishing swap facilities, relate our measure of currency-specific liquidity shortages to the probability of a country receiving a swap line in that currency, and find a significant relationship in the case of the U.S. dollar, the euro, the yen, and the Swiss franc. We find that countries with larger U.S. dollar shortages on our measure, and economies that are large international
financial centers, have a statistically significantly higher probability of receiving a U.S. dollar swap line. We also find that actual U.S. dollar funding obtained by drawing on the Feds swap lines at end-2008 was statistically significantly larger for economies with higher U.S. dollar shortages on our measure, as well as for economies which are large international financial centers.
The views expressed are those of the authors and should not be taken to reflect those of the BIS. The authors would like to thank Naohiko Baba, Corinne Ho, Robert McCauley, Petra Gerlach, Philip Turner and seminar participants at the BIS, Cass Business School, the European Finance Association Annual Meeting 2010, the Hungarian National Bank, the IMF, the London School of Economics, and the U.K. Financial Services Authority for helpful comments. We would also like to thank Bilyana Bogdanova and Swapan Pradhan for excellent research assistance. An earlier version of this paper has been published as BIS Working Paper no 310.
31
The advent of the credit crunch in August 2007, and its subsequent intensification, has largely eroded the hitherto apparently sharp distinction between monetary and financial stability, and has led to a revival of central bank cooperation. In this paper, we study the main innovation in central bank cooperation during the financial crisis of 2008-09, namely the establishment of bilateral central bank swap facilities, which have evolved to form interconnected swap networks. Obstfeld et al. (2009) describe these facilities as one of the most notable examples of central bank cooperation in history. The credit crisis was initiated by a widespread, though not uniform or complete, loss of confidence in the creditworthiness of banks. It began suddenly in August 2007, and varied in intensity throughout the following year. Perceived counterparty credit risks increased sharply, owing to uncertainty about other banks credit exposures and the size of potential losses, and banks started hoarding liquidity. Spreads between Libor rates and Overnight Index Swap rates (OIS) widened and became highly volatile. The credit crisis damaged the functioning of all financial markets, including the wholesale deposit and foreign exchange swap markets [Baba et al. (2008); Baba and Packer (2009); and Allen and Moessner (2010)]. The crisis became much more acute after the failure of Lehman Brothers in September 2008, which destroyed the widespread belief in financial markets that governments would not allow any systemically-important financial institution to fail, and thereby dramatically heightened perceptions of credit risk among trading counterparties in financial markets. In many countries, banks had made loans in foreign currencies, particularly those currencies in which interest rates had been relatively low, notably the U.S. dollar, the yen, and the Swiss franc. They had financed those loans partly by taking deposits in the currency of the loan, typically in the international wholesale deposit market, and partly by taking deposits in their home currencies, and using the foreign exchange swap market to eliminate foreign exchange risk. When the credit crisis struck, it became much more difficult, or in some cases impossible, for many banks to secure foreign currency deposits in the wholesale markets. Even in domestic currency markets, the available range of maturities became much shorter. Many banks were forced to use the lending facilities of their home central banks to finance themselves. Such facilities were in normal times typically confined to their domestic currency and to short maturities. Consequently, in the absence of any new special facilities designed to help them, banks would have had to replace relatively long-maturity foreign currency financing of foreign currency assets with relatively short-maturity domestic currency financing. The financial market consequences would have made the disruption caused by the credit crisis substantially more serious, as described in Allen and Moessner (2010), and the main purpose of the swap networks 32 was to avoid those consequences.
Swap facilities can be used as a means of making the provision of central bank liquidity more effective by extending its geographical scope. Typically, central bank lending to domestic commercial banks is denominated in domestic currency, but if the commercial banks need foreign currency liquidity, then something more is required if the central bank wants to address this need. Swap facilities enable a central bank to provide liquidity to domestic banks in foreign currency.2 In this paper, we study the emergency provision of international liquidity through the establishment of bilateral central bank swap facilities, which have evolved to form interconnected swap networks. We present a measure of currency-specific liquidity shortages for the U.S. dollar, the euro, the yen, the pound sterling, and the Swiss franc for a large number of advanced and emerging economies, based on the BIS international locational banking statistics. We discuss the reasons for establishing swap facilities, relate our measure of currency-specific liquidity shortages to the probability of a country receiving a swap line in that currency, and find a significant relationship in the case of the U.S. dollar, the euro, the yen, and the Swiss franc. Moreover, we find that economies which are large international financial centers have a statistically significantly higher probability of receiving a U.S. dollar swap line from the Federal Reserve, as well as from any country. We also find that actual U.S. dollar funding obtained by drawing on the Feds swap lines at end-2008 was statistically significantly larger for economies with higher U.S. dollar shortages on our measure, as well as for economies which are large international financial centers.
Incidence of currency-specific liquidity shortages evidence from the BIS international banking statistics
In normal market conditions, commercial banks can readily convert liquidity from one currency into another using foreign exchange swap markets. Thus, a bank which is in need of foreign currency liquidity but can only get domestic currency liquidity can swap the domestic currency into foreign currency using the commercial swap market, selling the domestic currency spot and buying it forward. However, commercial swap market liquidity was seriously impaired during the credit crisis, partly by concerns about settlement risk, and currency-specific liquidity shortages developed in many countries. Currency-specific liquidity shortages occurred when commercial banks needed to replace foreign currency deposits (including deposits taken in wholesale markets) which had been withdrawn, but were not able to do
Alternatively, the central bank could use some of its own foreign exchange reserves for that purpose, converting them into the required currency if necessary by means of market transactions.

International Liquidity Provision and Currency-Specific Liquidity Shortages
so. A number of advanced and emerging economies experienced currency-specific shortages as the financial crisis intensified.3 In principle, the size of the currency-specific liquidity shortage in any country is equal to the following: i. Banks total liabilities in the currency in question minus
200 New Zealand South Korea Indonesia Euro Area Lithuania Romania Australia 100 Hungary Bulgaria Sweden Canada Norway Estonia Iceland Poland Russia Turkey
Brazil
Chile
India
Peru
UK
0 Latvia Malaysia Czech Republic Mexico Philippines South Africa China Switzerland Hong Kong SAR Denmark Colombia Thailand Singapore Japan Chinese Taipei Venezuela Argentina
-100
-200
ii. Banks total illiquid liabilities in that currency minus iii. The total funds in that currency that banks can raise from depositors, from their affiliates or from other providers, including central banks, or by means of asset sales. Item (i) minus item (ii) are the banks liabilities that need to be refinanced, and item (iii) is the funding that can be raised for this refinancing. The difference between the two, [(i)-(ii)]-(iii), is, therefore, the currency-specific liquidity shortage. In practice, none of these components is available in published statistics. The sizes of currency-specific liquidity shortages have to be estimated using such proxy data as are available. As a proxy measure of U.S. dollar-specific shortages, Figure 1 shows the net outstanding U.S. dollar cross-border claims on BIS reporting banks by the economies shown, defined as cross-border total liabilities minus claims (in both foreign and domestic currency) of all BIS reporting banks vis--vis banks and non-banks located in the countries shown at end2008. The corresponding proxy measure of euro-specific liquidity shortages is shown in Figure 2, and the corresponding measures for the yen, the pound sterling and the Swiss franc are shown in Figures 3 to 5. On this measure, the largest currency-specific liquidity shortage was of U.S. dollars in the Euro Area (around U.S.$400 billion). The next largest were the shortage of yen in the United Kingdom (U.S.$90 billion equivalent), that of euros in the U.S. (about U.S.$70 billion equivalent), and that of Swiss francs in the Euro Area (about U.S.30 billion equivalent). There were only small shortages of pound sterling (with the largest being around U.S.$6 billion equivalent for Norway). The measure used in Figures 1 to 5 is based on the BIS locational international banking statistics by residence of counterparty. These statistics record the aggregate international claims and liabilities of all banks resident in the BIS reporting countries broken down by instrument, currency, sector, country of residence of counterparty, and nationality of reporting banks. Both domestic and foreign-owned banking offices in the reporting countries report their positions gross (except for derivative contracts for which a master netting agreement is in place) and on an unconsolidated basis, i.e. including banks positions vis--vis their own affiliates.4 There are several reasons why we base our measure of currency-specific shortages on these statistics. First, in a financial crisis gross positions can matter, including of banks vis--vis their subsidiaries, rather than just net
3 4 5 U.S. dollar shortages have been analysed in McGuire and von Peter (2009). This is consistent with the principles of national accounts, money and banking, balance of payments, and external debt statistics [BIS (2008)]. Some countries supervisors would like the subsidiaries of foreign banks in their country to hold their own liquidity, because it would fall to the host countrys authorities to undertake any bailout of subsidiaries of foreign banks. Moreover, it has been suggested that the unwinding of global banks in an emergency would be facilitated if their subsidiaries managed their own liquidity and funding needs, modeled on the example of some large global banks (such as HSBC) which are currently set up as holding companies [Pomerleano (2009)].
250 200 150 Czech Republic 100 50 0 Peru Malaysia Chinese Taipei Russia Singapore Japan UK Chile Mexico Switzerland China Hong Kong SAR Colombia Thailand Venezuela Argentina -50 -100 -300 -400
Figure 1 Net outstanding U.S. dollar cross-border claims on BIS reporting banks by economies shown1 (in billions of U.S. dollars, December 2008)
Sources: BIS locational international banking statistics, authors calculations.
Figure 2 Net outstanding euro cross-border claims on BIS reporting banks by economies shown (in billions of U.S. dollars, December 2008)
positions. Many bank assets are normally illiquid. And head offices may be unable or unwilling to provide their subsidiaries with necessary liquidity during a crisis (indeed they might withdraw it).5 Moreover, because of differences in time zones, subsidiaries may not have timely access to liquidity from their head office. Secondly, our measure of currency-specific shortages is available for a wide range of countries, including many emerging economies which are not BIS reporting countries, and it is available for the U.S. dollar, the euro, the yen, the pound sterling, and the Swiss franc. By contrast, currency-specific net foreign positions of banks by nationality of head office based on the BIS locational international
United States
New Zealand
South Korea
South Africa
Philippines
Indonesia
Lithuania
Denmark
Romania
Australia
Hungary
Bulgaria
Sweden
Canada
Norway
Estonia
Iceland
Poland
Turkey
Latvia
Brazil
India
33
statistics by nationality of head office and the BIS consolidated international banking statistics6 (see McGuire and von Peter (2009)) are available only for a subset of BIS reporting countries, and only for the U.S. dollar, the euro and the yen, but not the pound sterling or the Swiss franc.
Central bank swap networks

The central banks response to currency-specific liquidity shortages was to set up swap facilities so that the home central bank of the currencies in short supply could provide those currencies to the commercial banks outside the home country that needed them. They did so indirectly, using as intermediaries the central banks of the commercial banks that were short of liquidity. In effect, they used foreign central banks to extend the geographical scope of their liquidity-providing operations.7
100
The mechanics of an inter-central bank swap are very simple. Central bank A credits the account of central bank B in its own books with As currency; in return, central bank B credits the account of central bank A in its books with an equivalent amount of Bs currency. Thus A lends its currency to B and B lends its currency to A; each loan is collateral for the other. There may be a provision for the amounts of the loans to be adjusted as exchange rates change. In principle, both A and B may use the foreign currency which the swap has put at their disposal, but in practice, only one party normally uses the swap proceeds; the other party simply holds them on deposit as collateral for the loan. In total, four overlapping swap networks were established:
South Africa Chinese Taipei Peru Mexico Switzerland China Hong Kong SAR Euro area Venezuela Argentina Lithuania Romania Bulgaria Estonia
Czech Republic
United States
New Zealand
50
South Korea
Philippines
Singapore
Indonesia
Colombia
Denmark
Malaysia
Australia
Thailand
Hungary
Sweden
Canada
Norway
Iceland
Poland
Russia
Turkey
Latvia
Brazil
Chile
India
UK 0 -50 -100
Figure 3 Net outstanding Japanese yen cross-border claims on BIS reporting banks by economies shown (in billions of U.S. dollars, December 2008)
40
30
The Fed network, set up to supply dollars (the Fed also set up swap facilities with certain foreign central banks under which it could obtain foreign currencies from them).
20 New Zealand
10 Canada Norway Euro area
The euro network, under which the ECB supplied euros. There were also what we regard as extensions to the euro network enabling Danmarks Nationalbank, Norges Bank, and Sveriges Riksbank to provide euros to other central banks.
South Korea
Colombia
Denmark
Romania
Australia
Hungary
Sweden
India
China
Singapore
Japan
United States
Hong Kong SAR
Euro area
Thailand
Chinese Taipei
Lithuania
Bulgaria
Estonia
Poland
Indonesia
Russia
Brazil
Iceland
Malaysia
Latvia
South Africa
Peru
Turkey
Philippines
Chile
Czech Republic
Switzerland
Mexico
Venezuela
Argentina
-10
The Swiss franc network. The Asian and Latin American network.
Figure 4 Net outstanding pound sterling cross-border claims on BIS reporting banks by economies shown (in billions of U.S. dollars, December 2008)
The entire network of swap facilities is illustrated in Figure 6, and the swap lines set up are listed in Table A1 in the appendix.
The Fed network

10 Lithuania Denmark Romania Australia Hungary Sweden Canada Norway Estonia Iceland Poland Turkey Latvia
The Federal Reserve was the first in the field. It set up its first swap
Philippines
Czech Republic
New Zealand
South Korea
lines in December 2007, and the number and size of its swap lines increased steadily in the following months. In reporting the initial phase of
Bulgaria Colombia Thailand China Chinese Taipei Hong Kong SAR Singapore Japan United States Brazil Malaysia South Africa Peru Chile Argentina
Venezuela
Indonesia
Mexico
Russia
India
UK
-10
-20
-30
34
Figure 5 Net outstanding Swiss franc cross-border claims on BIS reporting banks by economies shown (in billions of U.S. dollars, December 2008)
The BIS consolidated international banking statistics provide information on the country risk exposures of the major banking groups of various countries vis--vis the rest of the world. The consolidated banking statistics report banks on-balance sheet financial claims vis-vis the rest of the world and provide a measure of the risk exposures of lenders national banking systems. The data cover contractual and ultimate risk lending by the head office and all its branches and subsidiaries on a worldwide consolidated basis, net of inter-office accounts [BIS (2009)]. By doing so, they avoided the credit risk of lending directly to foreign commercial banks. That risk was taken by the intermediary central banks.

However, we did explicitly coordinate to address problems in dollar

Poland Switzerland Hungary UK Norway
funding markets. The Federal Reserve entered into foreign exchange

ECB
swaps with a number of other central banks to make dollar funding available to foreign banks in their own countries. By doing so, we reduced the pressure on dollar funding markets here at home [Kohn (2009)].
Canada
USA
Mexico
Sweden
Iceland Estonia
During this period, foreign commercial banks were a source of heavy demand for U.S. dollar funding, thereby putting additional strain on global bank funding markets, including U.S. markets, and further squeezing credit availability in the United States. To address this problem, the Federal Reserve expanded the temporary swap lines that had been estab-
Denmark Brazil
Singapore
Latvia
India
Australia
NZ
Japan Korea China

Hong Kong
lished earlier with the European Central Bank (ECB) and the Swiss National Bank, and established new temporary swap lines with seven other central banks in September and five more in late October, including four in emerging market economies [Bernanke (2009)]. There is a relatively large Eurodollar market, i.e., a market for deposits denominated in U.S. dollars outside the U.S., and there is some evidence that Eurodollar interest rates could affect domestic U.S. short-term interest rates [Hartman (1984)]. This would be one channel through which U.S. dollar funding problems of foreign banks could affect domestic U.S. dollar funding markets. The minutes of the conference call held by the Federal Open Market Committee on 6 December 2007, at which it was decided to establish the first of the swap lines, records that the swap proposal was aimed at improving market functioning [Federal Open Market Committee (2007)]. The extension of swap lines by the Federal Reserve took place in four main phases, as market liquidity deteriorated. The first swap lines were set up in December 2007, and they were extended, both in size and in geographical spread, in March 2008, May 2008, and September/October 2008. The last phase of extensions was by far the largest. It followed the failure of Lehman Brothers on 15th September. In response to the ensuing deterioration in market conditions, it was announced that foreign central banks (the ECB and the central banks of Japan, Switzerland, and the U.K.) would auction term and forward dollar funding, in parallel with the Feds domestic Term Auction Facility. To facilitate these auctions, the upper limits on the amounts of the Feds swap lines with these central banks were removed entirely. In April 2009, the Fed announced that, as a precautionary measure, it had established swap lines to receive foreign currency from the ECB and the central banks of Switzerland, the U.K., and Japan, so that it would
Belarus
Malaysia
Indonesia
Argentina
Figure 6 Swap facilities network
the extension of swap facilities in December 2007, the Federal Reserve Bank of New York commented as follows: From mid-November to yearend, trading liquidity in the foreign exchange swaps market was severely impaired. The re-emergence of funding pressures in term dollar, euro, and pound sterling money markets caused by balance sheet constraints and typical year-end funding pressures made it difficult to identify the appropriate interest rates at which to price forward transactions. These factors were exacerbated by increased demand for dollar funding by offshore banks that are typically structurally short U.S. dollars and that use the foreign exchange swaps market to obtain such funding. As a result, trading volumes in the foreign exchange swaps market diminished considerably, trade sizes contracted, and bid-ask spreads on transactions became much wider than normal. Additionally, concerns about counterparty credit risk prompted some market makers to temporarily withdraw from the market. Credit tiering also became evident, with counterparties viewed as less creditworthy finding it more difficult and costly to enter into transactions than counterparties perceived to be more creditworthy. Despite the impairment to the swaps market, spot foreign exchange market liquidity for major currencies was generally healthy during the quarter [Federal Reserve Bank of New York (2008)]. The swap network was part of a broader program of facilities that the Fed established to provide liquidity to financial markets. Access to other Federal Reserve liquidity facilities is confined to banks and primary securities dealers in the U.S.8, so that banks outside the U.S. needing to raise dollars did not have access to them. The swap lines established by the Federal Reserve also had the aim of reducing U.S. dollar funding market pressure in the U.S., as the following statements make clear:
For information about access to the discount window, see Federal Reserve System, Board of Governors (2005), http://www.federalreserve.gov/pf/pdf/pf_complete.pdf. For information about access to the Term Auction Facility and the primary dealer credit facility, see http://www.federalreserve.gov/newsevents/press/monetary/20071212a.htm and http://www.federalreserve.gov/newsevents/press/monetary/20080316a.htm, respectively.
35
have the means to relieve shortages of foreign currencies in the U.S. should they arise.9 These swap lines were not used in the second or third quarters of 2009. All of the Feds temporary swap lines were repaid and allowed to expire on 1 February, 2010. However, in May 2010, in response to the re-emergence of strains in U.S. dollar short-term funding markets in Europe related to market concerns about sovereign debt, the Fed re-established swap lines with the central banks of Canada, the UK, the Euro Area, Switzerland, and Japan . We do not discuss this episode in this paper.
10
and Magyar Nemzeti Bank on 25 June 2009, the SNB stated in its Monetary Policy Report that the aim of this measure is to further ease the situation on the short-term Swiss franc money market [SNB (2009)]. The provision of the swap facilities probably also partly reflected the SNBs concern about the appreciation of the Swiss franc, as well as its concern about conditions in credit markets. Indeed, on 12th March, 2009, the SNB announced that it would act to prevent a further appreciation of the Swiss franc against the euro, including by purchasing foreign currency on the foreign exchange markets.12 It, therefore, seems highly likely that the provision of Swiss franc swap facilities to the ECB and to the central banks of Hungary and Poland was partly motivated by the same concern.
The euro network

Euro-specific liquidity shortages developed in several European countries outside the Euro Area; the market symptoms are noted in Allen and Moessner (2010). The ECB set up facilities with the Danish, Hungarian, Polish, and Swedish central banks to assist commercial banks in those countries in getting access to euro liquidity and thereby relieving localized shortages. In Hungary and Poland, commercial banks had made extensive domestic mortgage loans in foreign currencies, financing themselves in wholesale markets which became much less liquid as the credit crisis intensified.
11
The Asian and Latin American network

Before the credit crisis began, there was already an extensive network of inter-central bank swap lines in East Asia, created since 2000 under the Chiang Mai initiative [Kawai (2007)]. These facilities were set up after the Asian financial crisis of 1997-98 in order to enable East Asian central banks to provide mutual financial support in the event of a future crisis, and they are part of a larger program of economic integration in East Asia, as Kawai (2007) describes. The provisions of the swap facilities are fairly conservative, in that only the first 20 percent of the committed amount is available immediately. The remainder is provided under an IMF program. To the authors knowledge, no drawings were made under this network during 2007-09. The Chiang Mai network needed to be supplemented to address the pressures created by the credit crisis.13 During the crisis, the Bank of Japan established yen swap lines with the U.S. and Korea (Figure 6), which were among the six economies with the largest yen shortages according to the measure shown in Figure 3. In addition, in June 2008, the Bank of Japan, acting as agent for the Ministry of Finance established a U.S. dollar swap line with India. The Peoples Bank of China was active in establishing new swap lines during the crisis. It appears to have had two separate objectives: first,
Under the ECBs arrangements with Hungary and
Poland, euros were provided against euro collateral, rather than against the national currencies of the counterparty central banks. In addition to the swap lines provided by the ECB, further swap lines were provided indirectly to certain other European countries in an extended euro swap network (Figure 6), via central banks in countries with which the ECB had established swap lines (Sweden and Denmark), as well as separately by Norway, which had no swap line with the ECB.
The Swiss franc network

The Swiss National Bank provided Swiss franc swap lines to the U.S., the Euro Area, Poland, and Hungary (Figure 6). Under the SNBs swap lines with Hungary and Poland, Swiss francs were provided against euro collateral, not against the national currency of the counterparty central bank. The SNBs purpose in providing these swap facilities was to enable foreign central banks to provide their commercial banks with Swiss franc liquidity and thereby satisfy the strong demand for Swiss francs [Roth (2009)]. The SNB implements its monetary policy by fixing a target range for the three-month Swiss franc Libor rate. The SNB reduced the upper bound of its target range from above 3 percent to 0.75 percent in the course of the financial crisis, and sought to bring down the Libor rate within this target range. However, the efforts of foreign banks to obtain the Swiss franc funding that they needed put upward pressure on the Libor rate. Easing the Swiss franc funding problems of foreign banks by providing swap lines was, therefore, expected to help bring down the SNBs policy rate within the target range, thereby aiding in achieving the SNBs monetary policy objectives. Regarding the extension of the EUR/ 36 CHF foreign exchange swaps with the ECB, National Bank of Poland,
10
11
12 13
In each case, it already had in place a swap line under which it could supply dollars to the foreign central bank in question. Since the Fed described its dollar-supplying swap lines as reciprocal currency arrangements, it might be thought that the earlier swap lines, though initially set up so that the Fed could supply dollars, could be used in the reverse direction to enable the Fed to receive foreign currencies. However, that was evidently not the case. See Fed press releases http://www.federalreserve.gov/newsevents/press/ monetary/20100509a.htm and http://www.federalreserve.gov/newsevents/press/ monetary/20100510a.htm. See, for example, the 2007 IMF Article 4 reports on Hungary and Poland (http://www. imf.org/external/pubs/ft/scr/2007/cr07250.pdf and http://www.imf.org/external/pubs/ft/ scr/2008/cr08130.pdf). See also the NBPs Financial Stability Report, October 2008 (http:// www.nbp.pl/en/SystemFinansowy/Financial_Stability_October2008.pdf). See http://www.snb.ch/en/mmr/reference/pre_20090312/source. In late 2009, the Chiang Mai network was converted from a network of bilateral facilities into a multilateral facility. Each member country was assigned a contribution amount and a purchasing multiplier, and is able to draw U.S. dollars up to the product of its contribution amount and purchasing multiplier. See, for example, http://www.pbc.gov.cn/ english/detail.asp?col=6400&id=1451.

to help in dealing with financial stress, and second, to promote bilateral trade and investment in the partner countries own currencies, with a view to establishing these currencies as international trading and investment vehicles in the longer term. It is reasonable to believe that the PBOCs pursuit of both these objectives was motivated by the financial crisis. The need for liquidity was obvious. And the desire to promote nondollar currencies as trading and investment vehicles is consistent with the views on international monetary reform expressed by the Governor of the PBOC in a speech on 23rd March 2009 [Zhou (2009)]. These objectives were set out in the English versions of the PBOCs announcements of the establishment of the various swap lines, as Table 1 shows. The PBOCs second objective, of promoting bilateral trade in the trading partner countries own currencies, with a view to establishing their own currencies as international trading vehicles, is by its nature a longer-term project, and this is reflected in the fact that the PBOCs swaps all have three-year terms, much longer than the terms of the swaps set up by other central banks purely to address market liquidity strains. Only few economies had shortages in the pound sterling according to the measure shown in Figure 4, while many economies were close to balance in the pound sterling or had surpluses. This is consistent with the absence of any swap lines in the pound sterling (other than the swap line provided by the Bank of England to the Fed in April 2009, which was not used).
Date
Counterparty
Short-term liquidity? Yes Yes No No Yes No
Bilateral trade? Yes Yes1 Yes2 Yes2 Yes3 No
12 Dec 2008 20 January 2009 8 February 2009 11 March 2009 23 March 2009 2 April 2009
1
Bank of Korea Hong Kong Monetary Authority Bank Negara Malaysia National Bank of Belarus Bank Indonesia Central Bank of Argentina
This will bolster investor confidence in Hong Kongs financial stability, promote regional financial stability and the development of yuan-denominated trade settlement between Hong Kong and the mainland. 2 Announcement refers to bilateral trade and investment. 3 Announcement refers to bilateral trade and direct investment. Sources: English versions on PBOC internet site.
Table 1 Language of PBOC swap announcements
i.e., the U.S. dollar, euro, yen, or the Swiss franc, and sci is the measure of the shortage in that currency (where a shortage is taken here as positive), in billions of U.S. dollars or U.S. dollar equivalent.14 The estimated coefficients in the regression in equation (1) are shown in Table 2. We find that the probability of a country receiving a swap line in the euro, yen, and the Swiss franc depends significantly on the measures of currency-specific shortage in the currency considered. The relationship is significant at the 1 percent level for the euro and Swiss franc, and at the 5 percent level for the yen. The coefficient b2 on the currency-specific shortage is largest for the Swiss franc, followed by the euro and the yen. Consistent with this, the goodness-of-fit of the probit model, as measured by the McFadden R2 measure, is largest for the Swiss franc (at 0.44), followed by the euro (at 0.41) and the yen (at 0.19). The marginal effect of the estimated currency shortage on the probability of receiving a swap line in that currency is given by [Verbeek (2004)]: F(xci b)/sci =f(xci b)*b2 (2)
Relationship between currency-specific shortages and central bank swap lines

In this section we study the relationship between the level of a countrys currency-specific shortages based on the BIS locational international banking statistics by residence of counterparty as shown in Figures 1 to 5, and whether the country received a swap line in that currency. We do so by considering a probit regression model of the dependent variable, yci, which equals 1 if country i received a swap line in the currency, c, under consideration and 0 otherwise, on the level of countries currency shortages, sci, as well as on a constant term. The sample of countries consists of those included in Figures 1 to 5. We define the vector of explanatory variables as xci = (1, sci). The probit model models the probability that a country i receives a swap line in currency c, yci =1, as a function of the explanatory variables, in our case a constant term and the currency-specific shortage, sci, according to P(yci =1| xci )= F(xci b) (1)
where f(.) denotes the standard normal density function. This marginal effect depends on the value of the shortage. For the values of the shortages in our sample of countries, it ranges from close to zero to around 0.05 for the Swiss franc, to around 0.02 for the euro, and to around 0.01 for the yen. Next we estimate the probability that a country i receives a swap line in currency c, yci =1, as a function of the explanatory variables, a constant
where F(.) is the standard normal cumulative distribution function, the vector of coefficients is b = (b1, b2), c denotes the currency considered,
14 We do not consider sterling since no swap lines were granted in that currency, with the single exception of the swap line provided by the Bank of England to the Fed in April 2009.
37
U.S. dollar Constant, b1 Currency-specific shortage, b2 McFadden R2 Number of observations -0.25 (0.21) 0.004 (0.003) 0.04 39
Euro -1.87** (0.50) 0.043** (0.014) 0.41 39
Yen -1.73** (0.38) 0.021* (0.01) 0.19 39
Swiss franc -1.85** (0.40) 0.130** (0.048) 0.44 39
By contrast, in the specifications reported in Tables 2 and 3 the relationship between the currency-specific shortage and the probability of receiving a swap line is not significant for the U.S. dollar. One possible explanation is related to differences in time zones. In countries with time zones remote from the U.S., U.S. financial markets are closed during part or all of the trading day. This is, for example, the case in the mornings in European countries. During times when U.S. markets are closed, commercial banks with U.S. dollar shortages in such time zones, for example in Europe, are likely to have tried to obtain U.S. dollar funding in the markets of other large international financial centers outside the U.S., such as Japan and Singapore. Thus U.S. dollar shortages were
** and * denote significance at the 1% and 5% level, respectively; standard errors are given in brackets.
Table 2 Results for probit model by currency
U.S. dollar Constant, b1 Currency-specific shortage, b2 McFadden R2 Number of observations -0.41 (0.34) 0.006 (0.005) 0.04 39
Euro -3.17** (0.95) 0.073** (0.025) 0.39 39
Yen -3.04** (0.80) 0.036* (0.018) 0.18 39
Swiss franc -3.53** (0.99) 0.252** (0.097) 0.46 39
likely to have been passed from one time zone to another. An international financial center which initially had a dollar surplus might experience large inter-bank outflows which had the effect of turning the surplus into a shortage. Furthermore, commercial banks in such countries may be less likely to have affiliates in the U.S. from which they could obtain U.S. dollar liquidity, and would therefore be more likely to look for U.S. dollar funding outside the U.S. Consequently, given the international role of the U.S. dollar, the Federal Reserve may have supplied U.S. dollar funding via swap lines to large international financial centers, so as to ensure that the latter could distribute U.S. dollar liquidity on to commercial banks in time zones remote from the U.S. while U.S. markets were closed.
Table 3 Results for logit model by currency
term, and the estimated currency-specific shortage, sci , using the logit model, which is an alternative binary choice model to the probit model where the standard normal probability distribution function is replaced by a logistic probability distribution function: P(yc =1| xc )= F(xci b) (3)
To test this hypothesis, we add a dummy variable, dilfc, in the probit regression for the probability of the country receiving a U.S. dollar swap line from any country, which equals one if an economy is a large international financial center (i.e., Australia, the Euro Area, Hong Kong, Japan, Singapore, Switzerland, and the U.K.), and zero otherwise. The vector of explanatory variables in the probit regression of equation (1) is now defined as x$i = (1, s$i, dilfc), and the vector of coefficients is b = (b1,b2,b3). The results are shown in the middle column of Table 4. We can see that the coefficient on the dummy variable for an economy being a large international financial centre is statistically significant at the 1 percent level, consistent with our hypothesis. Moreover, when controlling for whether a country is a large international financial centre, our measure of the U.S. dollar shortage becomes statistically significant at the 5 percent level
Here, F(.) is the standard logistic distribution function, F(w)=exp(w)/ (1+exp(w)). In the logit model, the marginal effect of the estimated currency shortage on the probability of receiving a swap line in that currency is given by [Verbeek (2004)]: F(xci b)/sci =exp(xci b)/(1+ exp(xci b))2*b2 (4)
in the probit regression. For the values of the U.S. dollar shortages in our sample of countries, the marginal effect of the estimated U.S. dollar shortage on the probability of receiving a U.S. dollar swap line ranges from close to zero to around 0.01, similar to what we found for the yen above. These results suggest that economies with larger U.S. dollar shortages on our measure, and economies that are large international financial centres, had a statistically significantly higher probability of receiving a U.S. dollar swap line. Next, we repeat this exercise in the probit regression for the probability of the country receiving a U.S. dollar swap line from the Federal Reserve,
This marginal effect again depends on the value of the shortage. For the values of the shortages in our sample of countries, it ranges from close to zero to around 0.06 for the Swiss franc, to around 0.02 for the euro, and to around 0.01 for the yen. We find that the probit and logit models give similar results for the significance of the coefficients and the magnitudes of the marginal effects of the currency shortages on the probability of receiving a swap line in that 38 currency for the Swiss franc, euro, and yen where the effect is significant.

From any country Constant, b1 U.S. dollar shortage, b2 Dummy for large int. financial center, b3 McFadden R2 Number of observations -0.73** (0.27) 0.026* (0.010) 4.94** (1.87) 0.31 39
From the Federal Reserve (Equation 5) -0.96** (0.30) 0.026* (0.011) 5.24** (1.93) 0.35 39
Canada ECB Switzerland Japan United Kingdom Denmark Australia Sweden Norway New Zealand Korea Brazil Mexico
0 291.352 25.175 122.716 33.080 15.000 22.830 25.000 8.225 0 10.350 0 0 0 553.728
Table 4 Probit model for probability of receiving U.S. dollar swap line
From any country Constant, b1 U.S. dollar shortage, b2 Dummy for large int. financial center, b3 Time zone difference to New York, b4 McFadden R2 Number of observations -0.61 (0.50) 0.026* (0.011) 4.99** (1.89) -0.02 (0.07) 0.31 39
From the Federal Reserve
Singapore Total
-0.40 (0.51) 0.027* (0.011) 5.74** (2.09) -0.10 (0.08) 0.39 39
Source: Federal Reserve Bank of New York, Treasury and Federal Reserve foreign exchange operations, various releases.
Table 6 Drawings of U.S. dollars on Fed swap lines (U.S.$ billions, end2008)
(1) is now defined as x$i = (1, s$i, dilfc, tzi), and the vector of coefficients is b = (b1,b2,b3,b4). The results are reported in Table 5 (middle column). We find that the estimated coefficient on the difference in time zones is not statistically significant in the regression. Next, we run the analogous regression for the probability of a country receiving a U.S. dollar swap line from the Fed. The vector of explanatory variables in the probit regression of equation (5) is now defined as x$i =(1, s$i, dilfc, tzi), and the vector of coefficients is b=(b1,b2,b3,b4). A similar result holds in this case (see Table 5, right-hand column). This regression is also consistent with the Federal Reserve having a statistically significantly higher probability of providing swap lines to large international financial centers. Next, we study whether the actual amounts drawn on swap lines pro-
Table 5 Probit model for probability of receiving U.S. dollar swap line
rather than from any country.15 The dependent variable in the probit regression is now the probability that a country i receives a U.S. dollar swap line from the Federal Reserve, yFedi =1, and the explanatory variables are again x$i =(1, s$i, dilfc), with the vector of coefficients being b=(b1,b2,b3), P(yFedi =1| xci )= F(xci b) (5)
vided by the Federal Reserve at end-2008, draw$i (in U.S. dollar billions), depend significantly on our measure of a countrys U.S. dollar shortage, s$i (in U.S. dollar billions), for the sample of countries which received a swap line from the Federal Reserve. Data on drawings on Fed swap lines at end-2008 are shown in Table 6. We estimate the following regression via OLS (with White heteroskedasticity-consistent standard errors), draw$i = b1 + b2*s$i + b3*dilfc+ i (6)
The results are reported in Table 4 (right-hand column). For swap lines provided by the Fed we also find that the coefficients on both the U.S. dollar shortage and the dummy for a country being a large international financial center are statistically significant. These results suggest that countries with larger U.S. dollar shortages on our measure, and countries that are large international financial centers, had a statistically significantly higher probability of receiving a U.S. dollar swap line from the Federal Reserve. Finally, we add a variable for the difference in time zones between each country and New York, tzi (in hours), in the probit regressions for the probability of the country receiving a U.S. dollar swap line from any country. The vector of explanatory variables in the probit regression of equation
as well as the regression also including a variable for the difference in
15 India and Indonesia received a swap line from the Bank of Japan, but not from the Fed. All the other countries receiving US dollar swap lines did so from the Fed.
39
Equation (6) Constant, b1 U.S. dollar shortage, b2 Dummy for large int. financial center, b3 Time zone difference to New York, b4 Dummy for large int. financial center * time zone difference, b5 Adjusted R2 Number of observations -2.53 (6.96) 0.36+ (0.17) 69.6* (31.5) 0.56 14
Equation (7) -16.7 (15.8) 0.39* (0.17) 59.0 (34.9) 2.85 (3.12) 0.53 14
Equation (8) -2.20 (8.63) 0.42* (0.18) 7.73+ (3.55) 0.55 14
a large number of advanced and emerging economies. We discussed the reasons for establishing swap facilities, related our measure of currency-specific liquidity shortages to the probability of a country receiving a swap line in that currency, and found a significant relationship in the case of the U.S. dollar, the euro, the yen, and the Swiss franc. Moreover, we found that economies which are large international financial centers had a statistically significantly higher probability of receiving a U.S. dollar swap line from the Federal Reserve, as well as from any country. We also found that actual U.S. dollar funding obtained by drawing on the Feds swap lines at end-2008 was statistically significantly larger for economies with higher U.S. dollar shortages on our measure, as well as for economies which are large international financial centers.
**, * and + denote significance at the 1%, 5% and 10% levels, respectively; standard errors are given in brackets; White heteroskedasticity-consistent standard errors.
References

Table 7 Drawings on U.S. dollar swap line from the Federal Reserve at end-2008
time zones, tzi , draw$i = b1 + b2*s$i + b3*dilfc + b4*tzi + i (7)
and a specification including the dummy variable for large international financial centers interacted with the difference in time zones, draw$i = b1 + b2*s$i + b5*dilfc*tzi + i (8)
Results of these regressions are shown in Table 7. We can see that the coefficient on the U.S. dollar shortage is significant at the 5 percent level in two of the specifications, and at the 10 percent level in the remaining one. The dummy variable for large international financial centers is significant at the 5 percent level in the first specification, and the dummy variable for large international financial centers interacted with the variable for the difference in time zones is significant at the 10 percent level in the third specification. Goodness of fit of these regressions, as measured by the adjusted R2, is around 0.55. These results suggest that actual U.S. dollar funding obtained by drawing on the Feds swap lines at end-2008 was statistically significantly larger for countries with higher U.S. dollar shortages on our measure, as well as for economies which are large international financial centers.
Conclusion
In this paper we discussed the main innovation in central bank cooperation during the financial crisis of 2008-09, namely the emergency provision of international liquidity through the establishment of bilateral central bank swap facilities, which have evolved to form interconnected swap networks. Based on the BIS international locational banking statistics, we presented a measure of currency-specific liquidity shortages for the 40 U.S. dollar, the euro, the yen, the pound sterling, and the Swiss franc for
Allen, W. and R. Moessner, 2010, Central bank co-operation and international liquidity in the financial crisis of 2008-9, Bank for International Settlements Working Paper no 310. Baba, N. and F. Packer, 2009, From turmoil to crisis: dislocations in the FX swap market before and after the failure of Lehman Brothers, BIS Working Paper no 285 Baba, N., F. Packer, and T. Nagano, 2008, The spillover of market turbulence to FX swap and cross-currency swap markets, BIS Quarterly Review, March Bernanke, B., 2009, Reflections on a year of crisis, speech at the Brookings Institutions conference on A year of turmoil, Washington, D.C, 15 September BIS, 2008, Guidelines to the international locational banking statistics, December BIS, 2009, Statistical annex, BIS Quarterly Review, June Federal Open Market Committee, 2007, Minutes of conference call on December 6, 2007, available at http://www.federalreserve.gov/monetarypolicy/files/fomcminutes20071211.pdf Federal Reserve Bank of New York, 2008, Treasury and Federal Reserve foreign exchange operations: October December 2007, http://www.newyorkfed.org/newsevents/news/ markets/2008/fxq407.pdf . Federal Reserve System, Board of Governors, 2005, The Federal Reserve System: purposes and functions Hartman, D., 1984, The international financial market and U.S. interest rates, Journal of International Money and Finance, 3, 91-103 Kawai, M., 2007, Evolving economic architecture in East Asia, ADB Institute Discussion Paper no. 84, December, http://www.adbi.org/files/dp84.evolving.economic.architecture.east.asia.pdf Kohn, D., 2009, International perspective on the crisis and response, speech at the Federal Reserve Bank of Boston 54th Economic conference, Chatham, Massachusetts, October 23 McGuire, P. and G. von Peter, 2009, The U.S. dollar shortage in global banking, BIS Quarterly Review, March Obstfeld, M., J. C. Shambaugh, and A. M. Taylor, 2009, Financial instability, reserves and central bank swap lines in the Panic of 2008, American Economic Review, Papers and Proceedings, 99:2, 48086 Pomerleano, M., 2009, Ring-fence cross-border financial institutions, FT economists forum, August 10, http://blogs.ft.com/economistsforum/2009/08/ring-fence-cross-border-financialinstitutions/ Roth, J.-P., 2009, Geldpolitik ohne Grenzen. Vom Kampf gegen die Internationalisierung des Frankens zur Internationalisierung der Geldpolitik, speech at Schweizerisches Institut fr Auslandsforschung, Zrich, 6 May Swiss National Bank, 2009, Monetary policy report, Quarterly Bulletin, 3/2009, p. 6-43 Verbeek, M., 2004, A guide to modern econometrics, Wiley Zhou, X., 2009, Reform the international monetary system, speech available at http://www. pbc.gov.cn/english/detail.asp?col=6500&id=178 , 23 March

Appendix List of swap lines extended between December 2007 and 2009
Date Counterparty central bank Amount (bln) Expiry date Date Counterparty central bank Amount (bln) Expiry date
Swap lines extended by the Fed to provide dollars. 12-Dec-07 European Central Bank Swiss National Bank 11-Mar-08 European Central Bank * Swiss National Bank* 02-May-08 European Central Bank* Swiss National Bank* 30-Jul-08 18-Sep-08 European Central Bank* European Central Bank* Swiss National Bank* Bank of Japan Bank of England Bank of Canada 24-Sep-08 Reserve Bank of Australia Sveriges Riksbank Danmarks Nationalbank Norges Bank 26-Sep-08 European Central Bank* Swiss National Bank* 29-Sep-08 Bank of Canada* Bank of England* Bank of Japan* Danmarks Nationalbank European Central Bank* Norges Bank* Reserve Bank of Australia Sveriges Riksbank* Swiss National Bank* 13-Oct-08 Bank of England* European Central Bank* Swiss National Bank* 14-Oct-08 28-Oct-08 29-Oct-08 Bank of Japan* Reserve Bank of New Zealand Banco Central do Brasil Banco de Mexico Bank of Korea Monetary Authority of Singapore 20 4 30 6 50 12 55 110 27 60 40 10 10 10 5 5 120 30 30 80 120 15 240 15 30 30 60 Unlimited Unlimited Unlimited Unlimited 15 30 30 30 30 Jun-08 Jun-08 Sep-08 Sep-08 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Jan-09 Apr 2009+ Apr-09 Apr-09 Apr-09 Apr-09 Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+ Apr 2009+
Swap lines extended by the European Central Bank to provide euros 20-Dec-07 16-Oct-08 27-Oct-08 21-Nov-08 Sveriges Riksbank Magyar Nemzeti Bank Danmarks Nationalbank National Bank of Poland 10 5 12 10 Not specified As long as needed Not specified
Other swap lines in the extended euro network a. Danmarks Nationalbank supplying euros 16-May-08 Central Bank of Iceland 0.5 Extended in Nov 08 until end of 2009
16-Dec-08
Bank of Latvia
0.125
b. Norges Bank supplying euros 16-May-08 Central Bank of Iceland 0.5 Extended in Nov 08 until end of 2009
c. Sveriges Riksbank 16-May-08 Central Bank of Iceland 0.5 (EUR) Extended in Nov 08 until end of 2009
16-Dec-08 27-Feb-09
Bank of Latvia Bank of Estonia
0.375 (EUR) 10 (SEK)
Swap lines extended by the Swiss National Bank to provide Swiss francs 15-Oct-08 07-Nov-08 28-Jan-09 European Central Bank National Bank of Poland Magyar Nemzeti Bank Not specified Not specified Not specified Jan 2009*+ Jan 2009*+ Apr 2009+*
*+ The expiry dates of these swap lines were later extended to April then to October 2009 and again until January 2010. +* The expiry date was later extended to October 2009. Asian swap lines a. Peoples Bank of China 12-Dec-08 20-Jan-09 08-Feb-09 11-Mar-09 23-Mar-09 02-Apr-09 Bank of Korea Hong Kong Monetary Authority Bank Negara Malaysia National Bank of Belarus Bank Indonesia Central Bank of Argentina 180 (CNY) 38,000 (KRW) 200 (CNY) 227 (HKD) 80 (CNY) 40 (MYR) 20 (CNY) 8,000 (BYR) 100 (CNY) 175,000 (IDR) 70 (CNY) 38 (ARS) 3 years 3 years 3 years 3 years 3 years 3 years
* Denotes an extension or enlargement of an existing facility. + The expiry date of these swap lines was extended first to October 2009 and later to February 2010. Swap lines under which the Fed can receive foreign currencies 6-Apr-09 Bank of England European Central Bank Bank of Japan Swiss National Bank GBP 30 EUR 80 JPY 10,000 CHF 40 Oct 2009** Oct 2009** Oct 2009** Oct 2009**
b. Bank of Japan 30-Jun-08 12-Dec-08 6-Apr-09 Reserve Bank of India Bank of Korea Bank Indonesia 3 (U.S.$) 20 (U.S.$) 12 (U.S.$) Jan 09***
*** The expiry date was first extended to October 2009, and later to February 2010. Sources: Central banks.
** The expiry date of these swap lines was later extended to February 2010.
41
42
PART 1
The Failure of Financial Econometrics: Stir-Fry Regressions as a Con Job

Imad Moosa Professor of Finance, School of Economics, Finance and Marketing, RMIT
Abstract
This paper demonstrates the hazard of stir-fry regressions, which are used extensively in financial research to produce desirable results by reporting only one or a small number of regressions out of the tens or hundreds that are typically estimated. It is shown, by using data on the capital structure of some Chinese shareholding companies, that the sign and significance of an estimated coefficient change with the set of explanatory variables and that adding more explanatory
variables to the regression equation changes the sign and significance of a coefficient on a variable that is already included in the model. It is demonstrated that it is possible to change coefficients from significantly positive to significantly negative and vice versa and that obtaining the desirable results can be achieved by introducing various forms of nonlinearities. Finally, it is shown that it is possible to support either the trade-off theory or the pecking order theory by changing model specification. 43
Introduction
In 1983 Edward Leamer published his provocative article, Let us take the con out of econometrics, in which he justifiably criticized a practice in which economists tend to engage that of estimating 1,000 regressions and reporting the one or few that they like [Leamer (1983)]. Almost thirty years later, this practice is still highly popular. In fact it has become more widespread because of the growth in the power of computing. It is particularly widespread in corporate finance where testable models are assembled by combining various hypotheses to come up with a crosssectional regression equation that has no corresponding theoretical model. The regression equation is subsequently twisted and turned until it produces the results that make a dream come true. The problem with cross-sectional regressions is that theory is not adequately explicit about the variables that should appear in the true model as determined by theory. This would be the case if, for example, the final model specification is derived by solving a theoretical optimization problem. In the absence of a theoretical model, the regression equation is constructed haphazardly, by specifying the dependent variable, y, to be a function of several explanatory variables, xj, where j = 1,.,n. The results typically turn out to be difficult to interpret for example, x1 is significant when the regression includes x2 and x3, but not when x4 is included. So, which combination of all available xjs is to be chosen? It is a common practice to report the most appealing or convenient regression or regressions after extensive search and data mining (given that we do not know what the true model is). While scientific research should be based on a quest for the truth, this practice is motivated by the desire to prove a preconceived idea, which is particularly alarming if the idea is driven by ideology. Gilbert (1986) casts significant doubt on the validity of the practice of assigning 999 regressions to the waste bin, because they do not produce the anticipated results. Because of this problem, Leamer (1983) suggested that econometricians confine themselves to publishing mappings from prior to posterior distributions rather than actually making statements about the economy. Leamer and Leonard (1983) argued strongly against the conventional reporting of empirical results, stating that the reported results are widely regarded to overstate the precision of the estimates, and probably to distort them as well. As a consequence, they pointed out, statistical analyses are either greatly discounted or completely ignored. They further argued that the conventional econometric methodology (or technology as they called it) generates inference only if a precisely defined model were available, and which can be used to explore the sensitivity of inferences only to discrete changes in assumptions. Hussain and Brookins (2001) point out that the usual practice of reporting a preferred model with its diagnostic tests need not be sufficient to convey the degree of reliability of the determinants (explanatory variables). 44 A regression equation that is constructed in the way described above
is what I cynically refer to as a stir-fry regression, because there is no unique recipe for a stir fry. It may contain beef, chicken, various kinds of vegetables, and perhaps three or four sauces. Experimentation with stir fry starts by throwing everything in a pot or a wok, then by a process of elimination and the use of various combinations of the ingredients, the stir fry is perfected. The analogous practice in financial econometrics is to throw numbers representing 15 or so variables and arrive at the best model through a process of experimentation with various specifications and variable definitions. There is, however, a big difference between preparing a stir fry and the estimation of a model by using stir-fry regressions. The stir fry will likely turn out to be a delicious meal, but the model will tell us nothing apart from what the researcher would like us to believe. Hence, the model will be useless at best and hazardous at worst. The estimation of stir-fry regressions means that many conflicting inferences can be drawn from a given data set. According to Leamer and Leonard (1983), this deflects econometric theory from the traditional task of identifying the unique inferences implied by a specific model to the task of determining the range of inferences generated by a range of models. The objective of this paper is to demonstrate the hazard of stir-fry regressions, using a dataset (obtained from the OSIRIS database) on the capital structure and its determinants of 343 Chinese shareholding companies. Specifically, five propositions are presented to demonstrate that (i) the sign and significance of an estimated coefficient change with the set of explanatory variables, (ii) adding more explanatory variables to the regression equation changes the sign and significance of a coefficient on a variable that is already included in the equation, (iii) it is possible to change coefficients from significantly positive to significantly negative and vice versa, (iv) obtaining the desirable results can be achieved by introducing various forms of nonlinearities, and (v) it is possible by changing model specification to support either of two competing theories.
Stir-Fry Regressions in Studies of the Capital Structure

Studies of the capital structure are typically based on a cross sectional regression of the form LEVi = 0 + n j xji + i j=1 (1)
where LEV is the leverage ratio and xji is explanatory variable j for company i. The most commonly used explanatory variables, which are also used in this study, are size (SIZ), liquidity (LIQ), profitability (PRF), tangibility (TAN), growth opportunities (GOP), the payout ratio (POR), stock price performance (SPP), the age of the firm (AGE), and income variability (VAR). The problem of the sensitivity of the estimated coefficients with respect to model specification arises in studies of capital structure because no single theoretical model defines an explicit set of explanatory variables

to include in any empirical model. Huang and Ritter (2005) correctly argue that no single theory of capital structure is capable of explaining the time-series and cross-sectional patterns that have been documented. Likewise, Frydenberg (2008) points out that neither the pecking order theory nor the trade-off theory provides a complete description of empirical observations or explains why some firms prefer equity while others prefer debt. Titman and Wessels (1988) and Harris and Raviv (1991) argue along similar lines, pointing out that the choice of explanatory variables in the analysis of cross-sectional variation in capital structure is fraught with difficulty. The consequence is that researchers are tempted to try various combinations of the explanatory variables and report the ones they like, typically the ones that produce good results and/or confirm preconceived beliefs. This is why some firm-specific factors that are reported as important for capital structure may not be so they only appear to be so because of a model specification that contains a particular combination of explanatory variables. Li et al. (2009) use a sample of 417,068 firm-year observations over the period 2000-2004. They test the importance of nine explanatory variables: size, profitability, tangibility, asset maturity, industry concentration, industry leverage, state ownership, foreign ownership and marketization (classified into firm characteristics, ownership variables, and institutional variables). The results show that state ownership is positively associated with leverage (long-term debt), while foreign ownership is negatively associated with all measures of leverage. Other conclusions are derived on the effect of firm-specific factors. The Li et al. (2009) study provides the best illustration of the problem addressed in this paper. Five different model specifications are estimated: a full specification (all variables), firm characteristics only, ownership variables only, institutional variables only, and ownership/institutional variables. The diversity and inconsistency of the results can be demonstrated with just a few examples out of many: 1. For total leverage, industry concentration is highly significant in the full specification but not so in the specification that has firm characteristics only. 2. For short-term debt, size is highly significant in the full specification but not so in the specification that has firm characteristics only. The same goes for industry concentration. Marketization is insignificant when only institutional variables are included in the model but becomes significant when ownership variables are added. 3. In a model designed to determine the probability of having long-term debt, asset maturity is significant when firm characteristics only are used but not so in the full specification. It is exactly the other way round for industry concentration. 4. When the equations are estimated with fixed effects, industry concentration, state ownership, and marketization are highly significant when the model includes ownership and institutional variables only, but not so in any of the other model specifications.
So, what are we supposed to believe and how can we derive robust inference from these results? The answer is that we do not know what to believe and we cannot derive robust inference. The same problem can be observed in Li et al. (2009) who present three different models (i) with a constant and six explanatory variables, (ii) without a constant and six explanatory variables, and (iii) without a constant and four explanatory variables. The results turn out to be a mixed bag: size is significant only if the constant term is taken out, while profitability is significant only in the second regression. With this kind of results we cannot identify the factors determining capital structure with any degree of robustness. The best answer to the question, What determines capital structure? should be, Anything and nothing. This is what we will illustrate by presenting five propositions.
Proposition 1
By using stir-fry regressions we can change the sign and significance of the coefficients on the explanatory variables by estimating models with various combinations of these variables. To demonstrate this proposition, consider the general model, which will be our starting point: LEV = 0 + 1SIZ + 2LIQ + 3PRF + 4TAN + 5GOP + 6POR + 7SPP + 8AGE + 9VAR + (2)
We start by estimating this model and then deleting some variables and adding others to arrive at various results these results are reported in Table 1. Model 1 includes all of the nine explanatory variables as well as a constant term. Out of the nine explanatory variables, five of them turn out to have significant coefficients: liquidity, profitability, tangibility, growth opportunities, and stock price performance. For some, this is no good because size, presumably the most important explanatory variable, has an insignificant coefficient when this coefficient is supposed to be significantly positive. So, let us see what happens when we delete the insignificant explanatory variables to obtain Model 2. We end up with the same sign and significance of the same variables as in Model 1: no change and hardly any support for either the trade-off theory or the pecking order theory. But this is not the end of the matter. The finding that size is unimportant can be reversed by adding PRF as an explanatory variable, which gives Model 3. Hence size is unimportant in Model 1 but important in Model 3. Again, what are we supposed to believe? Consider now Model 4, which is obtained by deleting the insignificant variables in Model 1 with the exception of size. We get a great result: size is still important and all other coefficients are significant. It is tantalizing to report this regression as representing the truth, the whole truth, and nothing but the truth. Let us now move to Model 5, in which VAR (but none of the other variables) has a significant coefficient. All of a sudden, the only variable that determines the capital structure is 45
Model
1 76.829 (19.02) 0.149 (1.58) -0.566 (-9.35) -0.891 (-6.15) -0.113 (-3.21) -2.889 (-4.19) -0.002 (-0.99) 0.027 (3.11) -0.120 (-0.52) 0.00002 (1.08)
2 80.578 (38.19)
3 76.782 (35.43) 0.230 (5.07)
4 76.777 (35.43) 0.239 (5.11) -0.585 (-10.12) -0.862 (-6.04) -0.125 (-3.70) -2.802 (-4.09)
5 46.656 (10.01)
7 49.72 (11.19)
Model
8 50.588 (48.07) 0.366 (6.09)
9 67.138 (47.18) 0.272 (5.66) -0.522 (-9.58) -1.088 (-8.13)
10 75.230 (35.17) 0.233 (4.91) -0.053 (-9.49) -0.884 (-6.12) -0.114 (-3.35) -2.407 (-3.53)
11 76.782 (35.43) 0.238 (5.07) -0.057 (-9.43) -0.882 (-6.11) -0.121 (-3.56) -2.830 (-4.13) -0.002 (-0.97) 0.027 (3.09)
12 76.865 (19.02) 0.238 (5.06) -0.057 (-9.40) -0.881 (-6.10) 0.121 (-3.53) -2.831 (-4.12) -0.002 (-0.86) 0.026 (3.09) -0.006 (-0.02)
13 76.829 (19.01) 0.149 (1.58) -0.56 (-9.35) -0.891 (-6.15) -0.112 (-3.21) -2.889 (-4.19) -0.002 (-0.98) 0.027 (3.12) -0.012 (-0.05) 0.00002 (1.09)
0
SIZ LIQ PRF TAN GOP POR SPP AGE VAR
0
SIZ LIQ PRF TAN GOP
0.262 (1.93) -0.27 (-3.18) -0.771 (-3.70) 0.056 (1.29) 0.165 (3.57) 0.964 (1.01) -0.005 (-1.94) -0.002 (-0.86) 0.001 (0.09) 3.689 (20.60) 0.0002 (0.86) -0.005 (-2.03) -0.242 (-2.23) 0.34 (1.11)
-0.059 (-9.99) -0.840 (-5.68) -0.127 (-3.63) -3.472 (-4.99)
-0.568 (-9.43) -882 (-6.12) -0.121 (-3.57) -2.830 (-4.13) -0.002 (-0.87)
POR SPP AGE VAR
0.025 (2.84)
0.027 (3.09)
0.027 (3.16)
-0.019 (-1.75) 0.272 (0.89) 0.00006 (4.84)
Table 1 Models 1-7 (t-statistics in parentheses)
income variability, and it has a positive influence on the leverage ratio, which does not make any sense. By moving from Model 5 to Model 6, we have age with a significantly positive coefficient, but not so for size, while tangibility changes to be significantly positive. How about finding a result that shows the pay-out ratio to be important? That is easy, as it appears significantly negative in Model 7. By estimating just seven model specifications, we have managed to obtain all sorts of results. With nine explanatory variables it is possible to estimate 508 models containing from nine to three explanatory variables. Imagine what flexibility we have to produce the results we like to report, particularly if we play around with the definitions of the variables and (as we are going to see later) experiment with nonlinear specifications.
report any of the models 8-12, but we keep quiet about Model 13. This is the magic of stir-fry regressions.
Proposition 3
It is possible, with the help of stir-fry regressions, to change the estimates of some coefficients from significantly positive to significantly negative. This proposition is illustrated by reporting the results of estimating Models 14 and 15, which are reported in Table 3. Model 14 is the same as Model 1 with AGE deleted. Notice that in Model 14, TAN and GOP exert significantly negative effect on the leverage ratio. Notice also that this model shows that size has an insignificant coefficient. For some, these results are bad because the trade-off theory says that the effect of TAN is positive while the pecking order theory says that the effect of GOP is positive. This is indeed disturbing, but it is not an insurmountable problem. By taking out the constant term, both coefficients become significantly positive, as in Model 15. But that is not all: the coefficient on LIQ turns from significantly negative to significantly positive while the coefficient on SPP turns from significantly positive to significantly negative. And despite the presence of VAR in Model 15, the coefficient on size is significantly positive. All of this magic happened because we took out the constant term from Model 14.
Proposition 2
Adding more explanatory variables to the model changes the significance of the variables already in the model. This proposition can be illustrated by starting with a model in which the only explanatory variable is size, then we add variables as we go on. For this purpose we estimate Models 8-13, which are reported in Table 2. In Model 8, where the only explanatory variable is size, it turns out to be significantly related to the leverage ratio. Because it is insignificant in Model 1, it may be interesting to find out what causes the shift from significance to insignificance by adding variables as we move towards Model 13. Size remains a significant variable until income variability is added as in Model 13. It is not clear why this happens, given that VAR is itself insignificant, but who cares if that is the result that we want to report? What this tells us is that if we wish to 46 show size as an important determinant of the capital structure, we must
Proposition 4
It is possible, with the help of stir-fry regressions, to obtain the desirable results by the selective and haphazard introduction of nonlinearities without any theoretical or empirical justification. It is invariably the case that in those papers that report regressions in which size is an explanatory

Model
14 76.651 (35.32) 0.149 (1.58) -0.056 (-9.38) -0.891 (-6.16) -0.112 (-3.23) -2.888 (-4.21) -0.002 (-0.99) 0.027 (3.12)
15
selectively is a biased introduction of nonlinearities. As far as I can see, this selectivity is an integral part of the data mining process motivated by the desire to produce the optimal results. It does not stop at us-
0
0.557 (2.74) 0.027 (2.22) -0.381 (-1.22) 0.609 (9.98) 8.539 (6.49) -0.002 (-0.43) -0.041 (-2.24)
ing logs, as I have seen transformations by the selective use of squares and square roots, typically without justification. This malpractice serves a very important purpose of data miners: increasing the number of combinations, hence the probability of converging on what they want to see and report. Consider Table 4, which reports the results of estimating Models 16-20. Model 16 is exactly like Model 1, except that we now have log(SIZ) as opposed to SIZ. The coefficient is significantly positive. For someone wanting to prove that size is an important determinant of capital structure, log(SIZ) should be used as an explanatory variable because it gives that result. But now GOP is insignificant. Next try Model 17, which has log(PRF) as an explanatory variable. Size is no longer important, neither are TAN and GOP, but VAR is now important (it has a significantly positive coefficient). In Model 18, we have log(VAR), which makes TAN and GOP important determinants. In Model 19, the square root of size is an explanatory variable, which gives a brand new result: size has a significantly negative effect on the leverage ratio. This is a big change from any of the previous results where the coefficient on SIZ was either significantly positive or insignificant. In Model 20 there is an exotic combination of
0.433 (7.81) -0.000006 (-2.42) -0.0003 (-0.94) 0.00008 (3.22) 0.014 (2.27) 0.00004 (0.04) 0.0006 (1.05) 0.219 (31.87) -0.000002 (-2.64)
e
0.00002 (1.09)
0.00006 (1.41)
Model
16a 67.377 (16.74) 6.15 (7.06) -0.055 (-9.70) -0.993 (-7.31) -0.132 (-4.06) -1.234 (-1.79) -0.001 (-0.84) 0.028 (3.47) -0.043 (-1.98) -0.00006 (-0.40)
17b 95.971 (5.86) -0.276 (-1.12) -0.107 (-6.59) -6.978 (-3.96) -0.128 (-1.98) -1.700 (-1.60) 0.0009 (0.032) 0.060 (4.64) -0.629 (-0.69) 0.0001 (2.48)
b c
18c 25.026 (2.64) 0.034 (0.61) -0.054 (-9.40) -1.010 (-7.26) -0.093 (-2.83) -2.300 (-3.49) -0.002 (-1.17) 0.028 (3.43) -0.079 (-0.36) 4.38 (5.99)
d
19d 75.972 (18.81) -0.002 (-2.56) -0.055 (-9.28) -0.921 (-6.40) -0.092 (-2.66) -3.031 (-4.48) -0.002 (-1.05) 0.028 (3.22) -0.002 (-0.008) 0.00008 (4.90)
20e
0
a
nonlinearities as the explanatory variables include the log of size and the square roots of LIQ, PRF, and TAN. The dependent variable is log(LEV) and there is no constant term. In this case tangibility has a significantly positive effect, variability has a significantly negative effect, and age has a significantly positive effect. The limit, it seems, is the sky. Perhaps a comment is warranted on the finding that the coefficient on size turned out to be significantly negative. Size is typically taken to have a positive effect on the leverage ratio because big firms have easier access to debt markets and lower cost of debt financing than small firms. However, a finding of a significantly negative coefficient on size can be justified. For example, Chen (2004) suggests that the negative relation between size and leverage can be explained in terms of the better access of large firms to equity finance because of their reputation in the markets and the attraction of capital gains in the secondary market. For the special case of China, it is also suggested that the negative relation is due to the fact that bankruptcy costs are low since the legal system is incomplete. It seems, therefore, that some data miners may give up too soon and choose instead to find a plausible explanation for a finding that could, with some patience, be reversed.
log(SIZ) is an explanatory variable; log(PRF); log(VAR); log(SIZ), and square roots of LIQ, PRF, TAN, and GOP.
square root of SIZ; log (LEV),
Table 4 Models 16-20 (t-Statistics in parentheses)
variable that this variable is measured as the logarithm of total assets, sales, or something like that. The question is why the logarithm? Is it because size is a big number, therefore the log is used as a scaling factor? What is overlooked in this case is that using logs (for whatever reason) changes the underlying model from linear to nonlinear and that using logs
Proposition 5
It is possible, with the help of stir-fry regressions, to support either of the two competing theories. This proposition is illustrated by running a match between two competing theories of the capital structure: the 47
SIZ 21 22 23 24 25
a b c d
LIQ + + + -
PRF 0 0 0 0 -
TAN + + + -
GOP + 0 + 0 0
POR 0 * * * * * * * * *
SPP + + + 0 0 + + + +
AGE 0 * * * * * * * * +
VAR 0 0 0 + 0 + 0 0 + 0
correctly signed the theory gets 2.5 points, otherwise it gets one point for an insignificant coefficient and zero for a coefficient that is significant but incorrectly signed. For the purpose of running the match, we estimate ten models (21-30) where 21 is effectively the basic Model 1. Under Model 21, the trade-off theory gets 3.5 points while the pecking order theory scores 6 points, as shown in Figure 1. When we reach Model 25, the trade-off theory scores 6 points while the pecking order theory scores 3 points. In models 29 and 30, the trade-off theory scores 1 point while the pecking order theory scores 6 and 5 points, respectively. Table 5 shows how the coefficients change in terms of sign and significance as we move from Model 21 to Model 30. Consequently, if I have a bias towards the trade-off theory, I will report Model 26, but if I want to show that the pecking order theory is better, I will report Model 29. This is indeed a con job.
0 0 + * + + + + 0 0
26 27
28
29e 30
f
0: insignificant, significantly negative, + significantly positive, * not included in the regression. a log(LEV), log(SIZ), log(LIQ), log(PRF); b log(LEV), log(SIZ); c log(LEV), log(SIZ), log(LIQ), d e f log(PRF); log(SIZ), log(LIQ), log(PRF); log(LIQ), log(PRF), log(TAN); log(LIQ), log(PRF), log(TAN).
Table 5 Changes in sign and significance (Models 21-30)
Concluding Remarks
Although Edward Leamer made a strong case for taking the con out of econometrics back in 1983, the con job of stir-fry regressions is quite
trade-off theory and the pecking order theory. According to the tradeoff theory, a firm sets a target debt level and moves towards it gradually. As its name implies, the trade-off theory explains observed capital structures in terms of a trade-off between the costs and benefits of debt. The theory postulates that firms raise their debt level to the extent that the marginal tax advantages of additional borrowing are offset by the increase in the cost of financial bankruptcy. The pecking order theory, which was pioneered by Myers and Majluf (1984), suggests that firms do not have a leverage target and that they focus on information costs and signaling effects. They demonstrate that firms prefer to finance projects from internally generated cash flows namely, retained earnings and depreciation expenses. When this source of funds is exhausted, they move on to debt. Additional equity is issued only when the latter is not sufficient to meet financing needs. This hierarchy is justified by differences in financing costs: issuing additional equity is the most expensive source of financing as it involves information asymmetries between managers, existing shareholders, and potentially new shareholders. In view of the fixed payments associated with debt financing, it is less sensitive to information asymmetries, while internally generated resources do not produce issuing costs. In designing the match between the two theories, the winner is the theory that scores more in terms of the sign and significance of the estimated coefficients in the underlying regression. The trade-off theory predicts that PRF and TAN affect the leverage ratio positively whereas GOP and VAR affect it negatively. On the other hand, the pecking order theory predicts a negative influence for PRF, LIQ, and SPP and a 48 positive influence for GOP. If the underlying coefficient is significant and
conspicuous in financial research, particularly in the field of corporate finance. As a matter of fact the use of stir-fry regressions has intensified because of the exponential growth of computer power since the early 1980s. This paper has shown that it is possible, by estimating a large number of regressions and reporting a few of them, to come up with results that support the researchers preconceived beliefs or results that look nice to boost the probability of getting the paper accepted for publication. Furthermore, the results presented in this paper highlight the delusion of economists believing that economics (and finance) is as rigorous as physics. The phenomenon of stir-fry regressions does not arise in physics research. Take, for example, an experiment conducted to test Boyles law on the relationship between the volume of gas and the pressure exerted
Score
7
Trade-off Pecking order
0 21 22 23 24 25 26 27 28 29 30
Model
Figure 1 Trade-off versus pecking order

on it. In this experiment, the pressure exerted on a gas is raised gradually while the temperature is held constant. Observations are recorded on the volume of gas at each level of pressure. These observations are then used to run a regression of volume on pressure and the result will be unambiguous as the regression equation produces a perfect fit in the form of a rectangular hyperbola. The same can be said of the relation between the boiling point of water and pressure. Physical laws are universal they hold everywhere. We cannot say the same thing about the relation between the leverage ratio and size, as it can be anything. This is why the phenomenon of stir-fry regressions appears in economics and finance but not in physics, where one regression equation is estimated from experimental data. Sadly for our discipline, stir-fry regressions constitute a con job that is inconsistent with the spirit of scientific research of going on a quest for the truth. This is yet another manifestation of the failure of financial econometrics.
References

Chen, J. J., 2004, Determinants of capital structure of Chinese-listed companies, Journal of Business Research, 57, 1341-1351 Frydenberg, S., 2008, Theory of capital structure: a review, Available at http://papers.ssrn. com/sol3/papers.cfm?abstract_id=556631 Harris, M. and A. Raviv, 1991, The theory of capital structure, Journal of Finance, 46, 297-355 Gilbert, C. L., 1986, Professor Hendrys econometric methodology, Oxford Bulletin of Economics and Statistics, 48, 283-307 Huang, R. and J. R. Ritter, 2005, Testing the market timing theory of capital structure, working paper, Available at http://bear.warrington.ufl.edu/ritter/ TestingOct2805(1).pdf. Hussain, M. and O. S. Brookins, 2001, On the determinants of national saving: an extreme bounds analysis, Weltwirtschaftliches Archiv, 137, 151-174 Leamer, E., 1983, Lets take the con out of econometrics, American Economic Review, 73, 31-43 Leamer, E. and H. Leonard, 1983, Reporting the fragility of regression estimates, Review of Economics and Statistics, 65, 307-317 Li, K., H. Yue, and L. Zhao, 2009, Ownership, institutions, and capital structure: evidence from China, Journal of Comparative Economics, 37, 471-490 Myers, S. and N. S. Majluf, 1984, Corporate financing and investment decisions when firms have information that investors do not have, Journal of Financial Economics, 13, 187-221 Titman, S. and R. Wessels, 1988, The determinants of capital structure choice, Journal of Finance, 43, 1-19
49
PART 1
Lehman A Case of Strategic Risk

Patrick McConnell Visiting Fellow, Trinity College Dublin, and Honorary Fellow, Macquarie
University Applied Finance Centre (MAFC)
Abstract
The bankruptcy of Lehman Brothers in 2008 was a critical event in the Global Financial Crisis, exposing serious faultlines in the structure of global financial markets and leading to widespread economic disruption. But how did Lehman, a company of over 150 years experience in commodities markets, reach such a precarious position? The underlying reason for the failure goes back to a change in corporate strategy in 2006 in which Lehman decided to shift from a moving or securitization business to a storage business, with the firm making and holding longer-term, riskier investments. This strategic positioning was fully endorsed by the board, although certain senior risk management executives had expressed concerns about the extent of the change to the firms risk appetite. Using the official report into the firms bankruptcy, this paper describes how Lehman overextended itself in its strategic execution, failing to put in the place the proper governance structures needed to manage, arguably, the greatest risk to any firm, strategic risk. To prevent similar failures in future, there is a demonstrable need for banking regulators, especially of significantly important banks, to address such critical deficiencies in corporate governance and strategic risk management. 51
Like the assassination of Archduke Franz Ferdinand in 1914 (which led to the outbreak of the First World War), the bankruptcy of Lehman Brothers in September 2008 had ramifications far beyond the narrow confines of the event itself. The liquidation of Lehman did not cause the global financial crisis (GFC) but the firms failure exposed serious faultlines in the structure of the global financial markets where a cats cradle of opaque and complex interconnections collapsed like a house of cards when Lehman was removed.
conflict of interest in that the board and management own a firms strategy but that they are at the same time also responsible for implementing the strategy and managing the strategic risks. There is no independent review of the strategic risks taken by many firms, which constitutes a serious deficiency in corporate governance, particularly, as in the case of Lehman, where the CEO (who proposed the strategy) was also Chairman (who approved it). But Lehman was not unique in this respect. The Financial Crisis Inquiry
Lehmans bankruptcy was large but its cause was not unusual; in essence, the firm just could not pay its debts to its creditors. The proximate cause of the insolvency was that the firm was overly exposed to the commercial property market and was sitting on a large warehouse of securities, so-called Collateralised Debt Obligations (CDOs), the value of which were falling rapidly as a result of agency downgrades [Valukas (2010)]. Much of the commentary on Lehmans bankruptcy concentrates on hypotheticals, such as what would have happened if the U.S. Federal Reserve Bank had not allowed Lehman to fail and instead arranged a shotgun marriage with another bank, as happened earlier in the year with Bear Stearns [FCIC (2011)]. But at the time, the Fed was not able/willing to save Lehman and the rest is history [Tibman (2009); Valukas (2010)]. This paper does not look at the frantic activities in months leading to Lehmans bankruptcy, which are well covered elsewhere [Tibman (2009); Valukas (2010)], but instead asks the question: how did a firm that had been trading successfully for over 150 years come to a point where it was so catastrophically insolvent? The board and management were experienced in the businesses in which Lehman operated and therefore should not have been caught out so completely by the events in the credit markets in 2007 and 2008. The paper argues that the events can be traced back to a decision in 2006 to radically change the strategic direction of the firm and that there was a failure to manage the risks in that change of strategy. In short, the firm failed to properly manage its strategic risks. The Examiners report into Lehmans bankruptcy runs to some nine volumes and 2,100 pages, explaining in excruciating detail the events that led up to the demise of the firm. For the purposes of this paper, however, it is sufficient to note that in 2006, Lehmans management developed a new strategy, which was fully endorsed by the board, although the risks inherent in the strategy were not fully disclosed to the board nor properly mitigated by management, with the result that the flawed strategy [Valukas (2010)] resulted in the bankruptcy of Lehman in 2008.
Commission (FCIC) was set up by President Obama and the U.S. Congress with the aim of examining the causes, domestic and global, of the current financial and economic crisis in the United States [FCIC (2011)]. One of its key conclusions was that [d]ramatic failures of corporate governance and risk management at many systemically important financial institutions were a key cause of this crisis [emphasis added]. Illustrating with examples from some of the largest financial institutions in the world, the commission reported our examination revealed stunning instances of governance breakdowns and irresponsibility [emphasis added] [FCIC (2011)]. Failure to manage strategic risk was not the cause of the GFC but, as this paper argues, it was a major cause of Lehmans failure and thus contributed to the failures of other casualties of the crisis. Nor was the U.S. alone in deficiencies in corporate governance; it appeared to be a worldwide problem. For example, in one of several reports into the banking crisis in Ireland, it was concluded that the increasing wholesale funding and property lending risks were often accommodated by growing levels of governance failings [emphasis added] [Nyberg (2011)]. The reason for addressing the critical issue of corporate governance, in general, and strategic risk, in particular, is best put by the FCIC commission: These are serious matters that must be addressed and resolved to restore faith in our financial markets, to avoid the next crisis, and to rebuild a system of capital that provides the foundation for a new era of broadly shared prosperity [emphasis added] [FCIC (2011)]. This paper first provides a short history of Lehman Brothers and describes some of the key players and the events that precipitated the collapse of the firm. Following a consideration of the concepts of strategy and strategic risk, the paper then considers the change in corporate strategy that eventually led to the firms collapse and the firms catastrophic failure to manage its strategic risks arising from the change of strategy.
The rise and spectacular fall of Lehman Brothers

Strategic risk is arguably, because of the immense uncertainty in the global economy, the greatest risk facing all firms [McConnell (2012)]. However, strategic risk management, or the management of the risks to a firms long-term corporate strategy, is not a well-developed discipline. 52 The lack of maturity in the discipline stems, in part, from a fundamental
Before the crash

The history of Lehman Brothers is well documented by, among others, Tibman (2009). The firm was founded in the 1840s in Alabama by three German immigrant brothers, called Lehman, first as a general store and then expanding into cotton broking, which involved buying cotton from

local planters and selling to merchants. Rebuilding its operations after the devastation of the Civil War, the firm opened an office in New York in the 1870s, helping to form the New York Cotton Exchange, and then expanding beyond cotton into other commodities, such as coffee and petroleum. On the back of helping to sell bonds for the State of Alabama after the Civil War, Lehman Brothers expanded into the financial community, underwriting securities to fund the rapidly expanding U.S. railroad network in the 1880s. By 1900, the firm, still run by the Lehman family, was a stalwart of Wall Street and the New York Stock Exchange, and over the next 30 years (often in conjunction with Goldman Sachs) helped to underwrite the expansion of growth industries, such as retailing, communications, oil, airlines, and entertainment. Keeping with this strategy, Lehman survived the Great Depression and, from the 1930s onward, began to underwrite securities for the emerging electronic industries such as television and computers. From the 1960s Lehman expanded overseas, first to Paris, then London, and eventually through the 1970s and 1980s, into all major financial centers. In 1969, the link with the founding family was ended with the death of the benevolent dictator Robert (Bobby) Lehman, after which the firm went through a period of hard times. Following the death of Bobby, the firm increasingly became an active trader in the financial markets, in addition to its traditional business of investment banking. This was a change of culture, which created internal tensions that caused several management changes and contributed to the firms relative decline [Tibman (2009)]. In 1984, Lehman Brothers was acquired by American Express, creating an investment bank, Shearson Lehman Brothers, which in turn acquired EF Hutton and Kuhn, Loeb & Co. The firm returned to profitability over the next 10 years. In 1993, realizing that their acquisition strategy was not entirely successful, Amex changed direction and spun off the investment bank as Lehman Brothers. For the next decade, the firm prospered despite significant setbacks such as the Asian financial crisis and the destruction of the firms main offices in the 9/11 terrorist attacks [Tibman (2009)]. At first, Lehman was considered to have weathered the so-called subprime crisis well [Tibman (2009)], but as the crisis grew the firm was eventually overwhelmed and on 15th September 2008, Lehman Brothers Holdings filed for Chapter 11 bankruptcy, citing some U.S.$770 billion of debt with assets of only U.S.$640 billion. After 158 years of generally successful operations, weathering economic catastrophes such as civil wars, depressions, world wars, banking failures, and terrorist attacks, Lehman blew up within a few years. The reasons why Lehman failed so disastrously is important not only for the firms shareholders but for regulators of similar strategically important financial institutions (SIFIs).
A disastrous change in strategy

In March 2006, at a Global Strategy Offsite, Lehman CEO Richard S. (Dick) Fuld explained a new move toward an aggressive growth strategy, including greater risk and more leverage, describing the change, as a shift from a moving or securitization business to a storage business, in which Lehman would make and hold longer-term investments, [FCIC (2011)]. Lehmans board fully endorsed this growth strategy that was presented in the form of a 50 percent increase in risk appetite limit, which Lehman needed to compete [Valukas (2010)]. Not that the strategy was without controversy, with the Chief Risk Officer (CRO) at the time, Madelyn Antoncic, arguing for a much lower limit but being overridden by senior management, eventually leading to her replacement as CRO [Valukas (2010)]. As a result of this change in strategy, Lehmans mortgage related-assets almost doubled to some U.S.$111 billion over the next year, with the firm regularly increasing and then exceeding1 its own internal risk limits [FCIC (2011). Lehman was still buying mortgage assets into 2008, long after the U.S. housing market had peaked and property prices were falling [Shiller (2008)]. Lehmans bankruptcy examiner noted that instead of pulling back from what had become a losing strategy, Lehman made the conscious decision to double down hoping to profit from a countercyclical strategy [Valukas (2010)]. Lehmans management were well aware of the risks that they were running: Financial institutions generally engage in transactions designed to hedge their risks. But illiquid investments [such as mortgage-backed CDOs] are typically more difficult to hedge. In fact, Lehman decided not to try to hedge its principal investment risks to the same extent as its other exposures for precisely this reason its senior officers believed that hedges on these investments would not work and could even backfire, aggravating instead of mitigating Lehmans losses in a downturn. As a result, Lehman acquired a large volume of unhedged assets that ultimately caused Lehman significant losses [emphasis added] [Valukas (2010)]. It was the inability of Lehman to fund itself after the massive losses incurred in their storage strategy that eventually led to its bankruptcy and precipitated the wider economic chaos that subsequently ensued.
The key players in the Lehman collapse

Valukas (2010), Tibman (2009) and others have identified some of the key players in the dramatic events that led to the demise of Lehman, in particular Dick Fuld (Chairman and CEO), Joe Gregory (President and COO), Erin Callan (CFO), Chris OMeara (CRO, and ex-CFO), Madelyn Antoncic (ex-CRO), and various other business executives, such as Skip McGee
Valukas (2010) notes that limit excesses were approved by the executive Risk Committee, which consisted only of the CEO, CFO, and CRO.
53
(head of Investment banking), and Michael Gelband (head of the Fixed Income Division). While the Lehman Board consisted of experienced business executives and directors, such as John Akers (ex-CEO of IBM), their backgrounds were not in areas directly related to financial services. Although criticized [Valukas (2010)], this lack of specific banking expertise was not necessarily a significant disadvantage, given Lehmans traditional business in servicing large corporations, provided that appropriate oversight was exercised. Blame for Lehmans failure has been spread liberally around [FCIC (2011)], most notably singling out Dick Fuld. After a short period in the military, and leaving in somewhat colorful circumstances [Tibman (2009)], Dick Fuld joined Lehman Brothers in 1969 and worked there until the company collapsed, eventually rising to be Chairman and CEO after the company was floated off by American Express. Dick Fuld seems to engender extreme reactions among colleagues and competitors with some regarding him as the main culprit in the demise of Lehman [McDonald and Robinson (2009)]. Fulds nickname in Lehman was Gorilla, which appears to have arisen not merely from his stature and his unashamedly aggressive and combative nature [Tibman (2009)] but, in contrast to the traditional demeanor of a polished investment banker, he was, for such a senior executive, surprisingly inarticulate.2 However, Fuld was not the one-dimensional villain often portrayed by politicians and the media [FCIC (2011)]. Fuld was, at the time of Lehm3
arising from natural forces, are not created overnight. It is rare that an individual, by virtue of a single error, can create a disastrous outcome in an area believed to be relatively secure. To achieve such a transformation he or she needs the unwitting assistance offered by access to the resources ... of large organizations, and time [emphasis added]. If Turners (1976) insights are correct, the failure of a firms strategy is unlikely to be the fault of just one person, but more likely the collective responsibility of the board and management of the firm, its senior staff, and external bodies tasked with oversight, such as regulators, rating agencies, auditors and, institutional investors. Before considering why Lehmans strategy failed so spectacularly, the next section considers the questions: what exactly is strategy and who should be responsible for managing the risks in that strategy?
Strategy and strategic risk

What is strategy?
Johnson and Scholes (2002) define strategy as the direction and scope of an organization over the long-term. Alfred Chandler, one of the founders of modern management theory, defined strategy as the setting of long-term goals and objectives, the determination of course of action, and the allocation of resources to achieve the objectives [Koch (2006)]. The father of strategic planning, George Steiner (1979) notes that strategy answers the questions: (a) what should the organization be doing, and (b) what are the ends we seek, and (c) how should we achieve them? And in discussing strategy in terms of competition, Michael Porter (1980) defines competitive strategy as a combination of the ends (goals) for which the firm is striving and the means (policies) by which it is seeking to get there. Strategy, then, encompasses both the objectives a firm is striving to achieve over the long term and the means by which the firm plans to achieve those objectives. Within a firm, strategies may be developed at different levels, most importantly (Grant 2008):
ans collapse, one of the longest serving CEOs on Wall Street and in 2006 had been named by the influential magazine, Institutional Investor, as the Top CEO in the private sector [Institutional Investor (2007)]. Fuld was the consummate Wall Street insider and, until close to the end of the crisis, he had served on the board of the Federal Reserve Bank of New York. Inside the firm, he was respected as the architect of the One Lehman policy, which brought together the traditionally antagonistic investment banking and trading businesses into an integrated operating unit. Tibman (2009) describes the normally taciturn Fuld becoming extremely emotional in the days after the 9/11 terrorist attacks, during which the companys main premises were destroyed, and credits his determined leadership with saving the firm at that time. In an often-cited work in management literature, Turner (1976) examined the official reports of a number of disasters from an organizational perspective, and concluded that disasters build up gradually over time and the signs should be apparent to management, but instead go unnoticed or ignored because of cultural rigidity which manifests itself in erroneous assumptions and reluctance to face unpalatable outcomes. Turners insight is borne out by Tibman (2009), who reports that the staff of Lehman refused to face the inevitable collapse of the firm until very near the end, and were foolish enough to swallow the widely distributed, popular 54 Lehman doctrine. Turner (1976) argues that: Disasters, other than those
Corporate strategy which describes the scope of the firm in terms of the industries and markets in which it will compete; and Business strategy which describes how the firm will operate within a particular industry or market, also sometimes called competitive strategy [Porter (1980)].
2 3
As an example of Fulds distinctive combative style, Tibman (2009) reports that the CEO had, for a time, installed a large stuffed gorilla near his office. For example, Time magazine voted Fuld one of the 25 people most responsible for the GFC.

Koch (2006) describes corporate strategy as, the strategy for the entire firm [and] is about the evolution of a firm, how it grows and develops over time. Gilad (2003) describes it as addressing portfolio issues or looking at the firm as a whole. In a lengthy definition, Kenneth Andrews (1980) defines corporate strategy as: The pattern of decisions in a company that determines and reveals its objectives, purposes, or goals, produces the principal policies and plans for achieving those goals, and defines the range of business the company is to pursue, the kind of economic and human organization it is or intends to be, and the nature of the economic and non-economic contribution it intends to make to its shareholders, employees, customers, and communities. Corporate strategy, therefore, is not only about setting long-term objectives for the firm, as a whole, but also articulating how the firm is going to achieve its objectives. It is not merely about articulating a practical vision but, arguably more important, stating how the vision is to be achieved.
In June 2010, the Financial Reporting Council5 (FRC) published the U.K. Corporate Governance Code, which describes corporate governance as being based on the underlying principles of all good governance: accountability, transparency, probity and focus on the sustainable success of an entity over the longer term [FRC (2010)]. According to the FRC the boards role is to [p]rovide entrepreneurial leadership of the company within a framework of prudent and effective controls which enables risk to be assessed and managed. And in particular an effective Board: provides direction for management; demonstrates ethical leadership; creates a performance culture that drives value creation without exposing the company to excessive risk of value destruction; is accountable, particularly to those that provide the companys capital [emphasis added] [FRC (2010)]. In its analytic guidelines for evaluating corporate ratings for firms, Standard and Poors (S&P) considers management and corporate strategy a key element of the criteria that form the foundation of the financial strength rating process and in particular, when assessing the companys strategic positioning, it is important to establish what managements goals are and how its strategy was developed [S&P (2009)]. Note that S&P concentrate on managements development of strategy noting only that the board is, for the highest rating, independent, highly qualified, and willing to exercise proactive judgment. However, such an emphasis would appear to entrench the inherent conflict of interest in the role of management vis--vis independent oversight of strategic risk. The National Association of Corporate Directors (NACD), the U.S. body
Role of the board in setting strategy

In its Principles of Corporate Governance, the Organisation for Economic Co-Operation and Development (OECD) describes the responsibilities of the board as: Reviewing and guiding corporate strategy major plans of action, risk policy, annual budgets and business plans; setting performance objectives; monitoring implementation and corporate performance; and overseeing major capital expenditures, acquisitions and divestitures [emphasis added] [OECD (2004)]. In the wake of the corporate governance failures identified by the Global Financial Crisis [FCIC (2011)], the Basel Committee of the Bank for International Settlements (BCBS) has issued a set of Principles for Enhancing Corporate Governance including defining the boards overall responsibilities: The board has overall responsibility for the bank, including approving and overseeing the implementation of the banks strategic objectives, risk strategy, corporate governance and corporate values. The board is also responsible for providing oversight of senior management [BCBS (2010)]. And in detail the board should Approve and monitor the overall business strategy of the bank, taking into account the banks long-term financial interests, its exposure to risk, and its ability to manage risk effectively [emphasis added]. The Committee of European Banking Supervisors (CEBS), likewise, has expanded on corporate governance for the banks that it supervises noting that The [Board4] has overall responsibility for the institution and should set the institutions strategy and risk appetite. The responsibilities of the management body should be clearly defined and approved. And To achieve good governance, an institutions management and supervisory functions should interact effectively to deliver the institutions agreed strategy, and in particular to manage the risks the institution faces [emphasis added] [CEBS (2010)].
supporting directors, have identified that [f]or most companies, the priority focus of board attention and time will be understanding and providing guidance on strategy and associated risk based on the underlying understanding of the companys strengths and weaknesses, and the opportunities and threats posed by the competitive environmentand monitoring senior managements performance in both carrying out the strategy and managing risk [emphasis added] [NACD (2008)]. The report into the Causes of the systemic banking crisis in Ireland neatly summarized the division of responsibilities between the board and management: A board is responsible for the safety and soundness of a bank, its depositors and shareholders (or members in the case of a building society). It determines the strategic direction, governance culture, risk appetite and procedures, which are designed to generate a return for the owners, while ensuring prudence at all times. [On the other hand] A CEO is appointed by a board to lead the day-to-day management of a bank and to ensure that the strategies, risk appetite and procedures, as set by
Note, recognizing the different corporate structures in member countries within the European Union, the CEBS uses the term management body to cover both unitary (i.e., U.K.) and dual structures (i.e., Germany). The FRC describes itself as the U.K.s independent regulator responsible for promoting high quality corporate governance and reporting to foster investment.
55
a board, are followed. A CEO is also responsible for providing executive advice and for formulating policy proposals for consideration by a board [emphasis added] [Nyberg (2011)]. While, regulators and governance bodies all seem to agree that the board plays a central role in strategy, the actual responsibilities are somewhat ambiguous. Boards are required to direct strategy [Nyberg (2011)], to set strategy (CEBS), to guide strategy (OECD, NACD), to approve strategy (BCBS), and review and monitor strategy [S&P (2009)]. The ambiguity is understandable, as a board may not have the experience necessary or the time available to do the extremely hard work of actually developing strategy. Formulation of a workable strategy requires deep understanding of the industry, the firms competitors, technological advances, customer demographics, regulatory initiatives, economic drivers, and a host of other factors. In practice, development of corporate strategy is usually undertaken by a highly skilled specialist group, typically called strategic planning and often reporting to the office of the CEO or similar high-level group. Strategic planning is an arduous time-consuming activity, which precludes the day-to-day involvement of independent directors of the board. It is an activity that is, quite reasonably, delegated to management. This situation might raise conflicts of interest since, even if not chairman, when the development of strategy is delegated to the CEO, he/she naturally becomes the owner and, human nature being what it is, may be somewhat defensive when the assumptions and outcomes are questioned. Where the CEO is also chairman, as in the case of Dick Fuld of Lehman, the conflict is even greater since the chairman controls the agenda for any board level review of strategy.
risk would be the effect of uncertainty on strategic objectives. In its 2010 Annual Report, the Bank of America has a comprehensive definition: Strategic risk is the risk that results from adverse business decisions, ineffective or inappropriate business plans, or failure to respond to changes in the competitive environment, business cycles, customer preferences, product obsolescence, regulatory environment, business strategy execution, and/or other inherent risks of the business including reputational risk [emphasis added] [BOA (2010)]. However, firms do not always recognize and manage risks to their strategy: An effective overall corporate strategy combines a set of activities a firm plans to undertake with an adequate assessment of the risks included in those activities. Unfortunately, many firms have forgotten the second part of that definition. In other words, there can be no real strategic management in financial services without risk management [emphasis added] [Kroszner (2008)]. Because, by its nature, strategy is long-term and the business environment in which the firm is operating is constantly changing, the achievement of strategic objectives is highly uncertain. How firms address this strategic uncertainty [Raynor (2007)], through the discipline of strategic risk management, will determine their long-term success and even viability. Kroszner (2008) warns: Strategic decisions about what activities to undertake should not be made unless senior management understands the risks involved; assessing potential returns without fully assessing the corresponding risks to the organization is incomplete, and potentially hazardous, strategic analysis [emphasis added]. In a seminal article in the Harvard Business Review, Slywotzky and Drzik
At this point it should be noted that a board is completely within its rights to select any strategy that it considers appropriate and to change their strategies at any time, provided that they judge the strategy and subsequent changes to be in the best long-term interests of the companys shareholders, i.e., they fulfill their fiduciary duty. Given this unfettered delegation of authority by shareholders to the board, only full disclosure of relevant details of a strategy and its risks will allow a shareholder to make rational investment decisions as to whether to accept the boards strategy or to withdraw capital. And, of course, shareholders must be able to hold the board accountable for such decisions, especially if they can ruin the shareholders capital.
(2005) described strategic risk management as a: means to devise and deploy a systematic approach for managing strategic risk. A 2010 survey of risk executives in financial institutions, by the Economist Intelligence Unit [EIU (2010)], found that Strategic risk management remains an immature activity in many companies, although the executives saw strategic threats as their biggest risks they face over the next 12 months. The survey concluded that these strategic risks can make the difference between survival and extinction but, in many cases, companies do not have the structured framework for identifying or mitigating them [EIU (2010)]. Federal Reserve Bank Governor Kroszner has identified the need for a new approach to strategic risk management: The current environment certainly presents some fundamental challenges for banking institutions of all types and sizes. Their boards of directors and senior management, who bear the responsibility to set strategy and develop and maintain risk management practices, must not only address current difficulties, but must also establish a framework for the inevitable uncertainty that lies ahead. Notably, the ongoing fundamental transformation in financial services offers great potential opportunities for those institutions able to integrate strategy and risk management successfully, and I will argue that survival will hinge upon
What is strategic risk?

Allan and Beer (2006) define strategic risks as those threats and opportunities that materially affect the ability of an organization to survive. Slywotzky and Drzik (2005) identify strategic risks as those that can disrupt or even destroy [a] business. ISO 31000, the new risk management standard [ISO 31000 (2009)], defines risk, in general, as the effect of 56 uncertainty on objectives. From that definition, it follows that strategic

such an integration in what I will call a strategic risk management framework [emphasis added] [Kroszner (2008)]. If strategic risks are not identified at the outset and not managed properly during execution, the strategy is likely to fail, at least in part, and given the amount of resources needed to execute any non-trivial strategy, will cause losses to shareholders and may, as in the case of Lehman, ruin the firm.
The senior European banking regulator, CEBS, also requires that [a]n institution should have a holistic risk management framework extending across all its business, support and control units, recognizing fully the economic substance of its risk exposures and encompassing all relevant risks Its scope should not be limited to credit, market, liquidity and operational risks, but should also include concentration, reputational, compliance and strategic risks [emphasis added] [CEBS (2010)]. In looking at what makes an effective board, the U.K. Financial Reporting Council notes that the boards role is to provide entrepreneurial leadership of the company within a framework of prudent and effective controls which enables risk to be assessed and managed [emphasis added] [FRC (2011)]. Strategic risk is different to other risks. It is the board itself that directs, or at least approves, the strategy, thereby accepting the risks associated with the chosen strategy the board itself is the risk-taker! Furthermore, the board is reviewing and monitoring its own decisions, which is a clear conflict of interest. If a board adopts a particular strategy, such as that of Lehman, they will, jointly, be responsible for monitoring their own performance in this respect. Over time, there will be a natural tendency to be subjective rather than brutally objective about the likelihood of future failure. The danger here is that a board will, to fulfill its legal responsibilities, adopt a box ticking approach, uncritically accepting managements assessment of risks, which appears to have been the case in Lehman [Valukas (2010)].
Role of the board in managing strategic risks

Having articulated a strategy, the board faces a number of questions [Johnson and Scholes (2002)] is the strategy: (a) Suitable does the strategy make economic and business sense? (b) Acceptable is the strategy acceptable to stakeholders; in particular are the risk/return objectives understood and satisfactory? (c) Feasible does the firm understand the need to acquire the resources necessary to execute the strategy and have the time to do so? And having embarked upon executing a strategy, other serious questions are raised in the light of changes in the business environment is the strategy still: (a) Relevant does the strategy still make sense, i.e., is it still suitable? (b) Achievable even if relevant, are the objectives of the strategy still realizable and do they remain acceptable? (c) Attainable is the plan to achieve the objectives of the strategy still on track, i.e. is the strategy still feasible? These questions, in fact, identify the highest level risks to fully achieving the objectives of the chosen strategy. Two major sets of strategic risks are apparent: 1. Strategic positioning risks does the strategy satisfy the Johnson and Scholes (2002) conditions, i.e., is it suitable, acceptable, and feasible? 2. Strategic execution risks while being implemented, is the strategy still relevant, achievable, and attainable? The OECD (2004) outlines the roles of the board and management with regard to risk: An area of increasing importance for boards and which is closely related to corporate strategy is risk policy. Such policy will involve specifying the types and degree of risk that a company is willing to accept in pursuit of its goals.6 It is thus a crucial guideline for management that must manage risks to meet the companys desired risk profile [emphasis added]. In their Principles for Enhancing Corporate Governance, the Basel Committee specifies boards responsibilities, including: Approve and monitor the overall business strategy of the bank, taking into account the banks long-term financial interests, its exposure to risk, and its ability to manage risk effectively [emphasis added] [BCBS (2010)].
The Financial Reporting Council warns boards of key factors, which can limit effective decision-making, such as: A dominant personality or group of directors on the board [such as a CEO and executive directors], which can inhibit contribution from other directors; and insufficient attention to risk, and treating risk as a compliance issue rather than as part of the decision-making process, especially in cases where the level of risk involved in a project could endanger the stability and sustainability of the business itself [i.e., strategic risk] [FRC (2011)]. Where there is a dominant personality, such as Dick Fuld, who will provide the challenge necessary to ensure that potentially firm-destroying risks are being properly assessed and managed?
Lehmans strategy and strategic risks

Lehmans strategic positioning
Valukas (2010 pp. 4) reports that Lehman made the deliberate decision to embark upon an aggressive growth strategy, to take on significantly greater risk, and to substantially increase leverage on its capital [emphasis added]. But Lehmans new strategy was not particularly new. It was, in fact, the expand at all costs strategy that was being pursued
6 This is often referred to as the firms risk appetite.
57
by all of the major investment banks that existed at the time [which] followed some variation of a high-risk, high-leverage model that required the confidence of counterparties to sustain [Valukas (2009)]. With regards to strategic positioning, there is evidence of major financial institutions acting as a herd since many of them have adopted the so-called universal banking model, being all things to everyone, everywhere [Nyberg (2011)]. Before the GFC, the worlds major banks operated in markets in many different countries, traded in many different markets in many different commodities, provided a full range of lending, deposit taking, wealth management and insurance services to a variety of customer segments, and provided support services such as funds transfers and custody around the world. The advantages of such a universal banking model are often characterized as [Molyneux et al. (1996)]: (a) economies of scale (being big); (b) economies of scope (growing bigger); (c) better management of capital; (d) reduced volatility of earnings (diversification); and (e) reduced risk. The major critique of the universal banking model is that it creates conflicts of interest between various parts of such a bank, most notably providing investment advice on one hand and trading on the other [Nyberg (2004)]. There is obviously considerable danger when everyone in a particular business sector believes roughly the same thing, such as the disastrous belief/assumption by firms, regulators, and rating agencies, before the GFC, that there had never been a nationwide housing crash in the U.S. since the Depression [FCIC (2011)]. In analyzing the flawed strategies that led to the collapse of all of Irelands major banks, Nyberg (2011) strongly criticizes regulators, external auditors, and the banks themselves for simultaneously adopting and accepting, with little challenge, the same business model, with the banks: Broadly ending up making fairly similar mistakes The systemic failure [to challenge assumptions] resulted in the dangers inherent in the business models remaining undetected until it was much too late [emphasis added]. While the wisdom of the crowd may be right in many situations, it only takes one collective misjudgment to precipitate chaos. Michael Porter (1996), the guru of competitive strategy, warns that [t]he more benchmarking companies do the more they look alike. The more that rivals outsource activities to efficient third parties, often the same ones, the more generic those activities become. As rivals imitate one anothers improvement in quality, cycle times and supplier relationships, strategies converge and competition becomes a series of races down identical paths that no one can win [emphasis added].
Lehmans overall strategic positioning, as JP Morgan and other banks have demonstrated, was not completely flawed, though its logic was seriously undermined by the events that led to the GFC. But in order to implement a strategy successfully a firm must not only have a clear strategic direction but also the capacity to achieve its strategic objectives. Capacity in this context means, sufficient capital, skilled and knowledgeable staff, and importantly time. Following the 9/11 attacks on New York, Lehman had taken a deliberate decision not to fire staff, in reaction to the inevitable market downturn, but, in a typically contrarian move, chose to hire high-performing staff that had been let go by competitors [Tibman (2009)]. In some respects, this was CEO Dick Fulds finest hour, and resulted in Lehman having the skills necessary to embark on its new strategy. But capital was certainly a problem, Lehman was already highly leveraged as regards to its low capital base [Valukas 2010)] and time was, some 18 months later, to become Lehmans enemy. Ultimately, Lehmans capital base was not strong enough to withstand the losses that occurred as a result of its excessive exposure to the declining U.S. property market. In just 12 months, Lehmans reported that its net leverage ratio7 had gone from 14.5:1 to a dangerously high 16:1 [Valukas 2010)]. However, Valukas concluded that this ratio was in reality higher as the ratio had been enhanced (i.e., lowered) through the use of so-called REPO 105 transactions (see below). The last few months of Lehmans life was a constant round of largely unsuccessful capital raisings to cover asset write-downs [Tibman (2009)] but, until the end, Fuld maintained, despite misgivings from Fed chairman Ben Bernanke, that there was no capital hole at Lehman Brothers [FCIC (2011)]. Turner (1976) argues that disasters occur because people ignore the importance of warning signs that, with hindsight, clearly point to the buildup of danger. A common feature of disasters is that those who are closest to the problem fail to call for help, which has been attributed variously to fears of causing unnecessary alarm, psychological denial of the danger, or the assertion of the individuals invulnerability [Turner (1976)] all of these were present in Dick Fulds approach to the crisis [Tibman (2009)]. So while the overall strategic position sought by Lehmans was not unattainable, the execution of the strategy was seriously defective Lehman just did not have the capital capacity to pursue their chosen strategy at the speed at which they attempted to do so.
Lehmans Strategic execution

Though written in the mid-1990s, Porters analysis could just as easily be used to describe the madness that took hold of Wall Street in the subprime CDO market [Levin-Coburn (2011)] or the folly of unsustainable 58 lending that destroyed Irelands banking system [Nyberg (2011)].
7 Net leverage ratio is the ratio of net assets against its tangible equity capital base [Valukas (2010)].
It is not possible to know what would have happened to Lehman (and

the global financial system) if the new Lehman strategy had been executed at a more leisurely pace, with proper management of all of the risks involved. With hindsight, Lehman began an ill-advised program to purchase and hold mortgage assets, not only the toxic CDOs that caused trouble for firms such as AIG and Bear Stearns [FCIC (2011)], but also massive investments in leveraged lending and commercial real estate assets [Tibman (2009)].
whom? There are several stakeholders that have an interest in knowing the overall direction of a firms strategy, in particular: 1. Regulators will the firm be viable in the long term? 2. Investors is the firm a good long-term investment? 3. Employees how can decisions be made to advance the chosen strategy? The last category is pivotal: if employees do not understand the full
In their 2006 Annual Report, Lehman gave no clue as to the impending change in their core strategy, which was focused on traditional investment banking strengths: As trusted advisors, we have strategically positioned ourselves to help our clients take advantage of the forces of global change. Our carefully targeted global strategy remains rooted in our core competenciesincluding the effective management of risk, capital and expenses [Lehman (2006)].
implications of a firms strategy, how can they make rational trade-off decisions on short- versus long-term benefits and risks? If, however, employees know and understand the strategy but execution details are not released to investors (as happened with Lehman), the presence of insider information for one class of shareholders (employees) raises serious concerns of market transparency. Interestingly, the full significance of the new strategy was not widely
Valukas (2010) documents the explosive doubling during 2007 of Lehmans leveraged loan (or high-yield) investment banking activities. To gain this business Lehman began, to the chagrin of rating agencies, to loosen its contractual standards, soon exceeding the firms internal risk appetite limit for this business. To accommodate the growth in leverage lending, senior management loosened risk controls and, significantly, approved individual deals that exceeded single transition limits. In its Annual Report for the financial year 2007, Lehman announced record profits crediting not its new storage strategy but its client focused strategy as the key to its success [Lehman (2007)]. The Annual Report did state that the firm continued to drive diversified growth as a key pillar of its corporate strategy but did not stress the significant change in strategy from moving to storage other than to issue a pro-forma warning that global financial markets and economic conditions materially affect our businesses. Market conditions may change rapidly and without forewarning [Lehman (2007)]. With hindsight, the Annual Report held a prescient warning for Lehman shareholders that our liquidity could be impaired by an inability to access secured and/or unsecured debt markets, an inability to access funds from our subsidiaries, an inability to sell assets or unforeseen outflows of cash or collateral [Lehman (2007)]. By 2008 Lehmans management made a conscious decision to exceed the risk appetite limits on leveraged loans with the result that the leveraged loan exposure soon doubled the limit amount [Valukas (2010)]. There was internal opposition to this rapid growth, most notably from Michael Gelband, head of Fixed Income, and Madelyn Antoncic, Chief Risk Officer, who were both to leave their posts because of their opposition to the aggressive growth strategy [Valukas (2010)]. The development of a corporate strategy raises an interesting question should the strategy and its strategic risks be disclosed and if so to
known within the firm. Tibman (2009), a long-term senior investment banker in Lehman, notes that while staff knew that the firm was investing heavily in property assets its natural conservative approach to credit risk, plus the belief that the firms risk management had over the years improved so greatly meant that most felt immune to the massive losses from time to time registered by other firms. This, of course, was a delusion shared by many staff until the end. Some insiders, such as Larry McDonald, a trader in the Lehman mortgage business, claim that the writing was on the wall for over a year before the collapse but that no one was able to do anything about it because of the aggressive posture of management [McDonald and Robinson (2009)]. Nor was the board kept fully informed. Lehman was required by its regulator, the Securities and Exchange Commission (SEC), to run stress tests against its portfolio of risks which were designed to measure tail risk a one in ten year type event [Valukas (2010)]. But after changing its strategy in 2006, Lehman did not update the stress tests to cover the new risks that involved principal investments, and did not inform the board of the full implications [Valukas (2010)]. While the boards Finance and Risk Committee received the results of the stress tests they were not informed by management that many of the firms commercial real estate and private equity investments were excluded from the firms stress tests. Nor did management inform the full board of the implications of relaxing certain risk limits to facilitate the strategy [Valukas 2010)]. Shareholders were also kept in the dark. To hide the growing problems in its overblown balance sheet, Lehman resorted to window dressing, moving assets from the balance sheet over reporting periods using socalled REPO 105 transactions short term asset sales with guaranteed repurchase at a later time [Valukas (2010)]. A former Financial Controller testified that the only purpose or motive for [REPO 105] transactions was reduction in the balance sheet and that there was no substance to 59
the transactions [Valukas (2010)]. The NACD (2008) recognizes the importance and difficulties of disclosure of strategic issues: Boards should also consider reaching out and developing stronger relationships with investors through candid and open dialogue. In particular, boards should consider ways to engage large long-term shareholders in dialogue about corporate governance issues and long-term strategy issues, recognizing that the boards fiduciary duties with respect to these issues mandate that the board exercise its own judgment.
FCIC, I blame the management teams [of various firms impacted by the GFC] and no one else [FCIC (2011)]. Though concluding that the firms growth strategy was flawed so much so that its very survival was in jeopardy Lehmans bankruptcy examiner, nonetheless, did not find that there were colorable claims8 against the directors nor senior officers for breach of fiduciary duties for failing to: (a) observe official risk management policies and procedures; (b) inform the board of the level of risk assumed; nor (c) monitor the firms risk taking activities9 [Valukas (2010)]. In retrospect, it might be surprising that, despite the fact that the board and management embarked on a seriously flawed strategy that led to the collapse of the firm within a few years and failed to monitor the risks in the execution of that strategy, shareholders appear to have no legal redress. The bankruptcy examiner explains that under the prevailing law,10 officers and directors business decisions are generally protected from personal liability by the business judgment rule and even if the business judgment rule does not apply, there is no liability unless the officer or director was grossly negligent [Valukas (2010)]. Valukas (2010) also notes the proof necessary to defeat the business judgment rule and establish gross negligence is particularly high with respect to risk management and financial transactions. Since it proved difficult to assign personal responsibility in the Lehman
Who was to blame for Lehmans failure?

The risks of Lehmans new strategy were apparent to the board and management from the very outset, as evidenced by the 50% increase in the firms risk appetite limit when the new strategy was agreed by the board in 2007 [Valukas (2010)]. Such a sizeable increase in a critical control alone should have highlighted to the board that there was a need to forcefully oversee the execution of the new strategy, especially as the CRO at the time had argued strongly for a lower limit. On the contrary, during 2007 and 2008, the firms overall risk appetite limit was almost doubled in three increases, with little review (other than the Executive Risk Committee), despite assurances to the SEC that the hard limit was a meaningful constraint on Lehmans risk taking. Management was also permitted to change the rules used to calculate the overall risk usage, removing some large transactions that would have exceeded the board agreed limit [Valukas (2010)]. Valukas (2010) reports that repeated concerns of risk management professionals, such as the removal of specific limits on certain transactions, and questions by external agencies, such as Moodys, were routinely ignored by management and not reported to the board. However, the lack of specific information cannot absolve the board from its share of responsibility, since it must have been obvious that all was not well with the execution of the strategy, given the turnover of very senior control personnel, most especially the CRO and CFO. It is the role of the board to question senior management on such serious issues but directors appear not to have been as involved as should be warranted by such a significant change of strategic direction. As Turner (1976) notes, large-scale disasters are rarely caused by one person, but are the result of the actions, or more properly lack of actions, by many people over a period of time. With hindsight it is apparent that collectively the board and management must be held accountable for failing to create an effective strategic risk management framework. Although the board agreed with the new strategy proposed by management, it failed to put in place the necessary structures to ensure that the attendant risks were being managed properly and reported fully to the board for oversight. A situation, such as this, is a failure of corporate 60 governance, summed up by the CEO of JP Morgan who testified to the
case it is important, in the interests of shareholders and taxpayers, that the issue of strategic risk management is addressed by regulators so that a similar event does not happen in future.
Regulatory response
The Lehmans case raises many issues concerning the operations of the global financial markets, but it also raises some serious questions about corporate governance, in general, and strategic risk, in particular:
How should the development of strategy be governed, and who is responsible for creating, approving, monitoring, and reporting on changes to strategy?
Who is responsible for identifying and managing the strategic risks inherent in any new strategy? In what form should changes to an existing strategy and the resulting strategic risks be communicated to shareholders and regulators?
A colorable claim is one that that on appropriate proof would support a recovery, in other words, a claim that has a good chance of success. 9 The examiner did, however, find that there were colorable claims against various officers of the firm for a variety of actions including hiding problems though the REPO 105 program [Valukas (2010)]. 10 Lehman was subject to Delaware state law.

Can a board that is chaired by a firms CEO ever do proper due diligence on managements strategic proposals since there is obviously a conflict of interest in the dual role?
(2011)]. In October 2008, RBS, one of the largest banks in the U.K., failed, requiring substantial support by the U.K. government, incurring losses of some 35 billion at the time of the reports publication. The FSAs report is some 450 pages long but, in summary, concludes that the failure was a result of poor management decisions, deficient regulation and a flawed supervisory approach [FSA (2011)]. A key risk factor highlighted by the report was the strategic acquisition of ABN AMRO in 2007, in a consortium with Barclays and others, which RBS funded mainly by debt that greatly increased RBSs vulnerability [RBS (2011)]. The FSA noted that RBS management and board undoubtedly made many decisions which, at least in retrospect, were poor. They took risks, which ultimately led to failure [emphasis added] The ABN AMRO acquisition illustrates the point. The due diligence conducted was inadequate to assess the risks [FSA (2011)]. The FSA report concluded that the decision to acquire ABN AMRO is generally acknowledged to have been a critical strategic error and that it was not apparent that the board discussed in sufficient depth the risks involved in the acquisition, including its exceptional complexity, unprecedented scale and how it was to be financed, especially as so little effective due diligence was possible. In a damning indictment, the report concluded that [t]he judgement of the RBS Board in respect of the ABN AMRO acquisition was not characterized by the degree of moderation and sensitivity to strategic risk appropriate to a bank [emphasis added]. There are many similarities to the Lehmans failure not least that Sir Fred Goodwin, the CEO at the time of the failure, was perceived to be an overly dominant personality, so much so that the FSA had, in 2003, identified Goodwins dominance to be a risk, in particular it could [r]esult in a lack of effective challenge by the board and senior managers to the CEOs proposals, resulting in risks being overlooked and strategic mistakes being made [emphasis added] [FSA (2011)]. In an interesting admission of their own deficiencies, the FSA noted that their approach had been in retrospect, insufficiently robust [FSA (2011)] and in particular: Did not define it as part of Supervisions role to question the overall business strategy. As a result supervisors did not always reach their own judgements on the key business challenges and strategic risks in firms business models, based on in-depth, rigorous review involving specialists where appropriate. In other words, the senior U.K. banking regulator did not feel able/con-
Although Basel II (2004) excludes strategic risk from specific capital rules, because it is not easy measurable,11 the Committee does, however, recognize that the analysis of a banks current and future capital requirements in relation to its strategic objectives is a vital element of the strategic planning process and requires, under Pillar 2 (Supervisory Review), A banks strategic plan to clearly outline the banks capital needs, anticipated capital expenditures, desirable capital level, and external capital sources. Senior management and the board should view capital planning as a crucial element in being able to achieve its desired strategic objectives [Basel (2004)]. (This prescription, however, does not assign accountability, blurring the roles of the board and management). In the push to implement Basel II in 2008, and overtaken by the turmoil of the GFC, the measurement of strategic risk was not a major focus of either regulators or banks it remained in the too hard basket. The U.S. regulations for implementation Basel II are clear on the need to manage strategic risk, in addition to other risks: A bank should ensure that adequate capital is held against all material risks, and that capital remains adequate not just at a point in time, but over time, to account for changes in a banks strategic direction, evolving economic conditions, and volatility in the financial environment [emphasis added] [Fed (2008)]. In May 2011, the Prudential Regulatory Authority (PRA) of the Bank of England, the revamped U.K. banking regulator, announced a new approach to banking regulation, stating that, in future, prudential regulation will be forward-looking: Seeking to assess whether, on the balance of risks, there are vulnerabilities in firms business models, capital and liquidity positions, governance, risk management and controls that cast into doubt their future financial soundness [and echoing Slywotzky and Drzik (2005), regulation will] consider whether and how the wider external macroeconomic and business context may affect the execution of a firms business model in a variety of different scenarios [emphasis added] [BOE (2011)]. In effect, these requirements are a charter for strategic risk management within regulated firms to identify and assess risks to future viability.
Lehman not unique

Unfortunately, Lehman was not unique in failing to manage its strategic risks prior to the GFC. In its 2011 report into the failure of the Royal Bank of Scotland (RBS), the Financial Services Authority (FSA), the U.K. banking regulator, was scathing about the approach of the RBS board to managing the risks inherent in its opportunistic growth strategy [FSA
fident at that time to question the business strategies, however flawed, being followed by a financial institution that was without doubt systemically important to the U.K. economy. The FSA report goes into detail of
11 The Committee does, however, expect the industry to further develop techniques for managing all aspects of these risks.
61
the poor management decisions that led up to the failure of RBS much of which is outside of this paper and provides an excellent starting point for further research into failures of strategic risk management.
References

Summary
Large financial institutions are unlike other private firms, since their failure has the potential (through mutual dependencies inherent in the global financial system) to quickly spread contagion to other such firms and then to disrupt not only the local but also the global economy, as was demonstrated in the GFC.

Strategic risk is arguably the greatest risk facing such financial institutions as it has the capacity to totally destroy the firm, as evidenced in the case of Lehman, which is the subject of this paper. In the wake of the GFC, corporate and prudential regulators have highlighted the need for improved corporate governance and risk management. However, the precise roles of the board and management as regards the development of corporate strategy and the management of risks to the strategy (i.e., strategic risks) remain somewhat blurred. It is clear that at the very least that boards should approve corporate strategy, and by implication accept the risks inherent in such a decision, but the ongoing management of such strategic risks creates a serious conflict of interest since the people responsible for making strategic decisions are also accountable for monitoring their effectiveness. In terms of large systemically important financial institutions (SIFIs), there appear to be no independent checks that their long-term strategies may be flawed, either in positioning or in execution. There is a need, therefore, to develop a systematic approach to the evaluation of risks to corporate strategy, to ensure that large banks do not collapse and spread contagion to the international banking system. To achieve this, the governance of how strategy is developed, implemented, monitored, and managed must change, and boards of directors must put in place the necessary governance structures to ensure that their shareholders are protected in the long term which is, after all, their primary responsibility. As the case of Lehman amply demonstrates, there is a need to improve corporate governance in this key area. In opposing the 1999 Gramm-Bailey legislation that repealed the GlassSteagall Act of 1933, which had separated retail banking from riskier investment banking activities since, Senator Byron Dorgan presciently warned that we will look back in ten years time and say we should not have done this because we forgot the lessons of the past [Dorgan (1999)]. Unless significant changes are made to the rules that govern large banking conglomerates, particularly with regards to their management of strategic risks, memories will fade and similar events to those at 62 Lehman Brothers will happen in future.
Allan, N. and L. Beer, 2006, Strategic risk Its all in your head, Working Paper 2006.1, School of Management, University of Bath Andrews, K., 1980, The concept of corporate strategy, 2nd Edition, Dow-Jones Irwin Basel, 2004, International convergence of capital measurement and capital standards a revised framework, Bank for International Settlements, Basel Committee on Banking Supervision Basel BCBS, 2010, Principles for enhancing corporate governance BCBS 176, Bank for International Settlements, Basel Committee on Banking Supervision Basel BOA, 2010, Bank of America 2010 Annual Report, Bank of America, Charlotte NC BOE, 2011, Our approach to banking supervision, Prudential Regulation Authority, Bank of England CEBS, 2010, Consultation paper on the guidebook on internal governance (CP 44), Committee of European Banking Supervisors, London Dorgan, B, 1999, Speech Opposing the Gramm Bailey Act, 4 November, U.S. Senate EIU, 2010, Fall guys risk management in the front line, Economist Intelligence Unit, November FCIC, 2011, Final report of the National Commission on the Causes of the Financial and Economic Crisis in the United States, Financial Crisis Inquiry Commission, Washington, January FED, 2008, Supervisory guidance: supervisory review process of capital adequacy (Pillar 2) related to the implementation of the Basel II advanced capital framework, July, Federal Reserve Board, Washington, July FRC, 2010, The UK corporate governance code, Financial Reporting Council, London FRC, 2011, Guidance on board effectiveness, Financial Reporting Council, London FSA, 2011, The failure of the Royal Bank of Scotland, Financial Services Authority, London Gilad, B., 2003, Early warning, Amacom, New York Grant, R. M., 2008, Contemporary strategy analysis, 6th Edition, Blackwell, Oxford Institutional Investor, 2007, Best CEOs, Institutional Investor, 22nd January ISO 31000, 2009, AS NZS ISO 31000 2009 risk management Guidelines on principles and implementation of risk management, SAI Global, Australia Johnson G. and K. Scholes, 2002, Exploring corporate strategy, 6th edition, Pearson Education Ltd, Harlow Koch, R., 2006, The Financial Times guide to strategy, 3rd Edition, Pearson, London Kroszner, 2008, Strategic risk management in an interconnected world, Federal Reserve Board, Washington, October Lehman, 2006, Lehman Brothers 2006 Annual Report, Lehman Brothers Holdings, New York Lehman, 2007, Lehman Brothers 2007 Annual Report, Lehman Brothers Holdings, New York Levin-Coburn, 2011, Wall Street and the Financial Crisis: anatomy of a financial collapse, Permanent Subcommittee on Investigations United States Senate, Washington, April McConnell, P. J., 2012, The governance of strategic risk in systemically important banks, Journal of Risk Management in Financial Institutions, Vol. 5 No. 2, Forthcoming McDonald, L. G. and P. Robinson, 2009, A colossal failure of common sense the incredible inside story of the collapse of Lehman Brothers, Random House, New York Molyneux, P., Y. Altunbas, and E. Gardener, 1996, Efficiency in European banking, Wiley, Chichester NACD, 2008, Key agreed principles to strengthen corporate governance for U.S. publicly traded companies, National Association of Corporate Directors, London Nyberg, L., 2004, How do conflicts of interest in universal banking emerge and what are the arguments for a separation of commercial and investment banking? Speech, August 2004, Swedish Central Bank, Stockholm Nyberg L., 2011, Misjudging risk: causes of the systemic banking crisis in Ireland, Ministry of Finance, Dublin, March OECD, 2004, Principles of corporate governance, Organization for Economic Co-Operation and Development, Paris Porter, M. E., 1980, Competitive strategy, Simon & Schuster, New York Porter, M. E., 1996, What is strategy, Harvard Business Review, November-December, 61-78 Raynor, M. E., 2007, The strategy paradox, Doubleday, New York S&P, 2009, Management and corporate strategy, Standard and Poors, New York Shiller, R. J., 2008, The subprime solution, Princeton University Press, Princeton Slywotzky, A. J. and J. Drzik, 2005, Countering the biggest risk of all, Harvard Business Review, April Steiner, G., 1979, Strategic planning, Free Press Tibman, J., 2009, The murder of Lehman Brothers, Bricktower Press, New York Turner, B., 1976 The organisational and interorganisational development of disasters, Administrative Science Quarterly, 21, 387-397 Valukas, A. R., 2010, In re Lehman Brothers Holdings Inc. Report of Anton. R Valukas Bankruptcy Examiner, March 11, United States Bankruptcy Court Southern District Of New York, New York
PART 1
Explaining Credit Default Swaps Pricing for Large Banks

nci tker-Robe International Monetary Fund Jiri Podpiera International Monetary Fund1
Abstract
The credit default swap spread appears to be a potent predictor of financial distress, however, its determinants have not yet been fully understood. This paper suggests a banking-industry-specific approach to the pricing of banks credit risk. It shows that the pricing of banks credit default swap spreads could be approximated by a risk frontier derived from portfolio theory, where banks are viewed as leveraged portfolios. The commonly applied structural credit risk model adds very little to the explanatory power of the portfoliobased approach, thereby underlying the importance of the new approach to banks credit risk.
1 The authors would like to thank Christian Schmieder, Turgut Kisinbay, Manmohan Singh, and participants at the IMF seminar for helpful comments and suggestions.
63
The financial crises experienced in recent decades prompted efforts to identify indicators that would be helpful in predicting crises. Early detection of distress could trigger prompt interventions to take preemptive steps toward reducing the risk of a crisis. The ongoing global financial crisis has intensified these efforts, with policymakers looking for indicators and methods that could assist in a timely identification of highly vulnerable banks and banking systems. The credit default swap (CDS) spread appears to be one of the most promising indicators that can quite accurately reflect conditions in the financial sector at an early stage. Figure 1 provides some support it shows a strong link between the CDS spread prior to an official government intervention in a bank and the extent of needed recapitalization (a form of intervention), with the CDS spread appearing to correlate positively with the size of the losses.
2
3,5
Natixis Raiffeisen Zentralbank Unicredit SPA UBS AG Royal Bank of Scotland Anglo-Irish Bank Intesa Sanpaolo Erste Group Bank Royal Bank of Scotland Monte dei Paschi di Siena Lloyds BNP Paribas Societe Generale Credit Agricole BNP Paribas Dexia SA
2,5
1,5
Commerzbank Commerzbank
0,5
0 0,0 0,5 1,0 1,5 2,0 2,5 3,0
Notwithstanding the current usefulness of the CDS spread as a compound statistic of credit default risk, understanding its underlying pricing mechanism would permit a deeper analysis of factors behind distress and thus possibly lead to even better predictability of mounting default risks. However, to date, structural credit risk models have not satisfactorily explained variability in financial and non-financial corporate credit spreads, which has been established in the literature as the credit spread puzzle [see Duffee (1998)]. Consequently, this paper tries to address the credit spread puzzle by emphasizing specifics of the financial industry for pricing credit default risk for banks. The contribution of this paper is threefold: first, it focuses on a specific industry, large banks, unlike most of the relevant literature, which mixes industries [for instance, CollinDufresne et al. (2001); Blanco et al. (2005); and Ericsson et al. (2009)]; second, it formulates a testable hypothesis for pricing credit risk of large banks; and third, it tests the hypothesis on large European banks and verifies its robustness against an exhaustive set of variables. As such the findings in the paper invite new research in credit risk models for banks. The commonly used model of credit risk is based on the value of a firm relative to its debt. In a nutshell, the more the value of a company approaches the value of its debt, the more risky the company becomes, and vice versa (i.e., measuring the distance to default). Since Merton (1974), the equity is viewed as a call option on a firms assets with maturity T; the equity price is the spot price and the maturing debt at time T per share is the strike price. Using equities as proxy for a companys value, the credit default risk (corporate credit spread) is a function of the debt per share, volatility of equity price, and the risk-free interest rate. The theory can be applied to Credit Default Swaps (CDS) as well, as shown by Benkert (2004), but since CDSs are already spreads, the risk-free interest rate becomes non-essential. The pricing of call equity options involves the same arguments, however, the volatility of the equity price is the one perceived (expected) by investors. As such it informs about the investors view on the future value of the company. 64
2
The actual needed recapitalization of LCFIs in the sample during August2008 through May2009, measured by the ratio of injected capital (and in one case asset purchase) over banks total assets, that is, the capital needed to cover losses over total assets. The credit default swap (CDS) 5Y spread is measured as a 30-day average before the government intervention took place. Figure1 CDS spreads (vertical axis, p.p.) and recapitalization (percent of total assets)
Mertons model can be applied to CDSs for banks as well, since there is certain similarity between banking and other industries; a banks equity increases when the bank runs profits and decreases when it incurs losses. For this purpose, we use the Expected Default Frequency (EDF), produced by Moodys KMV, which is basically a Merton-type statistic, combining a measure of leverage and volatility of equities. However, banking distinguishes itself from other industries by managing borrowed funds and having prescribed capitalization standards. The value of a bank is crucially dependent on its operations, since losses of only a few percentage points of total assets can ruin the equity (equity accounts only for a small portion of banks assets). Consequently, financial soundness indicators, especially those related to profitability, serve as leading indicators for the value of the bank and its credit default risk. Banking is a leveraged portfolio management and thus, as proposed in this paper, pricing banks credit spreads might be based on the portfolio theory [Markowitz (1952)], since they leverage on depositors and other fund lenders and invest in risky projects. Every portfolio has its risk and return. Consequently, the portfolio theory can be applied to banks to assess their risk-return profiles. Credit spreads for banks, in general, are priced by risk-neutral investors investors that are indifferent between buying and selling credit risk protections. Thus, broadly, the CDS spread measures the portfolios risk (proportion of losses) lowered for its rate of return, since the return can be used to cover losses.
The regression yields recapitalization = 0.62(s.e. 0.09) CDS spread; Rs (adj.) = 0.73.

In this paper, credit default risk is measured by CDS spreads. In a CDS contract, the buyer of a protection against default is entitled to compensation in the event of the issuers default on the underlying asset. The seller of the protection receives periodic payments until the default event occurs or the contract matures. The spread is the price of the credit default risk on the underlying asset. There are several well-recognized advantages of using CDS spreads rather than corporate bond credit spreads for assessing credit risk:
Deriving the CDS spread: the risk frontier

CDS spreads determine banks funding costs, profitability, and the value of their equities. Banks with low CDS spreads have better access to cheaper funding, which allows them to charge a competitive net interest margin without taking excessive risk. Such banks are expected to have low impaired assets, low operating expenses (due to low monitoring and collection charges), and thus high ROAs, hence low CDS spreads. High returns will also increase the value of banks equity. Consequently, banks aim to minimize the credit risk from the point of view of risk-neutral investors, who monitor them and supposedly view banks as leveraged portfolios and price their CDS spreads according to a risk frontier that can be derived from a portfolio theory [Markowitz (1952)]. For a bank in country k, the task is to:
N N Minwj,tFtAt { j=1wj,t rj,t q [( j=1wj,t ij,t + Ft Pt C(At))(1 )]}
First, the CDS spread is a more direct measure of credit risk than corporate bonds credit spreads. The credit spread from corporate bonds is derived by subtracting a risk-free interest rate from the corporate bond yield. However, the choice of the reference risk-free asset and the method to derive zero-coupon yield curve affects the level of the spread. In contrast, the CDS is already a spread, since it is priced such that no arbitrage exists between investing in a riskless bond and buying a defaultable bond, and protecting against default by purchasing the CDS.
(1)
by choosing wj,t (the weight of asset j in the portfolio at time t), fees per dollar of assets Ft, and the size of assets At. There are N assets. The ij,t stands for the interest rate margin on asset j (a return on an asset less the interest rate paid on borrowed funds). By analogy to the portfolio theory, the expression N wj,t rj,t measures the amount of risk of the portfolio, j=1 while the return is represented by [(N wj,t ij,t + Ft Pt C(At))(1 )]. The j=1 risk tolerance is denoted by letter q; Pt stands for provisions for impaired assets relative to total assets; and C(.) is a cost function. The applicable corporate income tax rate is denoted by . The minimization is constrained by: N wj,t ij,t + Ft Pt C(At) 0 , j=1 (2)
Second, the CDS spread does not seem to contain liquidity components as corporate bond credit spreads. Driessen (2005) shows that more than half of the variation in corporate bond credit spreads relates to time-varying compensation for liquidity and to some extent the tax treatment of corporate bonds (while corporate bonds are subject to state tax, government bonds are not). In contrast to credit bond spreads, the CDS spread is not distorted by tax issues and according to Fabozzi et al. (2007) does not contain any liquidity premium.
Third, the CDS spread leads in the credit risk price-discovery process over corporate bond credit spreads. According to Blanco et al. (2005), the non-default components in corporate bond credit spreads make responsiveness of the credit spreads to changes in credit quality slower and weaker than that of the CDS spread. The authors also show that the CDS spread represents the upper bound of credit risk, while credit spreads the lower bound. which states the condition of non-negativity of return (break-even condition), and ij,t = 1-j,t b 1 + s , r j,t E(1-j,t) k,t (3)
The hypothesis that the pricing of banks CDS spreads is based on the performance of their current and expected leveraged portfolios was tested here and robustly confirmed on a set of 29 large European banks during 2004-2008. Large European banks were chosen for their homogeneity in CDS contract specifications and definition of financial soundness indicators. In addition, five-year CDSs for large European banks are well liquid. It has been confirmed, in an extensive robustness analysis, that the pricing of CDS spreads is well approximated by variables implied by portfolio theory. At the same time, the EDF added only little explanatory value to the portfolio model, suggesting that a banking industry-specific approach to pricing credit risk could potentially address the credit spread puzzle for banks. In practice, these findings suggest that financial soundness indicators related to profitability serve as leading indicators for banks credit default risk.
which describes the structure of the interest rate margin. It links ij,t to the sk,t (the slope of the yield curve for all banks in a country k) and to the risk rj,t predetermined according to the credit risk manual for each type of project, the share of impaired assets j,t, and the expected default rate E(j,t) of assets. Banks earn on the maturity transformation and thus changes in the slope of the yield curve translate into banks interest rate margin regardless of projects riskiness. The ratio E(1-j,t)/1-j,t= 1 + j,t, where j,t is the discrepancy between projects expected and actual default rates. Other basic conditions need to be satisfied: C (.)>0; N wj,t j=1 =1; Ft F*; Pt = (t). By letter F* we denote a cap on fees per dollar of assets. 65
The banks risk objective function (after incorporating the equation (3)) along with the break-even condition (and Lagrange multiplier ) yields:
N N Minwj,tFtAt [{ j=1wj,t ij,t (1+j,t) bsk,t ( j=1wj,t (1+j,t) qROAt}-
value of a bank based on expected profitability of the banks portfolio. For instance, if investors anticipate upcoming losses for a bank, they will expect declining price of equity in the future. This is associated with an increasing volatility implied by option prices. Hence the volatility implied by option prices should be reflected in the CDS pricing as well.
{1
Pt + C(At) N wj,t ij,t + Ft j=1
}],
(4) Consequently, the final reduced from specification reads as follows: CDSt = CDSt-1 + + t,
N-1 j=1 w*j,t ij,t +
where ROAt is the return on assets at time t. Given the parameters (,b,q,), the equation (4) describes a risk frontier for various choices of wj,t, Ft, and At, and thus can be used to price the credit default risk of banks.
t bsk,t qROAt + EFFt + Vt + c (8)
where t~N(0,) and V stands for the implied volatility from option prices. If banks CDS spreads are indeed priced as if banks were leverage portfolios, the CDS spreads would be derived from the risk profile of banks portfolios. Assuming that the risk frontier is an approximation for CDS
N N spreads, and by denoting t = j=1wj,t ij,t j,t bsk,t ( j=1wj,t j,t), the expression (4) can be rewritten as follows:
Parameters ,b,q,,,, and represent the sensitivities of risk-neutral investors to the variables in the risk frontier.
Estimating the CDS spread

Definition of variables
The dependent variable in the model is the CDS spread. Banks CDS spread daily last price quotes represent swaps on senior debt with a maturity of five years (most liquid maturity) and were downloaded from Bloomberg. The dependent variable is the yearly average of daily data. Independent variables in the model are presented in the structure of CAMELS an acronym for Capital Adequacy, Asset Quality, Management Quality, Earnings Potential, Liquidity, and Sensitivity to Market Risk. Capital adequacy is measured by four alternative variables: Tier I and Tier II ratios, Leverage ratio, and Z-score.
CDSt = N w*j,t ij,t bsk,t qROAt + j=1
Pt + C(A*t) N w*j,t ij,t + F*t j=1
+ t,
(5)
where w*j,t, F*t, A*t are each banks optimal choices of wj,t, Ft, and At, respectively. The error term t has a nown-trivial structure and would require using specification with bank-specific effects (e) and inertias. Autocorrelation is also suggested by findings of unit root in credit spreads by Pedrosa and Roll (1998) and Bierens et al. (2003); and Blanco et al. (2005) used a specification with a lagged dependent variable. Consequently, the specification (5) is amended with a lagged dependent variable.
Tier I capital represents the ratio of capital (shareholders capital, reserves, and hybrid capital to certain limits) divided by risk-weighted assets and is taken as reported by each bank. Tier II capital is a ratio of secondary capital (undisclosed reserves, revaluation reserves, general provisions, hybrid instruments, and subordinated term debt) to risk weighted assets. Tier II is computed as the residual from total capital adequacy ratio and Tier I capital ratio obtained from Bloomberg. Both capital ratios represent capital buffers for loss absorption and thus they should correlate negatively with the credit risk. Leverage measures the size of average total assets relative to average total common equity. It is one of the standard indicators implied by the structural approach to the pricing of default risk. Higher leverage (lower capitalization) would correlate positively with default risk. Z-score is a derivative measure of bank capitalization and gauges available funds for loss absorption. Following Boyd and Runkle (1993) it is computed as a sum of ROA and equity to total assets ratio scaled by the standard deviation of ROA. A higher value of Z-score indicates a higher resistance to shocks and implies a lower credit risk.
CDSt = CDSt-1 +
N w*j,t ij,t j=1
bsk,t qROAt + EFFt + e + t,
(6)
t~N(0,). The expression Pt + C(A*t)/N w*j,t ij,t + F*t denotes a ratio j=1 usually called Efficiency Ratio (EFF) operating expenses over revenues. Further, let iN be the rate of return from trading. Securities designated for trading are meant to generate short-term income and are purchased with the intention of a quick resale at a profit (the fair value of traded securities is recognized each quarter). Consequently, assuming that the return from trading is more volatile than the other returns, the rate of return from trading relative to the rest of the portfolio would proxy the trading risk surcharge over the rest of the portfolio [t = (wN iN/N w*j,t ij,t + F*t)]. Thus: j=1 CDSt = CDSt-1 + N-1j=1 w*j,t ij,t + t bsk,t qROAt + EFFt + c + t. (7) where c = e . Equation (7) represents a portfolio theory-based structural determination of the CDS. However, there might be additional variables to test for their significance in determining CDS. One that is most directly 66 related to the portfolio theory is that investors speculate about the future
The quality of banks assets is measured by three ratios: loan-loss provisions to total loans, non-performing loans to total loans, and

the loan-loss reserves to non-performing loans. While the loan-loss provisions ratio measures the credit risk from a banks portfolio generated within one year, the share of non-performing loans represents the stock of non-performing loans in the loan book. Both indicators should correlate positively with the credit risk, since higher newly expected losses and a higher stock of non-performing loans increase banks vulnerability and default risk. Loan-loss reserves serve to cover non-performing loans and thus their ratio to non-performing loans should, other things being equal, correlate negatively with the CDS spread.

default risk. A higher ROE or ROA indicates better profit prospects for growth and resilience to shocks, and thus should be associated with lower credit risk. A banks liquidity/funding position is measured by four ratios: loan to deposit ratio, share of wholesale funds to total liabilities, shortterm borrowing to total liabilities, and liquid assets to total assets. The higher the loan to deposit ratio, the higher the dependence on non-deposit funding. Since deposits are viewed to be more stable and usually cheaper sources of funding, greater dependence on wholesale funding may signal higher funding risk and higher riskiness for a bank. The share of wholesale funds to total liabilities is an alternative measure of banks reliance on wholesale funding. Short-term borrowing to total liabilities and liquid assets to total assets measure the degree to which banks can withstand a sudden liquidity distress. A bank with a higher share of short-term borrowing would be more vulnerable in the event of a bank run (hence a higher CDS spread). A bank with a higher share of liquid assets would prove more resilient to liquidity pressures.
The quality of management is assessed based on management flexibility, long-term issuer default rating, and differences in the business models. Management quality (approximated by cost efficiency scores) has been associated with bank failures in a number of recent studies [Barr and Siems (1994), Wheelock and Wilson (2000), and Kick and Koetter (2007)]. Cost efficiency is approximated by a simple ratio of operating expenses to total revenues, denoted as the efficiency ratio, which measures management flexibility to adjust costs to changes in the business development signaled by revenues. The higher the efficiency ratio, the higher is the default risk.
Market risk is represented by both bank-specific and common market factors. The historical variance of a banks equity price, volatility implied by the options prices, cost of funds, expected default frequency (EDF), and effective GDP growth and GDP volatility affecting a bank are measured at the bank level, while the slope of the yield curve along with the actual and expected GDP growth and GDP volatility are collected at the country level. Bloomberg Euro 500 stocks index and VSTOXX were used as common market factors.
Long-term issuer default rating contains judgment on all the risks stemming from a banks business. Produced by Fitch Ratings,3 this composite rating takes into consideration a number of external and internal qualitative and quantitative factors risks pertaining to credit, market, operational, funding, liquidity, banks structure, business diversification, management and strategy, and capitalization. On the numerical scale, a higher rating value means a higher credit default risk.
Both the variance of equity price and the options implied volatility (bank specific factors) contain distinct information about the CDS spread. The variance of equity price approximates the variance of a firms value in the structural pricing theory of credit derivatives. Firmspecific equity volatility is found to be an important determinant of bond credit spreads [Campbell and Taksler (2003)]. It is computed as the historical variance of a banks equity price in a particular year. In addition, Cremers et al. (2004) argue that besides the historical equity price volatility, the options implied volatility contains additional information for pricing credit risk. The implied volatility was extracted from the equity options with one year maturity using Bloomberg. Since volatility essentially measures the uncertainty of return to investment in the banks equity, it should positively correlate with credit default risk.
Trading income as apercentage of total revenues accounts for the differences in banks business models. Investment banking-oriented banks would report a high portion of revenues from trading income, while banks with dominantly commercial banking business would have a lower share. Since the trading income is more volatile than interest income, hence more risky, banks with a higher ratio may be expected to pay a relatively higher default risk premium.
Earnings potential is represented by three profitability measures: net interest income ratio, return on equity (ROE), and return on assets (ROA). The net interest income ratio is calculated as thepercentage interest yield of interest bearing assets; ROE (ROA) is computed as net income divided by total common equity (total assets).
Moodys KMV EDF (a bank specific factor) is a widely used marketbased statistics for credit default risks. The underlying information in the EDF (five years) is the market value and volatility of equity and the book value of liabilities the Mertons type statistics. A tight relation between equity and CDS spread is expected since these two markets are linked through capital structure arbitrage. A higher EDF in general
The net interest income ratio is a rough measure for the lending margin charged by a particular bank a risk taking measure. Since loans are priced according to their risk score, a higher lending margin may signal higher risk-taking. Consequently, assuming negligible differences in funding costs across banks, a higher net interest income ratio would imply higher default risk.
The ROE (ROA) measures the profit a bank can generate given shareholders capital (total assets), hence should correlate negatively with
The rating is not available for DnB NOR ASA. No change in rating was assumed in estimation.
67
implies a higher CDS spread.
GDP growth of the country in which a particular bank is incorporated, as well as the effective expected GDP growth of countries to which the bank has exposure (weights given by the shares in a given banks revenue). Forecast for the upcoming year was extracted from the December surveys by the Consensus Forecasts (CFs); the use of CFs comes naturally as the respondents to the CF surveys are mainly large banks and thus the expectations are their own.
Cost of funds (a bank specific factor) is approximated by the ratio of interest expense to total liabilities and signals banks liquidity and risk-taking characteristics. An increasing cost of funds means that a bank is willing to pay extra premium on funds. It does so usually in connection with problems in managing liquidity. Persistently higher cost of funds relative to its peers, other things being equal, suggests that a bank takes higher risk in its loan book to pay higher cost of funds. Consequently, cost of funds would correlate positively with a banks default risk.
Finally, the overall current economic climate measured by the stock market index and stock market volatility (common market factors) might play a role in the CDS market as well. Due to known time lags between economic slowdown (acceleration) and credit portfolio deterioration (improvements), the performance of the stock market leads the expectations about shocks to banks from the economic climate and affects their price of default risk. The stock market index (Bloomberg Euro 500 stocks index) and the European stock market volatility (VSTOXX) would correlate with the credit default premium negatively and positively, respectively.
The slope of the yield curve (a common market factor) reflects growth prospects of the economy and expected future short-term interest rates. A steepening of the yield curve signals positive prospects for economic growth (declining non-performing loans and improving recovery rate) and thus higher future short-term interest rates. A steepening of the yield curve (future improvement in economic conditions) should hence correlate negatively with credit default risk. The slope of the yield curve was derived here from the return on 10-year government bonds and the overnight interbank market rate. While the choice of the long-term rate is straightforward, the choice of risk free short-term rate is not. A usual choice in the literature is the three-month money market rate or the treasury bills rate. However, during the crisis years (2007 and 2008) the three-month rate was plagued by credit and liquidity risks.4 In order to filter out the credit and liquidity risks, the three-month overnight interest rate swaps (O/N IRS) would be the best choice, since only the interest is traded (no risk of collateral), though an O/N IRS is not available for all countries in the sample. Nevertheless, since there is very little difference between the overnight rate and three-month O/N IRS, the risk free short-term interest rate was approximated by the overnight rate. Banks were assigned the slope of the yield curve according to the country where a particular bank is incorporated.
Estimation method
The following dynamic specification of the CDS spread accounts for its potential persistence and for bank-specific credit risks. Formally: CDSi,t = CDSi,t-1 + Xi,tb + Zt + ui,t , (9)
where, CDSi represents the CDS spread value for bank i at time t; CDSi,t-1 denotes the one-period lagged value of CDS spread for bank i, and measures the degree of persistence. Further, Xi,t stands for a vector of explanatory variables observed for bank i in period t and b contains corresponding sensitivities to particular explanatory variables. Banks credit risk factors, summarized in Zt, represent banks common determinants of CDS spread (at the sample and country level) and contains the associated parameters. The composite error term ui,t is composed of a white noise component i,t and a bank-specific credit risk factor i. The model parameters are estimated using Arellano-Bond GMM method [Arellano and Bond (1991)]. There are several reasons for this choice. First, some explanatory variables such as ratings or variance of equity are endogenous to the CDS spread and need to be instrumented accordingly. Second, the bank-specific time invariant credit risk factors might be correlated with other explanatory variables which are accounted for
The economic climate banks face is measured by two indicators: (i) the GDP of the country where a bank is incorporated and (ii) the bank-specific GDP growth and its volatility (hence a common market and bank-specific factor, respectively). The first is given by the GDP growth and its volatility based on the banks country of incorporation.
5
For the second, we compute a weighted average of GDP growth of countries to which the bank has exposure, and its volatility, for each bank, where the weights are derived from the share in the banks revenue of countries to which the bank has exposure. GDP growth
6
is negatively correlated with the share of non-performing loans and positively with the recovery rate. On the other hand, the volatility of GDP means uncertainty in earnings. Consequently, a higher GDP growth (volatility) is expected to correlate negatively (positively) with the default risk.
The economic outlook (as both a common market factor and a bankspecific factor) was approximated by the one-year ahead expected
68
In well functioning markets, the spread between three-month and overnight rate is constant and minimal (up to 10 basis points). During the crisis, the spread increased significantly (to 70 basis points) as the three-month market had become illiquid. Following Tang and Yan (2010), the GDP volatility is estimated using the unexpected GDP growth rate. In particular, the following AR(1) process is estimated using quarterly GDP growth rates (gdp) for each country i: gdpi,t = i + igdpi,t-1+i,t. The GDP volatility in time t is equal to (/2)0.5| t |. Yearly data is obtained by averaging quarterly observations. Tang and Yan (2010) find that bank-level determinants are more informative than common macroeconomic variables.

by first differencing. Third, the presence of lagged dependent variable introduces autocorrelation in residuals and hence it is instrumented with its lagged value. Fourth, financial market variables often exhibit a random walk process and thus ought to be instrumented by differenced lagged values (hence Arellano-Bond method). Finally, the Arellano-Bond method is designed for samples with T/N 0. It is, therefore, the appropriate estimator since the panel dataset has a short time dimension (T=3) relative to the number of banks (N=29). The Arellano-Bond GMM dynamic panel data estimator is applied to firstdifference of the equation (9), which removes the fixed effects: CDSi,t = CDSi,t-1 +Xi,tb + Zt + i,t . (10)
availability limitations lead to a sample of 29 large banks in 12 European countries, spanning a period of five years during2004-2008 a balanced panel data structure. Characteristics of banks in our sample satisfy the emerging definition of systemically important banks. To date, there is no commonly agreed definition yet, given the difficulties in defining systemic importance [Huertas (2009)]. Size is an important but generally not sufficient condition for systemic importance [Marsh (2003)]. In general, there is a broad agreement that banks that are large (significant in size and activities) and complex (interconnected with other institutions and markets, or have significant cross-country exposures) can pose significant systemic risk. Recent work for the G-20 Group [IMF (2009)] set three key criteria for systemically important institutions and markets along these lines: size, interconnectedness, and substitutability. The first three categories of Thomson (2009) classification of systemically important financial institutions refer to size, contagion, and correlation. The following banks are included in the sample: Erste Group Bank and Raiffeisen International (Austria); Dexia SA and KBC Groep (Belgium); HSBC Plc, Barclays Plc, Royal Bank of Scotland, Lloyds, and Standard Charter (UK); Danske Bank (Denmark); BNP Paribas, Societe Generale, Credit Agricole, and Natixis (France); Deutsche Bank AG and Commerzbank (Germany); Bank of Ireland, Anglo-Irish Bank, and Allied-Irish Bank (Ireland); Unicredit SPA, Intesa Sanpaolo, and Banca Monte dei Paschi di Siena (Italy); DnB NOR ASA (Norway); Banco Santander and BBVA (Spain); Nordea Bank AB and Svenska Handelsbanken (Sweden); UBS AG and Credit Suisse (Switzerland). Data is sourced from multiple databases. Banks financial indicators, market prices, and macroeconomic variables are obtained from Bloomberg, Economic Intelligence Unit, Consensus Forecasts, Datastream, Stoxx. com, and Credit Edge. The data coverage of banks financial statements and ratios by Bloomberg for2004-2008 is about 70percent of the needed yearly data. The missing data has been filled directly from banks publically available statements, after a thorough data correspondence check. The descriptive statistics suggest that market indicators detect problems better, as compared to financial soundness indicators. Table1 contains the yearly average summary statistics of the sample. To illustrate, while the CDS spread tripled in 2007, several financial soundness indicators did not show signs of a crisis until2008. In contrast, the set of market related indicators were indicative of an upcoming crisis (in particular, rising implied volatility and the flattening slope of the yield curve). Importantly, banks show diverse credit risk developments amid the crisis. While the average CDS spread increased 11 times during the current crisis (from 9.6 in2006 to 108 in2008), the standard deviation of CDS 69
The actual estimation of equation (10) proceeds with one-step and two-step estimators. In the one-step estimation the i,t are assumed to be i.i.d., hence the (T-2) square matrix H in the weighting matrix W1 = N-1iMiHMi)-1 has twos on the main diagonal and minus ones on the first off-diagonals. Matrix W is used to produce consistent one-step parameter estimates: 1 = (DMW1MD)-1DMW1DCDS, where 1=[ ^, b^, ^], (11)
D= [CDSi,t-1, Xi,t, Zt ] and M is a matrix of
instruments containing the T(T-1)/2 sequential instruments for the endogenous explanatory variables (lagged levels of endogenous variables) and in addition the differenced exogenous variables and other instruments. The one-step parameter estimates are used to derive an optimal weighting matrix W2 = N-1iMii1 i1Mi)-1, where i1 = CDSi Di 1. The equation (11) run with W2 instead of W1 produces the two-step parameter estimates. As long as the estimates in the two steps differ, two-step estimates are preferable since they use an optimal weighting matrix (the covariance matrix is robust to panel specific autocorrelation and heteroscedasticity). The parameters in both steps are asymptotically equivalent if the errors are i.i.d. [Arellano and Bond(1991)]. Nevertheless, the two-step standard errors are found to be downward biased in small samples [Windmeijer (2005)], whereas the asymptotic standard errors in the one-step estimator tend to be unbiased. Since the sample in this paper is relatively small, the statistical inference in the two-step estimator is based on feasible estimator of the finite sample correction for the linear efficient two-step estimator, derived by Windmeijer (2005).
Data sample
The sample of large European banks consists of systemically important banks in Europe. They are either large in their domestic markets (local market interconnections represent a high local systemic risk, with their assets as a percent of domestic claims exceeding 10 percent), and/or they have large cross-border exposures (the share of foreign revenues in the banks total revenues exceeds 30percent). These criteria and data
2004 Dependent variable Credit Default Swap (CDS) (basis points) Standard deviation of CDS Capital adequacy Tier I ratio Tier II ratio Leverage (multiple of equity) Z-score Asset quality Ratio of loan-loss provisions to total loans Share of non-performing loans in total loans Loan-loss reserves ratio Management quality Efficiency ratio (ratio of operating costs to revenues) Fitch long-term issuer default rating (numerical scale) Trading income as percent of revenues Earnings potential Net interest income (percent of average earning assets) ROA (return on assets) ROE (return on common equity) Liquidity Loans to deposits ratio Short-term borrowing to total liabilities Wholesale funds to total liabilities Liquid assets to total assets Market risk Options implied variance of equity Slope of the yield curve Variance of banks equity price Cost of funds GDP growth Volatility of GDP growth Weighted GDP growth Volatility of weighted GDP growth Expected GDP growth Weighted expected GDP growth Moodys KMV EDF5 VSTOXX Bloomberg Euro 500 Stocks Index (index number) 13.8 3.9 8.3 3.4 25.8 3.5 0.34 1.88 114 61.6 4.1 13.98 1.78 0.7 16.32 128 22 59 13 16.8 1.85 99 2.4 2.5 0.41 3.9 0.3 2.3 2.9 0.09 18.9 175
2005 12.3 3.1 8.1 3.3 27.1 3.5 0.29 1.8 126 58.9 4.13 18.61 1.62 0.7 18.1 134 22 62 11.9 18.3 1.09 410 2.5 2.8 0.55 3.5 0.54 2.4 2.9 0.07 14.1 202
2006 9.6 2.7 8.3 3.2 27.2 3.8 0.33 1.75 106 56.2 3.83 19.38 1.52 0.8 19.86 138 21 63 11.3 21.6 0.83 211 2.9 3.1 0.45 4.2 0.43 2.4 3 0.04 16.5 240
2007 27.3 17.5 7.9 3 27.8 3.7 0.39 1.68 102 60.6 3.86 12.98 1.49 0.7 16.49 139 21 63 10.4 25.6 0.29 497 3.5 3.1 0.5 3.97 0.35 2.5 3 0.04 19.7 270
2008 108 47.4 8.4 3.3 31.2 1.9 0.85 2.3 95 104.4 4.28 -41.24 1.65 0.2 0.19 150 20 65 9.8 52.5 0.26 3345 3.3 -1.9 0.15 0.15 0.12 2.2 2.9 0.6 33.7 202
spread across banks increased by a multiple of 18 (Table1). Consequently, the structure of data variability is biased toward cross-sectional differences in banks characteristics.
Estimation results
Baseline model
The testable hypothesis of whether pricing of banks CDS spreads is based on portfolio theory is expressed in equation (8). The baseline unrestricted regression, therefore, contains this specification. However, since the specification covers only three categories of CAMELS, that is, management quality, earnings potential, and market risk, the baseline unrestricted specification contains also the most often used ratios representing the remaining categories in CAMELS. Estimation results for the baseline model are reported in Table 2 for two types of estimates: using the one- and the two-step estimator, respectively. While the one-step estimator assumes i.i.d. errors, the two-step estimator takes into account the actual structure of the residuals. Consequently, as long as there is a difference between the two estimators, the two-step estimates with finite sample corrected standard errors are preferred. The two-step estimator is furnished with two standard errors in the parentheses. The first pertains to the efficient asymptotic standard error, while the second is derived using the finite sample corrected variance by Windmeijer (2005). As Table2 suggests, the finite sample corrected standard errors exceed the asymptotic ones and tend to exceed even those of the one-step estimator, which indicates the importance of small sample bias correction [a similar result as in simulations by Windmeijer (2005)]. The instruments are valid and the estimation does not suffer from serial correlation in residuals. The validity of instruments is tested using Hansen test of over-identifying restrictions a preferred statistic for estimation with robust standard errors. The test does not reject the validity of instruments on the 26percent significance level. Further, as expected, the presence of the first order serial correlation is not rejected, while the second order autocorrelation is (probabilities 0.18 and 0.33 for one- and two-step estimator, respectively). The latter one implies no problems with serial correlation. Only variables predicted by the portfolio theory turned out to be significant. The results confirm the dynamic specification and underline the risk frontier as a pricing vehicle for CDS spreads. In particular, six variables turned out to be significant determinants of the CDS spread: the efficiency ratio, trading income ratio, net interest income ratio, ROA, the implied volatility, and the slope of the yield curve. These variables match those specified in equation (8), that is, confirm that the pricing of banks CDS is based on the portfolio theory. In relation to the literature, the slope of the yield curve and implied volatility from option prices proved to be significant determinants of CDS spreads already in previous studies [Ericsson et al. (2009); Abid and Naifar (2006); Blanco et al. (2005); and
Source: Bloomberg, Banks financial statements, EIU, Credit Edge, stoxx.com, and Datastream.
70
Table 1 Data summary statistics (cross-section average; in percentage)

Unrestricted One-step Lagged dependent variable Capital adequacy Asset quality Credit default swap (t-1) Leverage Loan-loss provisions ratio Share of non-performing loans Management quality Efficiency ratio Trading income ratio Earnings potential Net interest income ratio ROA Liquidity Market risk Loans to deposits ratio Equity options implied volatility Slope of the yield curve Bloomberg Euro 500 Stocks Index Test statistics 1st order autocorrelation 2nd order autocorrelation Hansen test of overid. restrictions Number of observations / banks F-test (for joint insignificance) R2 0.98**(0.41) -0.14 (0.33) 4.22 (15.9) 1.32 (2.39) 0.3***(0.08) 0.35***(0.08) 24.2**(10.2) -22.8*(11.7) 0.08 (0.09) 1.96***(0.34) -11.5***(3.9) 0.01 (0.07) -2.6***(0.01) 1.3 (0.18) 20.2 (0.26) 87 / 29 0.9 Two-step 1.13*(0.16;0.6) -0.26 (0.17;0.4) -0.45 (5.9;21) 0.63 (1.21;3.6) 0.3***(0.05;0.1) 0.33***(0.05;0.1) 23.2**(4.6;11) -21.7*(4;11) 0.11 (0.06;0.13) 1.86***(0.1;0.3) -12**(2.6;5.2) -0.01 (0.02;0.05) -1.64*(0.1) 0.97 (0.33) 20.2 (0.26) 87 /29 0.9 One-step 1.03***(0.25) 0.31***(0.1) 0.36***(0.09) 25.2***(9.3) -21*(10.7) 2.05***(0.3) -12.5***(3.1) -2.04**(0.04) 1.35 (0.18) 23.2 (0.39) 87 / 29 0.03 (0.99) 0.9
Parsimonious Two-step 1.07***(0.1;0.3) 0.3***(0.04;0.1) 0.35***(0.04;0.1) 25.1**(3.5;10.5) -24.6***(2.5;11) 1.92***(0.08;0.3) -11.5***(0.9;2.6) -1.79*(0.07) 0.99 (0.32) 23.2 (0.39) 87 / 29 0.65 (0.66) 0.89
Notes: Standard errors for coefficients and probabilities for test statistics are given in parentheses. The two-step estimates contain both the asymptotic (the first one) and the Windmeijers (2005) finite sample corrected standard errors. Stars denote significance level as follows: * 10%, ** 5%, *** 1%. Inference in two-step estimates is based on the finite sample corrected standard errors. Test statistics in the two-step estimator pertain to the regression with finite sample corrected variance. Instruments contain the GMM-style regressors (second and third lag) of the lagged dependent variable, ROA, GDP growth, expected GDP growth by Consensus Forecasts, and further the lagged differenced leverage, Tier I, Tier II, Loan-loss provisions ratio, the share of non-performing loans, efficiency ratio, the share of trading income in revenues, the slope of the yield curve, net interest income ratio, and loans to deposits ratio.
Table 2 Results: baseline model
Aunon-Nerin et al. (2002)]. On the other hand, other key financial soundness indicators, such as capital adequacy, asset quality, and liquidity turn out to be insignificant. The insignificance of capital adequacy and liquidity likely stems from the very little variation in leverage (the capital measure used in the estimation) and loan-to-deposit ratios (the liquidity measure used in the estimation), respectively, within the banks during the sample period. While the insignificance of the share of nonperforming loans is likely due to its backward-looking nature, the insignificance of the loan-loss provisions show that there is no extra explanatory power over the one expressed in the parameter in front of ROA (loan-loss provisions are, by construction, part of ROA). The model fits the data well. The R2 statistic presented in Table 2 is computed on first differences over the three-year estimation period during2006-2008 and reaches about 89percent. The lagged dependent variable contributes 37percent, while the risk frontier explains 52 percent. A baseline parsimonious model has been derived by dropping insignificant variables in the unrestricted model. According to the performed
F-test, all insignificant variables in the unrestricted baseline specification can be excluded from the model. The probability of not rejecting the joint restriction is 0.99 for the one-step estimator and 0.66 for the two-step estimator. The results in the parsimonious baseline model remain robust to the restrictions and are used in the robustness analyses, which follow in the next two sections.
Robustness to alternative CAMELS indicators

The parsimonious baseline specification showed robustness to a wide range of alternative financial soundness indicators and time dummy for the crisis year 2008. For the same categories of financial soundness indicators, nine alternative variables across the CAMELS structure were used: Tier I, Tier II, Z-score (for capital adequacy), loan-loss reserves (for asset quality), long-term issuer default rating (for management quality), ROE (for earnings potential), and short-term borrowing to total liabilities, wholesale funds to total liabilities, and the ratio of liquid assets to total assets (for liquidity). In addition, the CDS market in 2008 (especially in the second half) was possibly driven by factors outside the baseline model and thus a dummy for that year was tested (A10). The baseline model specification proved robust to all alternative CAMELS indicators as well as to the dummy variable for 2008. 71
The diagnostic estimations confirm validity of instruments and no autocorrelation problems in all ten regressions. Table3 displays the results for the two-step estimator with Windmeijers (2005) standard errors in parentheses. The Hansen test of over-identifying restrictions never rejects the validity of instruments at conventional significance levels. The set of
instruments has been adjusted in several regressions. The adjustments are listed in the notes to the Table3. The tests for serial correlation show a significant first order and insignificant second order autocorrelation. The alternative measures of capital adequacy are statistically insignificant
A1 Lagged dependent variable Credit Default Swap (t-1) Capital adequacy Tier I Tier II Z-score Asset quality Loan-loss reserves ratio Management quality Efficiency ratio Trading income ratio Long-term issuer default rating Earnings potential Net interest income ratio ROA ROE Liquidity Short-term borrowing to total liabilities Wholesale funds to total liabilities Liquid assets to total assets Market risk Options implied variance of equity Slope of the yield curve Dummy (year 2008) Test statistics 1st order autocorrelation 2nd order autocorrelation Hansen test of overid. restrictions Number of observations / banks R2 0.9***(0.3) -1.4 (2.8) 0.3**(0.1) 0.3***(0.1) 23*(12.5) -22.8*(13) 2.1***(0.4) -10.7***(3) -1.7*(0.09) 0.96 (0.33) 23.1 (0.34) 87 / 29 0.89
A2 1.2***(0.3) -3.9 (5.2) 0.3**(0.1) 0.3***(0.1) 21.5**(10) -22.3**(11) 1.9***(0.4) -11.5***(3) -1.8*(0.07) 0.96 (0.34) 24.7 (0.26) 87 / 29 0.9
A3 1 1.0***(0.2) -0.87 (2.4) 0.3***(0.1) 0.4***(0.1) 20.3*(10.8) -30.7***(10) 1.8***(22) -10.3***(2) -1.8*(0.07) 1.1 (0.26) 17.9 (0.65) 87 / 29 0.9
A4 2 1.0***(0.3) 0.03 (0.04) 0.3***(0.08) 0.3***(0.08) 26.3***(9.9) -20.4*(11.4) 2.04***(0.3) -11.9***(3) -1.7*(0.09) 0.8 (0.43) 22.8 (0.36) 87 / 29 0.9
A5 1.1***(0.3) 0.3**(0.12) 0.4***(0.1) -2.7 (7.6) 24.6**(12) -25.8**(12) 2***(0.3) -11.6***(3) -1.8*(0.07) 1.27 (0.2) 23.2 (0.33) 87 / 29 0.89
A6 1.1***(0.3) 0.3***(0.1) 0.4***(0.1) 24.4**(11) -34.6**(16) 0.45 (0.36) 1.9***(0.3) -12***(3) -1.9*(0.06) 1.44 (0.15) 22.7 (0.36) 87 / 29 0.89
A7 3 0.9***(0.2) 0.3***(0.1) 0.36***(0.1) 21**(10.1) -31.1***(11) 0.45*(0.23) 1.95***(0.2) -10***(2.1) -1.66*(0.1) 0.66 (0.51) 18.1 (0.64) 87 / 29 0.88
A8 4 1.0***(0.2) 0.3**(0.12) 0.3***(0.1) 19.7*(12) -29.4**(12) 0.06 (0.39) 1.9***(0.4) -10***(3.1) -1.8*(0.07) 1.05 (0.29) 17.6 (0.67) 87 / 29 0.88
A9 5 1.0***(0.2) 0.3***(0.1) 0.3***(0.1) 20.1*(10) -34***(10.1) 0.26 (0.66) 1.8***(0.3) -10.6***(3) -1.8*(0.07) 0.85 (0.4) 17.9 (0.66) 87 / 29 0.88
A10
1.1***(0.3) 0.3***(0.1) 0.35***(0.1) 25.9**(11.7) -23.1**(11.2) 1.9***(0.4) -11.7***(2.8) 0.6 (10.5) -1.8*(0.08) 1.0 (0.3) 22.7 (0.4) 87 / 29 0.89
Notes: Windmeijers (2005) finite sample corrected standard errors for coefficients and probabilities for test statistics are given in parentheses. Stars denote significance level as follows: * 10%, ** 5%, *** 1%. If not specified otherwise, instruments contain the GMM-style regressors (second and third lag) of the lagged dependent variable, ROA, GDP growth, GDP forecast by Consensus Forecasts, and further the lagged differenced Leverage, Tier I, Tier II, loan-loss provisions ratio, the share of non-performing loans, efficiency ratio, the share of trading income in revenues, the slope of the yield curve, net interest income ratio, and loans to deposits ratio. 1/ Liquid assets over total assets replaced the loans over deposits instrument. 2/ Loan-loss reserves to non-performing loans replaced the share of non-performing loans. 3/ Short-term borrowing to total liabilities replaced loans to deposits ratio. 4/ Wholesale funds to total liabilities replaced loans to deposits ratio. 5/ Liquid assets to total assets replaced the loans to deposits ratio.
72
Table 3 Results: two-step estimates of the baseline with alternative CAMELS indicators

and point to a homogeneity of banks in capitalization. The regression results (A1-A3) show the effects of the alternative capital adequacy specifications the Tier I, Tier II, and Z-score, respectively. None of the three indicators adds additional explanatory power to the parsimonious baseline specification. Their insignificance is likely caused by high sample homogeneity in capital adequacy measures that remain rather stable during the initial crisis years.
Moreover, the alternative measures of asset quality, management quality, and earnings potential also do not add value to the baseline model. The regression results (A4-A6) test the significance of loan-loss reserves ratio, long-term issuer default rating, and ROE, respectively. The insignificance of the loan-loss reserves ratio and the long-term issuer default rating is again most likely caused by the banks homogeneity and stability of reserves and the rating during the outbreak of the crisis. In the case of
A11 Lagged dependent variable Credit default swap (t-1) Management quality Efficiency ratio Trading income ratio Earnings potential Net interest income ratio ROA Market risk Options implied variance of equity Slope of the yield curve Variance of banks equity price Cost of funds GDP growth Volatility of GDP growth Weighted GDP growth Volatility of weighted GDP growth Expected GDP growth Weighted expected GDP growth Moodys KMV EDF5 VSTOXX Test statistics 1st order autocorrelation 2nd order autocorrelation Hansen test of overid. restrictions Number of observations / Banks R2 0.99***(0.2) 0.29***(0.1) 0.33***(0.1) 20.6*(10.3) -31.7***(9.7) 1.8***(0.25) -10.4***(1.9) .0002 (.0001) -1.9*(0.06) 1.25 (0.2) 18.1 (0.64) 87 / 29 0.88
A12 1.09***(0.3) 0.31**(0.14) 0.35***(0.1) 25.5*(14.9) -23.8**(11.1) 1.92***(0.3) -11.3**(4.7) 30.2 (464) -1.79*(0.07) 0.82 (0.42) 23.6 (0.32) 87 / 29 0.89
A13 1 0.9***(0.3) 0.26**(0.1) 0.33***(0.1) 24.8**(10) -36.1***(13) 1.8***(0.3) -11.8***(3.3) 0.007 (0.8) -1.7*(0.09) 0.56 (0.58) 19.6 (0.55) 87 / 29 0.87
A14 2 1.15**(0.5) 0.29***(0.1) 0.3***(0.1) 23.7**(10.3) -31.2***(10) 1.8***(0.6) -11.8***(1.2) 3.1 (4.4) -1.62*(0.1) 0.84 (0.4) 20.7 (0.48) 87 / 29 0.88
A15 3 1.2***(0.3) 0.28*(0.1) 0.3*(0.16) 18.2*(9.6) -21 (14) 1.7***(0.3) -10.6***(3.1) -1.4 (2.5) -1.67*(0.1) 1.29 (0.2) 20.2 (0.51) 87 / 29 0.9
A16 4 0.99***(0.3) 0.3***(0.1) 0.34***(0.1) 21.3*(12) -29.7***(11) 1.9***(0.3) -11.3***(2.2) 4.8*(2.4) -1.75*(0.08) 0.88 (0.38) 19.86 (0.53) 87 / 29 0.89
A17 5 1.17***(0.3) 0.28***(0.1) 0.33***(0.1) 22.1*(12) -28***(6.4) 1.9***(0.22) -7.46**(3.4) 10.4 (10) -1.76*(0.08) 1.11 (0.27) 15.35 (0.81) 87 / 29 0.88
A18 5 1.4***(0.4) 0.24**(0.1) 0.3**(0.1) 21**(8.9) -29.4***(7) 1.7***(0.3) -9.9***(2.2) 15.4 (10) -1.8*(0.07) 0.82 (0.41) 15.9 (0.77) 87 / 29 0.87
A19 4,6 1.17***(0.3) 0.3**(0.11) 0.3***(0.1) 17.4*(9.4) -29.4***(8) 1.7***(0.4) -11***(3.3) 5.2***(2) -1.52 (0.13) 0.56 (0.58) 19.9 (0.53) 87 / 29 0.9
A20 1.1***(0.3) 0.3**(0.11) 0.34**(0.1) 23.3*(13) -24.1**(10.5) 1.8***(0.6) -10.3*(5.7) 0.3 (1.1) -1.9*(0.06) 0.99 (0.33) 23.7 (0.3) 87 / 29 0.89
Notes: Windmeijers (2005) finite sample corrected standard errors for coefficients and probabilities for test statistics are given in parentheses. Stars denote significance level as follows: * 10%, ** 5%, *** 1%. If not specified otherwise, instruments contain the GMM-style regressors (second and third lag) of the lagged dependent variable, ROA, GDP growth, GDP forecast by Consensus Forecasts and further the lagged differenced Leverage, Tier I, Tier II, loan-loss provisions ratio, the share of non-performing loans, efficiency ratio, the share of trading income in revenues, the slope of the yield curve, net interest income ratio, and loans to deposits ratio. 1/ Weighted GDP growth replaced GDP growth instrument. 2/ Variance of GDP growth replaced the loans to deposits ratio. 3/ Volatility of weighted GDP growth replaced the expected GDP growth. In addition, the ratio of Liquid assets over total assets replaced the Loans over deposits instrument. 4/ Volatility of weighted GDP growth replaced the Loans to Deposits Ratio. 5/ Liquid assets to total assets replaced the loans to deposits ratio. 6/ EDF replaced the GDP growth instrument.
Table 4 Results: two-step estimates of the baseline with alternative market risk indicators
73
ROE, its insignificance is a result of its subordinate information content compared to ROA. Alternative liquidity measures show mixed results. The results for the baseline regression containing short-term borrowing to total liabilities, wholesale funds to total liabilities, and liquid assets in total assets, respectively are contents of the columns A7-A9. While the share of liquid assets in total assets and the share of wholesale funds in total liabilities turn out to be insignificant, a higher share of short-term borrowing in total liabilities is associated with a higher CDS spread and is marginally statistically significant. However, its contribution to the baselines overall explanatory power is negligible.
the country of banks incorporation. The GDP growth of the parent banks country and its volatility are statistically insignificant (regressions A13 and A14). While it is also insignificant, weighted GDP growth has a negative sign, which is in line with theory (regression A15). In addition, the volatility of GDP growth is marginally significant and positive (i.e., a percentage increase in volatility implies an increase in CDS spread by 4.8percent, regression A16). Nevertheless, also in this case, the explanatory power added by this variable to the baseline model is negligible. Expected GDP growth and stock market volatility turned out to be uninformative (regressions A17, A18, and A20). Even though banks, in their operations, depend, to a significant extent, on the accuracy of GDP forecasts, the forecasts (both expected GDP growth and weighted (effective) expected GDP growth) turn out to be insignificant in explaining CDS spreads. Similarly, European stock market volatility (VSTOXX) was insignificant, likely being dominated by highly significant bank-specific implied volatility from options.
Robustness to alternative market risk indicators

The parsimonious baseline specification is fairly robust, and explains a dominant portion of CDS spread variability under alternative market risk indicators. The baseline was tested against the following nine market risk variables: the historical variance of banks equity price, Moodys KMV EDF5, the cost of funds, actual and expected GDP growth, actual and expected GDP growth weighted by banks revenues, the volatility of GDP growth and of weighted GDP growth. Most of the variables turned out to be insignificant and those that were significant added a negligible explanatory power to the baseline model. The diagnostics of the two-step estimator, presented in Table4, confirm robust estimation results in all nine regressions. The validity of instruments has not been rejected by Hansen test of over-identifying restrictions at conventional significance levels. The second order autocorrelation is rejected in all nine regressions. The absence of the first order autocorrelation in the regression A19 is only marginal and is secondary to the importance of the second order autocorrelation test. Most alternative market risk indicators turn out to be statistically insignificant. The historical variance of banks equity price is insignificant, while Moodys KMV EDF5 correlates positively with the CDS spread. The historical variance of equity (regression A11) appears to carry lower information content than the implied volatility from options, which is statistically significant in the baseline model. On the other hand, apercentage increase in EDF5 (combining the equity price and its volatility the Mertons type statistics) leads to a rise in the CDS spread by 5.2percent (regression A19), but the contribution to the coefficient of determination is negligible. The cost of funds turns out to be not a significant yardstick in credit risk for the banks, suggesting that the banks form a group of homogenous entities that have similar access to funds (regression A12). 74 Weighted GDP growth and its volatility show better results than those of
Conclusion
This paper claims that the pricing of Credit Default Swaps (CDS) spreads for banks is based on a portfolio theory. Banks are viewed as leverage portfolios and thus their CDS spreads could be priced by risk-neutral investors according to a risk frontier (risk and return on their portfolio). The hypothesis is tested and confirmed, with extensive robustness checks, on a sample of 29 large European banks during 2004-2008. The choice of large European banks is motivated by the fact that CDS contracts for European banks are homogenous and five-year CDS spreads are liquid. The results confirm that CDS spreads for large European banks are indeed priced according to a risk frontier as suggested by the formulated hypothesis. In relation to the Mertons structural approach to the credit risk, variables implied by the portfolio theory explain most of the variability in CDS spreads. The EDF, constructed by KMV Moodys, is derived according to the structural approach to credit risk and thus is used as a proxy for Mertons model. However, the EDF adds only a very little explanatory power to the model based on the portfolio approach. These findings suggest that the credit spread puzzle for banking industry could be potentially addressed by considering banking industry-specific credit risk model based on portfolio theory.

References

Abid, F. and N. Naifar,2006, The determinants of credit default swap rates: an explanatory study, International Journal of Theoretical and Applied Finance, 9, 23-42 Arellano, M. and S. Bond,1991, Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, Review of Economic Studies, 58, 277-97 Aunon-Nerin, D., D. Cossin, T. Hricko, and Z. Huang,2002, Exploring for the determinants of credit risk in credit default swap transaction data: is fixed-income markets information sufficient to evaluate credit risk? Research Paper Series (HEC-University of Lausanne and FAME) Barr, R. S. and T.E. Siems,1994, Predicting bank failure using DEA to quantify management quality, Financial Industry Studies Working Paper (Federal Reserve Bank of Dallas) Benkert, Ch.,2004, Explaining credit default swap premia, Journal of Futures Markets, 24, 71-92 Bierens, H., J-Z Huang, and W. Kong,2003, An econometric model of credit spreads with rebalancing, ARCH and Jump Effects, Working paper (Penn State University) Blanco, R., S. Brennan, and I.W. Marsh, 2005, An empirical analysis of the dynamic relation between investment-grade bonds and credit default swaps, Journal of Finance, 60, 2255-81 Boyd, J. H. and D. E. Runkle,1993, Size and performance of banking firms, Journal of Monetary Economics, 31, 47-67 Campbell, J. T. and G. B. Taksler,2003, Equity volatility and corporate bond yields, Journal of Finance, 58, 2321-49 Collin-Dufresne, P., R. S. Goldstein, and J. S. Martin, 2001, The determinants of credit spread changes, Journal of Finance, 56, 2177-207 Consensus Economics,2003-2007, Consensus forecasts, Asia Pacific consensus forecasts, consensus forecasts Eastern Europe, Latin American consensus forecasts, November and December Issues Cremers, M., J. Driessen, P.J. Maenhout, and D. Weinbaum,2004, Individual stock-option prices and credit spreads, Working Paper (Yale ICF) Driessen, J., 2005, Is default event risk priced in corporate bonds? Review of Financial Studies, 18, 165-95 Duffee, G. R.,1998, The relation between Treasury yields and corporate bond yield spreads, Journal of Finance, 53, 2225-41 Ericsson, J., K. Jacobs, and R. Oviedo,2009, The determinants of credit default swap premia, Journal of Financial and Quantitative Analysis, 44, 109-32 Fabozzi, F. J., X. Cheng, and R. R. Chen,2007, Exploring the components of credit risk in credit default swaps, Finance Research Letters, 4, 10-18 Huertas, T. F.,2009, Too big to fail, too complex to contemplate: what to do about systemically important firms, Manuscript (Financial Markets Group Conference, London) International Monetary Fund,2009, Guidance to assess the systemic importance of financial institutions, markets and instruments: initial considerations, background paper (A report to the G-20 Finance Ministers and Central Bank Governors) Kick, T. and M. Koetter, 2007, Slippery slopes of stress: ordered failure events in German banking, Journal of Financial Stability, 3, 132-48 Markowitz, H., 1952, Portfolio selection, Journal of Finance, 7, 77-91 Marsh, I. W. and I. Stevens, 2003, Large complex financial institutions: common influences on asset price behavior? Financial Stability Review, December, 91-101 Merton, R. C.,1974, On the pricing of corporate debt: the risk structure of interest rates, Journal of Finance, 29, 449-70 Pedrosa, M. and R. Roll,1998, Systematic risk in corporate bond credit spreads, Journal of Fixed Income, 8, 7-26 Thomson, J. B.,2009, On systematically important financial institutions and progressive system mitigation, Policy Discussion Paper, No.27 (Federal Reserve Bank of Cleveland) Wheelock, D. C. and P. W. Wilson,2000, Why do banks disappear? The determinants of U.S.bank failures and acquisitions, Review of Economics and Statistics, 82, 127-3 Windmeijer, F., 2005, A finite sample correction for the variance of linear efficient two-step GMM estimators, Journal of Econometrics, 126, 25-51
75
76
PART 1
Investing in Private Equity Capital Commitment Considerations

Sameer Jain Head of Investment Content and Strategy Alternative Investments,
UBS Financial Services, Inc.
Abstract
This paper explores capital commitment and cash-flow management issues in private equity fund investing. It provides a theoretical framework to structure private equity capital commitment issues in a formal manner, and defines variables, inter-relationships, and boundaries in such a way that the problem can be worked upon. The papers findings suggest that achieving a targeted level of allocation to private equity is a function of the pace of capital deployment as well as dependent upon the desired amount of targeted exposure. It is also dependent on the spread of realized returns in private equity versus other asset classes, as well as on timing and realization periods for capital already invested. 77
Commitment issues
Private equity fundraising and deployment is referred to as making commitments because not all funding is made available immediately to the fund vehicle. Rather, funds are called up as projects covered by the private equity vehicles mandate become available. When committing to a private equity fund, the commitment is typically to provide cash to the fund on short notice from the general partner. General partners of a fund draw down capital from the limited partners as and when they make investments. General partners call down capital only as they require it, rather than in pre-set amounts according to a rigid timetable. If an investor fails to fund a capital call from a fund when due, the fund may exercise various remedies with respect to such investors to forfeit, or sell, all or a portion, of its interest in the fund or require that the investor immediately pay up the full amount of their remaining capital commitment. Investors are typically required to fund only a small percentage of their total capital commitment at the outset. This initial funding is followed by subsequent capital draw-downs (the timing and size of which are generally made known to the investor a few days in advance), as needed by the fund to make new investments. Just-in-time draw-downs are used to minimize the amount of time that a fund holds uninvested cash, which is a drag on fund performance. All of this introduces uncertainty in the cash flows of private equity investments. The unpredictability of cash flows applies to both capital calls, which the investor must fulfill at earlier stages, and distributions to the investor at later stages. In both cases the amount and timing of cash flows is at the discretion of the fund manager. Hence, the importance of carefully considering cash flow needs in portfolios and planning for commitments that have already been made.
If private equity, for example, grows faster than other asset classes, then one cannot easily rebalance the risks in ones portfolio necessarily; if it grows more slowly, then there is no issue. What does this mean for the portfolio over the long term? Clearly, one approach is to account for this before it happens and moderate ones investments. But, when it does, should one rebalance into more fixed income while one is over-allocated? Ignore it? This is particularly interesting when one considers that there might be tactical implications about timing of entry into the private space which also must be thoughtfully addressed. Transitional planning a related issue is transition planning in the sense that the commitment strategy will differ between an investor starting at a zero allocation and ramping up to a target allocation versus one who is replacing a piece of a regular program every year. The reason here is that cash outflows from a ramping up program are not the same as a regular, vintage-year diversified program. Contracting costs finally, there are other issues which are clearly important: tax, transaction costs, etc.
Framework for determining commitments and cash-flows

In order to bring rigor to an approach that may help elucidate important considerations in making private equity commitments, within the larger context of a portfolio allocation, we propose a theoretical model here. Unfortunately, there is no escaping the algebra behind this model, but we do hope that our conclusions follow logically from it. We assume a model in which each asset class grows at a deterministic rate, and the timing of private equity draw-down and liquidations is certain. We further assume that an investor would like to maintain a constant percentage allocation to private equity, and needs to choose a commitment strategy to obtain this desired allocation.
Complications
Uncertain cash-flows lead to uncertain commitment strategies it gives rise to questions of what is an appropriate commitment strategy given: (1) uncertainty about capital draw-downs, (2) uncertainty about capital distributions, and (3) meeting target allocations to portfolio allocation over long time periods. Other questions arise such as what does one do with undrawn, committed investments? Investors are required to make a range of decisions: keep it in cash (riskless), allocate it proportionately to the rest of the portfolio, and so on. Naturally this depends a lot on the investment objectives and the characteristics of the existing portfolio. Added risks (i.e., what happens if one puts the uninvested, but committed capital into a public proxy and markets are depressed at the precise time capital is being called?) are considerations worth investigating. Rebalancing is a related issue; clearly there is an asymmetry in the ability to rebalance a portfolio that consists of illiquid investments in private 78 equity and liquid traditional investments such as publicly traded stocks.
Model
We can model the investors portfolio as follows: IC Initial value of investors capital (i.e., total starting portfolio value) TPE Investors target private equity allocation, in percent G Growth factor of private equity investments R Growth factor of optimal portfolio C0 Initial commitment at t = 0 D Years to invest commitment (i.e., 1/draw-down rate) H Investment holding period It New amount invested, or drawn-down, at end of period t Lt Liquidations at time period t. This amount may be reinvested in the fund or returned to investors Pt Total private equity dollars invested at end of period t

Bacheliers Legacy: Applications of Theoretical Physics in Market Price Modeling
Intuitively, in order to maintain a constant steady state allocation to private equity, the growth in commitments over time must equal the growth in the optimal portfolio. This implies that Ct = C0Rt. Furthermore, the disbursements at any time t will equal the new amount invested at time t H, multiplied by the growth factor in private equity. In other words, Lt = It-HGH. Using this terminology, we can model the private equity dollars invested at any point in time as: Pt = GPt-1 + It Lt (1)
P0Rt GP0Rt-1 = C0/D [Rt(RH GH) D-1i=0 R i ]/RH+D-1 P0Rt/Rt [1 G/R] = C0/D [(RH GH) D-1i=0 R i ]/RH+D-1 C0 = P0[G/R 1][DRH+D-1]/[(GH RH) D-1i=0 R i ]
(8) (9) (10)
Equation (10) establishes the initial level of commitment an investor ought to make to private equity assuming that a variety of variables are perfectly predictable (as we have made simplifying assumptions). Despite the constraints we have imposed this is a powerful equation for it provides interesting insights.
Commitment strategy insights

Equation (10) illustrates that the private equity commitment level depends on five factors: Assumptions: G>R>1, H>1, D>1 (2) Rt-(D-1), we can re-write (2) as: We distinguish the sign of the first partial derivative by contemplating (3) equation (10) and find it to be positive. C = P[G/R 1] [DRH+D-1]/[(GH RH) D-1i=0 R i] = P D [(G R)/(GH 1. Target private equity investment, P higher levels of private equity allocations require higher commitment levels.
Each commitment, Ct, is invested uniformly at the end of each year for D years. Consequently, if an investor commits Ct in year t, Ct/D is invested each year between years t and t + D 1. Based on this, we can define It as: It = D-1i=0 Ct-i/D Since Ct-(D-1) is equal to C0 It = (C0Rt-(D-1)/D) D-1i=0 R i Similarly, we can define Lt as: Lt = GHIt-H = (GHC0Rt-H-(D-1)/D)
D-1 i=0
RH)] [(RH+D-1 RH+D-2)/(RD 1)] Ri (4)
dC/dP = D [(G R)/(GH RH)] [(RH+(D-1) RH+(D-2))/(RD 1)] > 0 2. Private equity growth rate, G faster growth in private equity
which may be either reinvested or returned to investors. Substituting (3) and (4) into (1) provides: Pt = GPt-1 + [C0Rt-(D-1)
D-1
reduces required commitments, since the investor can reach the same steady state allocation with less capital. ]/D (5) We distinguish the sign of the first partial derivative by contemplating equation (10) and find it to be negative. (6) For dC/dG, let K = P D [(RH+D-1 RH+D-2))/(RD 1)] > 0 GH RH = (G R) (GH-1 GH-2R + + RH-1) d[(G R)/(GH RH)/dG = [d[1/(GH-1 GH-2R + + RH-1)]]/dG = [(H 1) GH-2 + R(H 2)GH-3 + + RH-2]/(GH-1 + GH-2R + + RH-1)2 < 0 dC/dG = K [d(G R)/(GH RH)]/dG < 0
i=0
Ri
]/D
[GHC0Rt-H-(D-1) D-1i=0 R i
This can be simplified as: Pt = GPt-1 + 1/D(RH GH) C0Rt-H-(D-1) D-1i=0 R i
Given (6), how should the investor manage his commitments in order to reach his desired private equity allocation? Assume that the investor wants to achieve a target private equity allocation, TPE, of his portfolio value. If the portfolio is compounding at a factor R, then the investors desired dollar investment in private equity at time t is: Pt = TPE IC Rt = P0 Rt (7)
Alternatively, we may also use intuition. Substituting (7) into (6) and rearranging the terms yields: (G R)/(GH RH) = 1/(GH-1 + GH-2R + + RH-1) 79
As G increases, each term in the denominator increases, so 1/(GH-1 + GH-2R + + RH-1) decreases and C decreases as well.
This is self-evident from our definition: Lt = It-HGH when combined with Insight #2 i.e., dC/dG, < 0.
dC/dG, < 0 3. Portfolio growth rate, R faster portfolio growth increases required commitments, since the desired allocation in dollar terms increases as the portfolio growth increases. dC/dR is extremely complex in its analytical form so it is helpful to examine this intuitively C = P[(G/R) 1] [(DRH+D-1)/(GH RH)
D-1
Steady state investment and liquidation rates

Using the above model, we can compute the steady state investment and liquidation rates as a function of unfunded commitments and total investments, respectively. More formally, we would like formulas for RIN and RDI, where: It = RIN x Ct-1, and Lt = RDI x Pt-1G
i=0
(11) (12)
Ri ] We know that in period t, a portion of the new commitments from t-1, t-2,, t D will be invested. Consequently, we can restate It as: It = D Ct-i /D i=1 (13)
= P D [RH-1/[(GH-1 + GH-2R + + RH-1)] [RD-1/(1 + R + RD-1)] = P D [1/(GH-1/ RH-1+ GH-2 / RH-2 + + 1)] [1/(1 / RD-1 + 1 / RD-2 + + 1)]
Recognizing that the commitments grow each year at the same rate as As R increases, every term in the denominator decreases, which means that the denominator decreases and C increases. It = Ct-D/D D R i-1 i=1 i.e., dC/dR > 0 Equation (14) identifies the new amount invested, or drawn down, at end 4. Drawdown cycle, D longer drawdown periods require higher initial commitment levels to reach steady state allocations, since the capital accumulation rate is slower. of period t to compute the steady state investment rate for maintaining dynamic equilibrium. (14) the portfolio, we can rewrite (13) as:
Limitations of this framework

We distinguish the sign of the first partial derivative by contemplating equation (10) and find it to be positive. For dC/dD, let M = P [(G R)/(GH RH)] > 0 d[D [(RH+D-1 RH+D-2)/RD 1]/dD = -[(RDRH(R 1) (D ln R RD + 1))/ [R2(RD 1)2]
The model we have described here, while useful to frame issues deterministically suffers naturally from shortfalls; in the real world investors face many types of uncertainty that impact their private equity allocations such as:
Returns the returns for both the portfolio and the private equity allocation are uncertain. Cash flows the draw-down schedule and the investment horizon/ realization timings are uncertain. Valuation investors do not observe true private equity valuations; the observed valuations contain some error, since the prices are not marked-to-market.
RDRH(R
1)/R2(RD
1)2
> 0 and D ln R
RD
+ 1 < 0, so
d[D [(RH+D-1 RH+D-2)/RD 1]/dD > 0 For the latter, we use shows that dC/dD > 0 5. Liquidation period, L longer liquidation periods reduce initial commitment levels, since the capital liquidation rate is slower. 80 ex > 1 + x for all x > 0 and set x = D ln R which
These forms of uncertainty have two implications for investors. Firstly, they change the target private equity allocation for most investors. Secondly, they may change the optimal commitment strategy. There are no easy solutions to addressing these issues. Answers range from assuming that expected returns, draw-downs/cash-flows and valuations are (i) historically arrived at, or (ii) Monte Carlo simulated under simplifying distributional assumptions. In addition, in the real world one needs to make

Bacheliers Legacy: Applications of Theoretical Physics in Market Price Modeling
accommodations for exogenous factors. For instance, cyclical factors such as periods of under-commitment and over-commitment influence capital deployment rates. In general, drawdown rates are lower when commitment rates are high. The effect is more pronounced in first-year investment rates. This phenomenon may partially be explained by the following hypothesis:
Demand for private equity funds is cyclical and is a function of the business cycle. Supply of private equity funds is sticky in the short run and responds to demand with a lag. Increases in the supply of private equity imply tougher competition for capital deployment in subsequent periods (money chasing deals). Thus, we often observe anecdotally reduced capital deployment in the first three years.
The implications of this observation are that investors may expect their money to be invested at lower rates after strong fundraising periods. Consequent returns, as measured by IRRs, may be lower because distributions are also likely to take longer. Does this means that investors should stay away from private equity when fund raising is strong? Probably not. Does this means that investors should gravitate to private equity when fund raising is weak? Probably not, once again.
Conclusion
Investors in a private equity fund commit to providing a pre-agreed sum of capital to the fund over a specified period of time. This is called the limited partners capital commitment. However, the entire funding may not be needed immediately. The general partner draws down or calls the capital over a period of time when investment opportunities arise. Drawdown usually occurs over a four to five year investment period, though this can be sooner as seen in some recent vintages. The flexibility to call capital on an as needed basis, with a few weeks notice, reduces cash drag on overall fund returns. Some investors monitor for timing issues between draw-down and deployment when they qualitatively evaluate performance. Given uncertainty in capital calls investors need to plan their cash flows as well as develop investing strategies to meet desired exposure levels to private equity over a long period of time. This paper provides a framework to explore these issues still further.
81
PART 1
Moving the OTC Derivatives Market to CCPs

Manmohan Singh Senior Economist, Research Department, IMF1
Abstract
Recent regulatory efforts, especially in the U.S. and Europe, are aimed at reducing moral hazard so that the next financial crisis is not bailed out by tax payers. This paper looks at the possibility that central counterparties (CCPs) may be toobig-to-fail entities in the making. The present regulatory and reform efforts may not remove the systemic risk from OTC derivatives but rather shift them from banks to CCPs. Under the present regulatory overhaul, the OTC derivative market could become more fragmented. Furthermore, another taxpayer bailout cannot be ruled out. A reexamination of the
1 The views expressed are those of the author and do not reflect those of the IMF.
two key issues of (i) the interoperability of CCPs, and (ii) the cost of moving to CCPs with access to central bank funding, indicates that the proposed changes may not provide the best solution. The paper suggests that an appropriate levy/tax on derivative liabilities could make the OTC derivatives market safer, rather than the present regulatory push toCCPs.
83
In the aftermath of Lehman and AIG, regulators, especially in the U.S. and Europe, have focused on reducing the risks in the credit-defaultswap (CDS) market, initially, and subsequently in the overall over-thecounter (OTC) derivatives market. In order to achieve their objectives, the regulatory bodies require all subject to some exceptions users of derivatives to post the requisite margins so that there is no undercollateralization in this market. Recent surveys [BIS, 2011] have shown that this U.S.$600trillion OTC derivatives market is seriously under-collateralized and thus contributes to systemic risk. However, recent studies have shown that the associated demand for additional collateral to satisfy the envisaged regulatory efforts will be onerous [Singh (2010); Oliver Wyman (2011)]. Thus, the regulatory effort(s) are meeting resistance from the financial services industry. Another group that is lobbying to avoid posting collateral are the end-users, who (presumably) are genuine hedgers and thus do not contribute toward the systemic risk stemming from the use of OTC derivatives. This article provides an overview of the OTC derivatives market and the associated drawbacks in the proposed regulatory initiatives that continue to unfold. The financial crisis following Lehmans demise and AIGs bailout has provided the impetus to move the lightly regulated over-the-counter (OTC) derivative contracts from bilateral clearing to central counterparties (CCPs). The debate about the future of financial regulation has heated up as regulators in both the U.S. and the European Union seek legislative approval to mitigate systemic risks associated with systemically important financial institutions (SIFIs) that include large banks and non-banks such as hedge funds. In order to mitigate systemic risk that is due to counterparty credit risks and failures, either the users of derivative contracts will have to hold more collateral from bilateral counterparties, or margins will have to be posted to CCPs. There are several initiatives to move the systemic risk from SIFIs derivative books to CCPs, including the Dodd Frank Act in the U.S. and proposals by the European Union that are pending legislative approval. However, there has been very little research that looks at the overall costs to SIFIs of offloading derivative contracts to CCPs. Much of the initial discussion and research [Barclays (2008); ECB (2009)] on risks associated with derivatives was focused on credit derivatives (or the CDS market), which now represents only about 6percent of the overall notional OTC derivatives market, as reported in Bank for International Settlements (BIS) data. The OTC derivatives market has grown considerably in recent years. According to BIS surveys, notional amounts of all categories of the OTC contracts stood at $583trillion at the end of June,2010. These include foreign exchange (FX) contracts, interest rate contracts, equity linked contracts, commodity contracts, and credit default swap (CDS) contracts. Furthermore, especially in the present Basel IIIs regulatory environment, 84 the demand for high quality collateral has increased significantly, while
the supply of collateral has been reduced due to the hoarding of (unencumbered) collateral by SIFIs as reserves. It is envisaged that a critical mass of SIFIs derivative-related risks will be moved to CCPs so that this regulatory effort can bear fruit. A key incentive for moving OTC derivatives to CCPs is higher multilateral netting, i.e., offsetting exposures across all OTC products on SIFIs books. Intuitively, the margin required to cover the exposure of the portfolio would be smaller in a CCP world. However, if there are multiple CCPs that are not linked, the benefits of netting are reduced, because cross-product netting will not take place (since CCPs presently only offer multilateral netting in the same asset class and not across products). Present market practices result in residual derivatives payables and receivables, based on International Swap and Derivatives Associations (ISDA) netting agreements, because:
Sovereigns, AAA insurers/corporates/large banks/multilateral institutions (i.e., EBRD), and the Berkshire Hathway type firms do not post adequate collateral since they are viewed by SIFIs as privileged and (presumably) safe clients.
SIFIs (i.e., dealers) have agreed, based on the bilateral nature of the contracts, not to mandate collateral for dealer-to-dealer positions. In fact, dealers typically post no initial margin/default funds to each other for these contracts.
Regulatory infrastructure for OTC derivatives

A single CCP with an adequate multicurrency central bank liquidity backstop that is regulated and supervised and spans the broadest range of derivatives would have been an ideal first-best solution. Barring this, cognizant of the political realities (and subtleties of market organization), fewer CCPs would be better than a proliferation of CCPs from an exposure, netting, and collateral standpoint. In fact, recent developments have diverged substantially from this firstbest solution. A CFTC draft proposal has lowered the capital threshold for a CCP to U.S.$50million, which will encourage new entrants in this business. Furthermore, end-user exemptions along with not moving certain products like the foreign exchange OTC derivatives to CCPs may not only dilute the intended objectives, but actual outcomes may be suboptimal relative to the status quo. CCPs will require collateral to be posted from all members. In essence, both parties should post collateral to CCPs; no exceptions or exemptions. This is also called two-way CSAs (Credit Support Annexes) under ISDA. However, this is not happening since it is envisaged that there will be exemptions for some end-users. Also, many central banks, sovereigns, and municipalities do not yet post full collateral; thus, moving transactions to CCPs would make the under-collateralization obvious and require large increases in collateral.

By way of background, prior to the momentum to move OTC derivatives from SIFIs books, CCPs were viewed under the rubric of payment systems. After Lehman, regulators are forcing, en masse, sizable OTC derivatives to CCPs. This is a huge transition, primarily to move this risk outside the banking system. These new entities may also be viewed as derivative warehouses, or concentrated risk nodes of global financial markets. Figure 1 illustrates that on average, each of the top ten SIFIs carries about U.S.$100billion of derivatives-related tail risk this is the cost to the financial system from the failure of a SIFI. Yet, instead of addressing the derivatives tail risk, the present regulatory agenda is focused on offloading all (or most) of the OTC derivatives books to CCPs.
DB UBS U.S.$100 billion GS
Large banks active in OTC derivatives and relevant CCPs

Large banks CCPs
MS ICE US
CS LCH Swap clear
BARC
RBS ICE Europe Citi
Issues considered under the proposed regulations for OTC derivatives interoperability of CCPs
Interoperability, or linking of CCPs, allows a SIFI to concentrate its portfolio at a CCP of its choice, regardless of what CCP its trading counterparty chooses to use. Thus, at the level of each CCP, CCPi may hold or have access to collateral from another CCPj that may go bankrupt in the future, so that losses involved in closing out CCPjs obligations to CCPi can be covered. However, legal and regulatory sources indicate that crossborder margin access is subordinate to national bankruptcy laws (such as Chapter 11 in the U.S.) over discussions with regulators/attorneys. Thus, it is unlikely that CCPi in one country would be allowed access to collateral posted by CCPj registered in another country. A recent discussion paper by EuroCCP is reviving some interest in the linking of CCPs [EuroCCP/DTCC (2010)]. The basic premise of this paper is that when CCPs agree to interoperate, they should each increase their default fund as a function of the open positions between them. In other words, default funds in a linked-CCP world will be higher than the status quo, but the linked-CCP world may provide higher netting benefits to justify the larger default funds at CCPs. The augmented default fund proposal avoids the cross-border complexities potentially associated with collateral being trapped in a defaulted CCP. All else being equal, interoperability reduces both (i) the probability of default of CCPs since the overall netting will be higher, which results in smaller tail risks at CCPs, and (ii) the sizable collateral needs associated with offloading derivative risks to multiple CCPs. Also, note that interoperability of CCPs does not equate to one global CCP, since the former would straddle multiple legal jurisdictions. Nevertheless, interoperability allows a way to proxy the first-best solution outlined above to maximize netting and lower the associated collateral needs when moving OTC derivatives to CCPs.
JPM
BofA
CME
Figure1 The SIFIs will offload most of their OTC derivatives to CCPs (instead of tail risk)
argument that suggests the need to minimize the number of CCPs (and benefit from additional netting), rather than increasing their number. Thus, collateral needs will be higher in the proposed world. Most of the major SIFIs derivatives books are largely concentrated in one business (i.e., a legal entity) to run the derivatives clearing business so as to maximize global netting. Some clients like sovereigns and U.S.municipalities are presently not in a position to post collateral. A recent TABB Group [TABB Group (2010)] also estimates under-collateralization in the OTC derivatives market of around U.S.$2trillion and suggests that due to end-user exemptions a significant part of this market will not reach CCPs. Oliver Wyman (2011) also suggests sizable additional collateral needs that are even higher than those suggested by [Singh (2010)] and by TABB Group. ISDA has also acknowledged the sizable collateral needs resulting from moving derivative positions to CCPs, despite their (earlier) margin surveys indicating that most of this market is collateralized. The sizable collateral needs imply that CCPs may not inherit all the derivative positions from SIFIs.
Unbundling of netted positions

The SIFIs are reticent to unbundle netted positions, as this results in deadweight loss and increases collateral needs. Since there is no universally accepted formal definition of a standard contract (or contracts that are clearable at CCPs), there is room for SIFIs to skirt this definition despite the higher capital charge associated with keeping non-standard contracts on their books, since the netting benefits may be sizable relative to the regulatory capital charge wedge. This can be expected of SIFIs where risk management teams build high correlations across OTC derivative products for hedging purposes. Following the Dodd-Frank Act, there may be end-user exemptions including the possibility that U.S. Treasury exempts the entire foreign exchange market. This would 85
Sizable collateral requirements

Without interoperability, the 10 largest SIFIs will continue to keep systemic risk from OTC derivatives on their book and regulatory efforts will introduce more new entities (i.e., CCPs) that will hold systemic risk from OTC derivatives. This goes against the proposed [Duffie and Zhu (2009)]
entice SIFIs not to unbundle positions bundles with foreign exchange contracts (i.e., correlated foreign exchange/ interest-rate swap, or a correlated foreign-exchange/CDS swap, etc.).
members have fallen away) and if the CB provides liquidity support it will be taking credit/solvency risk on whatever the net CCP position is. A CCP failure should not be ruled out. As CCPs begin to clear more complex, less liquid, and longer-term instruments, their potential need for funding support in extremis will rise. In the most extreme scenario, where a temporary liquidity shortfall at a CCP has the potential to cause systemic disruption, or even threaten the solvency of a CCP, it is likely that a CB will stand ready to give whatever support is necessary. However, such an arrangement would create moral hazard. Under the Dodd-Frank Act in the U.S., the Federal Reserve cannot bail out any derivatives dealer. More generally, there is not complete clarity on whether non-banks would have access to CB liquidity. Sections 802 through 806 of Dodd-Frank Act generally authorize the Fed to provide liquidity support under unusual or exigent circumstances to CCPs that have been designated as systemically significant.
Adverse impact on risk management

In an environment where CCPs compete, unlimited loss sharing may not be a viable business model because market participants are likely to choose CCPs with the lowest loss sharing obligations in their rules, everything else being equal. Yet, pushing CCPs to clear riskier and lessliquid financial instruments, as the regulators are now demanding, may increase systemic risk and the probability of a bailout. Banks may also provide loans as collateral and not lose clients/business, so the tail risk may not leave their books.
Regulatory arbitrage likely

Gaps in coordinating an international agenda will result in regulatory arbitrage by SIFIs. Following Senator Lincolns push out clause under Dodd-Frank Act, SIFIs banking groups can keep interest rate, foreign exchange, and investment grade CDS on the banking book. Other OTC derivatives like equities, commodities, and below investment grade CDS have to be outside the SIFIs banking book. This will also lead to unbundling of positions (or a move to another jurisdiction like the U.K. that skirts the Lincoln push out).
Decrease in rehypothecation
The decrease in the churning of collateral may be significant since there is demand from some SIFIs and/or their clients (asset managers, hedge funds, etc.), for legally segregated/operationally commingled accounts for the margin that they will post to CCPs. In addition, the recent demand for bankruptcy remote structures another form of siloing collateral that stems from the desire not to legally post collateral with CCPs in jurisdictions that may not have the central banks lender-of-last-resort backstop (i.e., liquidity and solvency support) will reduce rehypothecation.
Concentration of systemic risk via risk nodes

Regulators are forcing, en masse, sizable OTC derivatives to CCPs. This is a huge transition, primarily to move this risk outside the banking system. If the intended objective(s) are achieved, these new entities should be viewed as derivative warehouses, or risk nodes in financial markets, and not under the payment/settlement rubric.
Supervision of more SIFIs

Regulators will have to supervise more SIFIs, as CCPs will effectively be SIFIs. Furthermore, existing SIFIs (i.e., large banks/dealers) will retain OTC derivative positions since non-standard contracts will stay with them. End-user exemptions and (likely exempt) foreign exchange contracts will not migrate to CCPs. SIFIs may keep some non-standard/standard combination on their books due to netting benefits across products and not move them to CCPs despite higher regulatory capital charge. Post-Lehman there has not been much progress on crisis resolution frameworks for unwinding SIFIs; thus, creating more SIFIs need to be justified. However, policies and regulations in these areas are evolving.
Central bank backstop (or taxpayer bailout)

A CCP may face a pure liquidity crisis if it is suffering from a massive outflow of otherwise solvent clearing members, in which case the risk is that it will have to realize its investment portfolio at low prices. Assume an external shock where everyone is trying to liquidate collateral simultaneously. This will lead to a problem if the CCP has repod out the collateral it has, cannot get it back, and for whatever reason does not want to pay cash to the members (i.e., effectively purchasing the securities at that price). In these circumstances, a central bank (CB) would be repo-ing whatever collateral the CCP would ultimately get back. In such instances, it would be more sensible to require the bank members (i.e., JPMorgan, Credit Suisse) of the CCP to access the CB and then provide the CCP with liquidity. The CCP may also need CB support if it has suffered a series of member defaults and is subject to a run because of credit concerns. In this 86 case, the CCPs book is not balanced (since the trades of the defaulting
An alternative to the CCP route: taxing derivative liability positions

In order to summarize the derivatives risk to the financial system, we measure the exposure of the financial system to the failure of a SIFI that is dominant in the OTC derivatives market, according to the SIFIs total derivative payables (and not derivative receivables). Derivative payables represent the sum of the counterpartys contracts that are liabilities of the SIFI. Similarly, derivative receivables represent the sum of the counterpartys contracts that are the assets of the SIFI.

Under no interoperability, tail risks are less likely to decline. Let p denote the probability of a bail-out in a CCP world, and let P measure the probability of the bailout of a SIFI in the (status quo) non-CCP world. For p < P, overall tail risks in the CCP world would be lower than the tail risks in the non-CCP world. Increased multilateral netting via interoperability is one way this could happen, but this is unlikely, because the needed legal conditions are not in place. Furthermore, no CCP offers crossproduct netting, so contracts that net at a SIFI book may need to be unbundled when moved to two non-linked CCPs. Similarly, between-product netting may also lead to collateral inefficiencies, since a standard/non-standard combination would have to be unbundled. The standard contract would move to a CCP along with the associated collateral, while the non-standard contract would stay with the SIFI and attract a regulatory charge. Such unbundling decreases overall netting. Thus, ex-ante, it remains unclear if the overall netting due to CCPs (primarily between products and not across products) will be higher than that from the unbundling of netted positions or other issues associated with moving derivatives to CCPs (i.e., reduced rehypothecation of collateral due to the siloing of collateral at CCPs, or demands for segregated collateral accounts by certain clients). The decrease in the churning of collateral under restricted rehypothecation may be significant. Another way to reduce tail risks is to take collateral from those who are not posting collateral. This can be done in the CCP world by regulatory incentives. But it could also be done in the status quo world by placing a levy/tax on derivative liabilities (that would result in revenue that could be used if a SIFI needs to be bailed out in the future). Now, let p1 and P1 denote the probability of a bail-out when the present under-collateralization is reduced. Note that p1 < p and P1 < P. Moreover, p1 is largely exogenous due to regulatory uncertainty, while P1 is endogenous since the tax, T, can be calibrated to reduce the risk metric (i.e., residual derivative payables in the non-CCP world). Thus, P1 can be less than p1 and further strengthens the tax argument analytically. However, we will make the worst-case assumption here that they are equal, i.e., p1 = P1. To summarize, the tail (or bailout) risk in the envisaged CCP world and the present SIFI-only world might well remain the same. However, the CCP world would have a bail-out cost of C (i.e., when a CCP and/or a SIFI goes under). The status quo world without CCPs may well have a similar bail-out cost C when a SIFI goes under, but this can be paid by the revenue T via the tax/levy that will be imposed on the large residual derivative liabilities of SIFIs who want to carry this systemic risk. Such tax could be set aside and akin to FDICs insurance premium, or SIPCs (Securities Investors Protection Company) annual assessments of SIFIs.
CCP world At present, probability of bailout Ex-post, probability of bailout Ex-post, cost, C, of bailout in n years p p1 p1C Status quo with tax P P1 P1C n T T=1
As argued above, since p1 is not less than P1, the status quo with tax is economically more efficient.
Box 1 Analytics of the tax on derivative payables
At present, a SIFIs derivative payables do not carry a regulatory capital charge and are not reflected in risk assessments. On the other hand, derivative receivables are imbedded in credit risk and there is already a capital charge/provision for potential non-receivables. By using derivative payables as a yardstick, we thus provide a readily available metric to measure systemic risk from derivatives, compared to other sources that focus on derivative receivables. The five largest European banks had about U.S. $700 billion in undercollateralized risk in the form of derivative payables as of December2008, as per their financial statements. The U.S.banks had around U.S.$650 billion in derivative payables as of end-2008 (as per their financial statements), since dislocations were higher then. The key SIFIs active in OTC derivatives in the U.S. are Goldman Sachs, Citi, JP Morgan,
Bank of America, and Morgan Stanley. In Europe, Deutsche Bank, Barclays, UBS, RBS and Credit Suisse are sizable players. It is useful to note that the International Swap and Derivatives Associations (ISDA) master agreements allow SIFIs to net (or offset) their derivative receivables and payables exposure on an entity. Thus, if Goldman has a positive position with Citi on an interest rate swap and a negative position with Citi on a credit derivative, ISDA allows for netting of the two positions. A tax or a levy on derivative payables (after netting) has been suggested in earlier research by the IMF [Singh (2010)]. This route may be more transparent than moving OTC derivatives to CCPs, especially if the costs to bail out CCPs are to be funded by taxpayers (see Box 1). Even if the CCP route is preferred for other reasons, it is argued that a tax/levy on derivative payables (after netting) may incentivize SIFIs to move to CCPs. 87
Unless the regulatory capital charge wedge between non-cleared derivatives and cleared derivatives is large, a SIFI is unlikely to offload its derivative books to CCPs and forgo the netting benefits across derivatives that exist on its books in the status quo.
These regulatory steps seem unlikely to adequately reduce systemic risks or excess rents from OTC derivatives, and the likelihood of future taxpayer bailouts appears to remain significant. Taxing derivative payables would be a good alternative (or a complementary solution while regulations are finalized).
Conclusions
At a minimum, CCPs should be viewed as very important SIFIs and abide by their principles. Some CCPs, such as LCH.Clearnet/Swapclear in the U.K., clear the bulk of the global interest-rate swaps market over U.S.$200trillion and almost half of this market and are thus interconnected to most other SIFIs active in the OTC derivatives. Furthermore, since LCH is owned by the large SIFIs, their shareholder structure unfolds systemically relevant network linkages and raises contagion and pro-cyclicality risks. Another notable CCP is the ICE Group, which includes ICE Trust in the U.S.and ICE Clear in the U.K. and has a niche in the credit derivatives (CDS) market. Most of their business has so far focused on clearing CDS indices. ICE Groups systemic importance will only grow for two reasons: (a) there will always be jump risk associated with CDS contracts and (b) they will inherit more risky business as SIFIs offload single name CDS to CCPs. Regulation of CCPs should encompass and adhere to basic SIFI rules that include resolution schemes, increased loss absorbency, and supervision by a global regulator that straddles cross border SIFIs. Present efforts to move OTC derivatives to CCPs involve the following:
References

Barclays Quantitative Credit Strategy, 2008, Counterparty risk in credit markets Bank for International Settlements, 2011, OTC Derivative Market Activity, semiannual report Duffie, D. and H. Zhu, 2009 Does a central clearing counterparty reduce counterparty risk? Working Paper, Stanford University, Graduate School of Business European Central Bank, 2009, Banking Supervisory Committee report on CDS and counterparty risk, August Oliver Wyman, 2011, The future of capital markets infrastructure, February (with Morgan Stanley) Segoviano, M. and M. Singh, 2008, Counterparty risk in the over-the-counter derivatives market, IMF Working Paper 08/258 Singh, M., 2010, Collateral, netting and systemic risk in the OTC derivatives market, IMF Working Paper 10/99, International Monetary Fund Singh, M., 2011a, Making OTC derivatives safe a fresh look, WP/11/66, International Monetary Fund, Washington, D.C. TABB Group Study, November 2010, The global risk transfer market
Annex1 Typical OTC derivative position from a SIFIs financial statement

March 2009
1. A significant increase in overall collateral needs. 2. Some netted positions will need to be unbundled. 3. Duplicating risk management teams (at CCPs) which already exist at large banks. 4. Public authorities will have to supervise more SIFIs, as CCPs will effectively be SIFIs. Furthermore, existing SIFIs (i.e., the large banks/ dealer) will retain OTC derivative positions (non-standard contracts, end-user exempted positions, foreign exchange contracts, netted positions, etc). 5. Regulatory arbitrage will increase; stemming from commodity caps in the U.S.and from the push out clause in the Dodd-Frank Act. 6. Rehypothecation, or churning of collateral will decrease, as much of the collateral at CCPs will be segregated at the clients request or in bankruptcy remote structures. 7. Derivative warehouses will be created that are more akin to concentrated risk nodes in global finance. 8. CCPs will be viewed under the payment/settlement rubric. They will thus likely garner CB support and taxpayers could well be on the hook again to bail-out CCPs.
Derivative contracts for trading activities Interest rates Credit Currencies Commodities Equities Subtotal
Derivative Assets
Derivative Liabilities
(in millions)
$ 1,171,827 469,118 92,846 80,275 100,291 $ 1,914,357
$ 1,120,430 427,020 85,612 77,327 92,612 $ 1,803,001
Derivative contracts accounted for as hedges under SFASNo.133 Interest rates Currencies Subtotal Gross fair value of derivative contracts Counterparty netting Cash collateral netting Fair value included in Trading assets, at fair value Fair value included in Trading liabilities, at fair value $ 24,347 50 $ 24,397 $ 1,938,754 (1,685,348) (149,081) $ 104,325 $1 31 $ 32 $ 1,803,033 (1,685,348) (27,065) $ 90,620
88
PART 1
Cultural, Political, and Economic Antecedents of Country Risk in Sixty-Two Countries

Moshe Banai, Professor of Management, Zicklin School of Business, Baruch College, City University
of New York
Abstract
This study presents and empirically tests a model of the relationships between cultural, political, and economic measures and business practices in sixty-two countries. GLOBE studys nine dimensions of cultures scores are regressed along scores on democracy, religiousness, GDP, corruption, and country risk. The results indicate that in-group collectivism and humane orientation are good predictors of democracy, religion, and corruption; performance orientation and uncertainty avoidance are good predictors of societal institutions; and corruption and societal institutions are good predictors of country risk. Implications for scholars and practitioners are offered. 89
Forty years ago numerous researchers had suggested that it would be worthwhile to correlate socio-cultural, political, and economic variables with various dimensions of business practices across nations [Ajiferuke and Boddewyn (1970a); Farmer and Richman (1965); Haire et al. (1966); Harbison and Myers (1959)]. Culture [Hofstede (1980), House et al. (2004), Dyck and Zingales (2004)] and political institutions [Child (1981); Geertz (1973); Glaser and Strauss (1967) have been claimed to influence nations economic performance. Yet, a model that would be able to establish relationships between culture, political institutions, economic performance, and management in organizations could not be fully and empirically tested at that time because of lack of data. Efforts were made to correlate a few variables at a time, however, there was no concentrated effort to offer and test a comprehensive theory that would be able to explain and predict the impact of cultural, political, and economic extraneous variables on business practices cross-culturally. In 2004, the GLOBE research group [House et al. (2004)] came up with nine dimensions of culture measured across sixty-two countries, an achievement that has made the research about the relationships between culture and management statistically possible. This study adopts the GLOBE measures and offers and tests a model of the relationships between cultural, political, and economic measures, and business practices of corruption and risk in sixty-two countries. Corruption and risk are measured at the national level rather than at an organizational level. They are used here as proxies for risk taking involved in the day to day operations of organizations. Hence, in this study the GLOBE studys nine dimensions of cultures scores are regressed along scores for democracy, religiousness, GDP, corruption, and country risk. These variables are discussed in the next section, first culture as the independent variable, and then democracy, religiousness, GDP, as well as corruption and country risk as dependent variables. Several of the relationships analyzed here have been described, explained, and tested in the past. However, no one study has gone so far as to empirically measure culture, political institutions, and economic indicators in one model, certainly not in a sample consisting of sixty-two countries. This is the uniqueness and the contribution of this study. The sample size and richness of the variables have the potential to establish a theory of the relationships between societies cultural values and their institutional and economic performance. When the market is global the ability to predict a nations performance by measuring its peoples social values could prove to be a powerful tool for scientists, practitioners, and politicians. Moreover, efforts to change some of these values may be used by educators to improve a nations performance.
in-group collectivism, humane orientation, uncertainty avoidance, power distance, assertiveness, performance orientation, future orientation, and gender egalitarianism. The measures and their relationships with the dependent variables are presented here. Institutional collectivism is defined as the degree to which individuals are integrated into groups within society. Individuals who are in societies high on institutional collectivism are proud of being in groups, and leaders of those societies encourage group loyalty even if individual goals suffer. The term relates to a cultural value that endorses the prioritization of the groups needs over the individuals needs. The GLOBE study has found a significant and positive relation between the level of institutional collectivism in a country and that countrys economic performance. In-group collectivism defines the degree to which individuals have strong ties to their small immediate groups [Triandis (1994); Triandis et al. (1986)]. In this society children live with their parents until they get married and in many instances continue staying in the same house or the same neighborhood even after their marriage. This type of culture holds different norms and standards for justice for two different groups (the in-group and the out-group) and facilitates the negotiation of justice for familial duty or loyalty based on in-group membership. Pressure for conformity among the in-group members can significantly influence overall levels of corruption in a society through its impact on obedience and loyalty to the group. The GLOBE study has found a strong and significant correlation between in-group collectivism and economic performance. Humane orientation is defined as the degree to which a society encourages and rewards a fair, generous, and kind behavior. This cultural value urges people to generally be very tolerant of mistakes, kind, generous, and always look towards others with a helping hand. This set of behaviors is advocated by all religions and therefore we would assume some relationships between humane orientation and religiousness. Humane orientation is positively correlated with assertiveness, institutional collectivism, in-group collectivism, and performance orientation in the GLOBE study (576). Some of those correlates (performance orientation) may indicate that humane orientation should lead to more democracy and yet some other correlates (in-group collectivism) indicate the opposite. The GLOBE study has found a significant and negative relation between the level of humane orientation in a country and that countrys level of economic performance. Uncertainty avoidance has been defined as the extent to which uncertainty is tolerated in a society. Uncertainty avoidance can be observed by the amount individuals show tolerance for ambiguity, seek feedback, and prefer constant communication of events and things. In organizations it can be observed in the extent of planning that companies do before they take business decisions, and the amount of innovation
Cultural dimensions
House et al.s (2004) measures were selected as the independent vari90 ables in this study. The measures include institutional collectivism,

that these organizations support. People in high uncertainty avoidance cultures will take actions to reduce uncertainty in life, including engaging in illegal activities, which aid in their survival. The GLOBE study has found significant and positive relation between uncertainty avoidance and a countrys economic performance.
between performance orientation and religious belief. However, they are only related to Catholicism and Protestant cultures. The GLOBE study has found a significant and positive relationship between performance orientation in a country and its economic performance. Future orientation the cultural characteristic of future orientation can
Power distance focuses on the degree of equality or inequality between people in a society or a country. The term power distance was coined by Mulder (1971) to mean the degree of inequality in power between a less powerful individual and a more powerful other, where both individuals belong to the same (loosely or tightly knit) social system. Victor and Cullen (1988) indicate that in high power distance cultures, not only would the superior be less tolerant to whistle-blowing behavior but also try to retaliate using all of his power. People in high power distance cultures assume that superior orders override any moral considerations that might apply in other situations, freeing them of responsibility for their actions. This can create a more deleterious environment for corruption. Power distance was found to be significantly and negatively correlated with economic prosperity of countries [House et al. (2004)]. Park (2003) has found power distance to significantly predict corruption. Assertiveness is the extent to which individuals in a society exert their will and opinion in their daily activities and their relationships with others. This is a characteristic that is also often related to masculine properties [Hofstede (1980)] and is treated in the same light in various societies. Consequently, societies that encourage gender egalitarianism are also often observed to encourage assertiveness in daily actions. Speed and scale are venerated values in assertive cultures. Being fast is seen as being efficient and being big is considered to be great [Hofstede (1983a)]. However, in many cases, fast and big achievements are only possible through cutting corners or illegal behavior. In order to achieve a goal faster than others, people feel pressed to provide grease money [Brademas and Heimann (1998)]. Park (2003) who has used Hofstedes (1980) four dimensions of culture found masculinity, of which assertiveness is one vector, to be related to corruption. The GLOBE study has found no relationship between the level of assertiveness in a country and the level of economic performance of that country.
also be viewed in terms of time and the pattern in which societies view time. It relates to the amount of importance a society assigns to longterm thinking and planning for the future. Societies having a future orientation focus on being thrift in the short run and work towards having an ample amount of savings. The GLOBE (2004) study has found a significant negative correlation between future orientation and dislike for democracy. The GLOBE study has found no significant relationship between future orientation and religiousness. The GLOBE study has found a significant, strong positive relation between the level of future orientation in a country and that country per capita GDP. Gender egalitarianism deals with the degree of equality placed on the sexes. It focuses on the degree of traditional gender role on achievement, control, and power. Gender egalitarianism focuses on the extent to which society seeks to minimize gender role differences as compared with societies that tend to maximize the differences by defining clich roles for men and women. In the latter, tasks or behaviors are categorized as either feminine or masculine. Masculine behaviors primarily focus on traits of assertiveness, aggression, and the ability to earn money. Feminine roles are more nurturing, forgiving and subdued. There is no indication in the literature that there is a relationship between the level of gender egalitarianism in a country and the level of democracy in the country or the level of religiousness of its citizens. However, some of the measures of democracy are freedom and equality. Hence, we assume that the more egalitarian a society is the more democratic it becomes. The GLOBE study has found a significant positive relation between the level of gender egalitarianism and the countrys economic performance. Seleim and Bontis (2009), using the Globes nine dimension of culture, found that perceived gender egalitarianism practices were associated with a countrys corruption.
Democracy
Performance orientation is the degree to which a society encourages and rewards group members for performance improvement and excellence. This cultural value reflects the extent to which a community encourages and rewards innovation, high standards, and performance improvements. The GLOBE (2004) study has identified a significant negative correlation between the level of performance orientation of a country and the disdain for democracy in that country. Hofstede and Bond (1988) showed that Confucian dynamism was strongly correlated with economic growth in 22 countries between 1965 and 1985. Trompennars and Hampden-Turner (1998) have suggested that there is no relationship Democracy comes from the Greek word demos meaning people. The U.S. Department of States considers democracies to be countries where the people hold sovereign power over the government, the judiciary, and the legislative. Democracy as a form of government often is assumed to go hand in hand with freedom. Commonly, freedom and democracy are often used interchangeably, and it is generally assumed that a free society is also a democratic one. For this reason, it is possible to identify the time tested fundamentals of constitutional government, human rights, and equality before the law that any society must possess to be properly called democratic. Democracy seeks to assure that the government is based upon the 91
consent of the governed and is accountable to its people. Democracy and corruption Staw and Szwajkowski (1975) found that private firms dealing with less munificent environments engaged in a higher number of illegal activities. Tax policies [Novitzky et al. (1995)], price controls [Mauro (1997)], multiple exchange rate systems, and foreign exchange controls [Levine and Renelt (1992)], and rules on government subsidies [Clements et al. (1995)] are typical examples of government regulations that can lead to illegal rent seeking behaviors. Alam (1995) argues that corruption occurs because the government has a monopoly power over certain resources. Government officials who hold this monopoly power may refuse or delay the authorization of certain economic activities in order to extract bribes from those who need authorizations or permits [Tanzi (1998)]. Bribery is also often used as speed money to skip cumbersome regulatory procedures and as grease for the wheels in over-regulated societies, such as found in many developing countries [Brademas and Heimann (1998)]. When too many regulations and rules frequently impede free economic activities, this creates a breeding ground for illegal rent-seeking behavior by public officials and increases the level of corruption.
lower quality bureaucracies, higher rates of tax evasion, lower rates of participation in civic activities and professional association, a lower level for importance of large firms in the economy, inferior infrastructure, and higher inflation.
Per capita gross domestic product (GDP)

GDP is the market value of a countrys output attributable to factors of production located in the countrys territory. GDP per capita is the countrys gross domestic product divided by the populations size. GDP per capita figures can be misleading because they do not consider differences in the cost of living in different countries. To account for those differences one can adjust the GDP per capita by purchasing power. Purchasing Power Parity (PPP) adjustment allows for a more direct comparison of living standards in different countries. The base for the adjustment is the cost of living in the U.S. [Hill (2004)]. In this study we have used GDP (ppp) as our measurement to compare economic performance in sixty-two countries. GDP and corruption Lesnik and Blanc (1990) describe scarcity as the father of rent seeking. The level of income could be one of the major economic variables that affect the level of rent seeking through its impact on scarcity. In many poor developing countries, people do not have enough financial resources to support their families in a normal way. As Staw and Szwajkowsky (1975) argue, people will understandably take any action, including engaging in illegal activities, when they experience difficulty acquiring the resources necessary for survival. The situation puts pressure on public servants as well to find alternative sources of income. Along the line of efficiency-wage mechanisms, Kraay and Van Rijckeghem (1995) found that when civil servants are not paid enough, they are forced to use their position and power to seek rent. As Leiken (1996) points out, when government does not pay civil servants decent salaries, they cannot buy a layer of insulation against patronage and bribery. In some cases, once a government subscribes to the view that civil servants earn sufficient income from corruption, they may reduce civil servants pay as a consequence [Lambsdorff (1998)]. This can cause a vicious cycle in some countries and encourages illegal activity by civil servants. GDP (ppp) per capita was not found to be related to a countrys level of corruption [Park (2003)]. When GDP was used as a control variable it did not change the positive relationships between in group collectivism and humane orientation and corruption [Seleim and Bontis (2009)]. GDP and religiousness with exception of the U.S., richer nations tend to place less importance on religion [The Pew Research Center (2002)]. One explanation for that finding has been offered by Larry Witham of the Washington Times: ... high rates of professed belief often have been attributed to religious freedom and separation of church and state, whereas in most other countries a state-established religion or a secular state policy has been enforced [Withan (2002)].
Religiousness
The Merriam Webster Dictionary defines religiousness as relating to or manifesting faithful devotion to an acknowledged ultimate reality or deity. Religion and economic performance Weber (1904/1978) argued that religious practices and beliefs had important consequences for economic growth. He suggested that the difference between Catholicism and Protestant doctrines is that while Catholicism focused on good works as the exclusive path to salvation, the Protestant ethic introduced the idea of work as a calling. Martin Luther (1483-1546) emphasized the notion that doing worldly work is not a distraction from godly life. McClelland (1961) proposed a more detailed process in which the Protestant values lead to greater emphasis among parents teaching their children to have high standards and to value independence, traits that resulted in higher need for achievement and better national performance. Religion is an important component of the national culture of a country. Some recent studies find that religion is an important determinant of economic behavior [Barro and McCleary (2003); Dyck and Zingales (2004); Guiso et al. (2003); Stulz and Williamson (2003); Treisman (2000)]. Recently, religious belief has been tested to influence a nations economic growth in the wider context of religions in general rather than focusing on the two Christian mainstreams alone [Barro and McCleary (2003)]. They have concluded that economic growth responds positively to religious beliefs. Religiousness and corruption LaPorta et al. (1997) have found that holding per capita income constant, countries with more dominant 92 hierarchical religions have less efficient judiciaries, greater corruption,

Corruption
Corruption has been defined in many different ways. Even though many articles devote their entirety to this subject alone [Gardiner (1993); Heidenheimer et al. (1989); Dolan et al. (1988)], there is little consensus about its definition. Bureaucracy and corruption theories of corruption range from corruption being induced as a product of bureaucratization in countries such as Britain, the U.S., and other capitalist industrialized countries, to corruption as a part of a permanent bureaucratic inertia, and mal-administration, in developing countries. However, both these viewpoints recognize that corruption is fundamentally a problem of governance [Park (2003)]. Corruption thrives where states are too weak to control their own bureaucrats, to protect property and contract rights, and to provide the institutions that underpin an effective rule of law. Park (2003) has found economic freedom to predict a countrys level of corruption. Corruption and economic performance Glynn et al. (1997) argue that corruption impedes economic development and distorts international trade and investment flow. It also undermines the very foundation of multilateralism, which is the backbone of free world trade. The World Bank and OECD denounce corruption as one of the major problems faced by a globalized world economy, and recently started to take the initiative in combating corruption. Most recently, Zhao et al. (2003) found a negative relationship between the level of corruption and inward foreign direct investment.
Methods
Measurements
Cultural values The GLOBEs [House et al. (2004)] scores on the nine dimensions of culture described earlier have been used in this study. The GLOBE study produces two set of scores, one for society practice on each dimension and the second for society value on each dimension. While the two dimensions measure people perceptions, there are differences between the two which are demonstrated in the following two questions: Practice in this society students are encouraged to strive for continuously improved performance; Value I believe that teen aged students should be encouraged to strive for continuously improved performance. In our study we have adopted the societal practices (perceived norms) rather that the societal values (desirable norms) for the 62 countries.
H H1: H2: H3: H4: H5: H6: H7: H8: Independent variable Power distance Power distance Power distance Group collectivism Group collectivism Assertiveness Assertiveness Uncertainty avoidance Uncertainty avoidance Institutional collectivism Institutional collectivism Institutional collectivism Humane orientation Humane orientation Humane orientation Performance orientation Performance orientation Performance orientation Future orientation Future orientation Future orientation Gender egalitarianism Gender egalitarianism Gender egalitarianism Democracy GDP Religiousness Corruption Relationship Positive Positive Negative Positive Negative Positive No relation Positive Positive Positive Negative Positive Positive No relation Negative Positive Negative Positive Positive Positive Positive Positive Negative Positive Negative Negative Positive Positive Dependent variable Religiousness Corruption GDP Corruption GDP Corruption GDP Corruption GDP Religiousness Democracy GDP Religiousness Democracy GDP Democracy Religiousness GDP Democracy Religiousness GDP Democracy Religiousness GDP Corruption Corruption Corruption Risk
Country risk
Country risk is a measure that predicts the likelihood that political, economic, and financial forces will cause drastic changes in a countrys business environment that will adversely affect the profit and other goals of a particular business enterprise [Hill (2004)]. There are five common measures of country risk: political risk, financial risk, economic risk, composite risk indices, and country credit ratings. Country risk measures are related to future equity returns so that the higher the risk the lower the return on investment [Erb et al. (1996)]. Borio and Packer (2004) have found country risk to be negatively related to good macroeconomic and structural policies.
H9: H10: H11: H12: H13: H14: H15: H16: H17: H18: H19: H20:
Hypotheses
The relationships suggested between dimensions of culture and democracy, religiousness, GDP, corruption, and country risk call for the testing of many hypotheses. Because of the unusually large number of hypotheses generated by the model the hypotheses are presented in Table 1. Not all possible combinations of relationships between variables are offered as hypotheses. Rather, only those relationships that were corroborated in the past in partial models are presented here for empirical investigation.
H21: H22: H23: H24: H25: H26: H27: H28:
Table 1- Summary of the hypotheses
93
Corruption several different data sources are available for the degree of corruption in a country. One of them is the CPI (Corruption Perception Index) developed by Transparency International, which ranks 85 countries and has been published annually since 1995. The validity and reliability of the CPI index has been endorsed by Lancaster and Montinola (1997), and the index has been widely used in recent studies [Husted (1999); Swamy et al. (1999)]. However, the reversed scale of TFs CPI index (Corruption free = 10 and total corruption = 0) can cause a problem when interpreting statistical results. Extra caution is needed to interpret the coefficient signs in statistical results because a negative coefficient sign means that the factor is positively related to the degree of corruption in a country. GDP per capita for national income, GDP (ppp) per capita data are used to eliminate the size effect. The data for these variables are drawn from the World Development Indicators published by the World Bank (1999). Country risk the measure used here to estimate countries combined political, financial, and societal risks has been offered by OECD (2005). OECD uses the Kneepan Package as a system for assessing country credit risk and classifying countries into eight country risk categories (0-7). The Country Risk Classification Method measures the country credit risk, i.e., the likelihood that a country will service its external debt. The classification of countries is achieved through the application of a methodology comprised of two basic components: (1) the Country Risk Assessment Model (CRAM), which produces a quantitative assessment of country credit risk, based on three groups of risk indicators (the payment experience of the participants, the financial situation, and the economic situation); and (2) the qualitative assessment of the models results, considered country-by-country to integrate political risk and/or other risk factors not taken (fully) into account by the model. The details of the CRAM are confidential and are not published. The final classification, based only on valid country risk elements, is a consensus decision of the sub-group of country risk experts of the participating export credit agencies. The sub-group of country risk experts meets several times a year. These meetings are organized so as to guarantee that every country is reviewed whenever a fundamental change is observed and at least once a year. Whilst the meetings are confidential and no official reports of the deliberations are made, the list of country risk classifications is published after each meeting. Democracy the Freedom in the World (www.freedomhouse.com) survey provides an annual evaluation of the state of global freedom. The survey, which includes both analytical reports and numerical ratings of countries and select territories, measures freedom by assessing two broad categories: political rights and civil liberties. Political rights enable 94 people to participate freely in the political process. This includes the right
to vote and compete for public office and to elect representatives who have a decisive vote on public policies. Civil liberties include the freedom to develop opinions, institutions, and personal autonomy without interference from the state. Freedom House assigns each country and territory a political rights and civil liberties rating, along with a corresponding status designation of free, partly free, or not free. The survey does not rate governments or government performance per se, but rather the real-world rights and freedoms enjoyed by individuals as a result of actions taken by both state and nongovernmental actors. The survey team does not base its judgment solely on the political conditions in a country or territory (i.e., war, terrorism), but on the effect that these conditions have on freedom. Freedom House does not maintain a culture-bound view of freedom. The methodology of the survey established basic standards drawn from the Universal Declaration of Human Rights. These standards apply to all countries and territories, irrespective of geographical location, ethnic or religious composition, and level of economic development. Religion The Pew Research Center conducted a poll whose results were released on the 19th of December 2002 that questioned whether people around the world consider religion to be personally important [The Pew Research Center (2002)]. Results (percentage of adults for whom religion is important) showing the balance of interest in secularism and religion were reported from 41 countries. Among the five countries in which religion is considered most important, four have been involved in recent religiously based conflicts in which persons of one religion engage in mass murder of those of another religion. Countries in which less than 60% of the public considers religion to be important have been free of inter-religious conflict, with the exception of the Protestant-Catholic troubles in Northern Ireland.
Statistics
Pearson Moment Correlations were run for all variables included in the model. This procedure allows us to investigate the direct relationships between all the variables. Five step-wise regression analysis processes were run. In the first stage all variables, namely, nine cultural dimensions, religion, democracy, corruption and GDP, were regressed along the dependent variable of country risk. In the second stage, eight cultural dimensions (institutional collectivism was omitted), religion, democracy, and GDP were regressed along the independent variable of corruption. In the third stage, the same variables were regressed along the dependent variable of institutional collectivism. In the fourth stage, the eight cultural dimensions were regressed along the independent variable of religion. In the fifth stage the same eight variables were regressed along the dependent variable of democracy. These procedures, while losing some information, have focused the analysis on the main relationships between the variables, thereby increasing the chances for causal relationships.

Results
Correlations between the variables are presented in Table 2. As indicated in Table 2, country risk is positively correlated with religion (.61), group collectivism (.61), and negatively correlated with uncertainty avoidance (-.53), future orientation (-.38), and per capita GDP (-.37). Corruption is positively correlated with country risk (.86), group collectivism (.71), religion (.65) and humane orientation (.41), and negatively correlated with uncertainty avoidance (-.54), future orientation (-.45), and per capita GDP (-.37). Institutional collectivism (institution building) is positively correlated with future orientation (.46), performance orientation (.43), and humane orientation (.43), and negatively correlated with power distance (-.43), and assertiveness (-.43). Religion is positively correlated with humane orientation (.57), and group collectivism (.49). Democracy is positively correlated with uncertainty avoidance (.42), future orientation (.36), and per capita GDP (.36), and negatively correlated with corruption (-.82), country risk (-.73), group collectivism (-.66), religion (-.45), and power distance (-.39). Following these correlations, variables have been included in a set of stepwise regression analyses. The results for the regression analyses are presented in Table 3. Five models of Stepwise Regression have yielded the following results: in the first model country risk has been significantly explained by corruption (B = .89; p < .001) and by institutional collectivism (B = -.23; p<.03). In the second model corruption has been significantly explained by democracy (B = .58; p < .001), by religion (B = .26; p < .026) and by group collectivism (B = .26; p < .04). In the third model religion has been significantly
Cultural values Humane orientation (+.57) Group collectivism (-.66) Performance orientation (+.83) Uncertainty avoidance (-.54)
Social institutions Religiousness (+.26) Democracy (-.58) +.26
Business practices
Performance
Corruption (+.89) Country risk
Institutional collectivism (-.23)
Figure 1 An empirical model of the relationships between cultural dimensions, religiousness, democracy, corruption and institutions, and country risk (all standardized coefficients are significant).
explained by humane orientation (B = .57; p < .005). In the fourth model democracy has been significantly explained by in-group collectivism (B = .66; p < .001). In the fifth model institutional collectivism has been significantly explained by performance orientation (B = .83: p < .001) and by uncertainty avoidance (B = .54; p < .027). The results indicate that per capita GDP (ppp) does not influence a countrys risk; humane orientation influences religiousness which, in turn, increases corruption; group collectivism directly and negatively decreases democracy and increases corruption; performance orientation increases and uncertainty avoidance decreases institutional collectivism; corruption increases a countrys risk and institutional individualism decreases a countrys risk. All together 10 out of the 28 hypotheses have been corroborated. These findings are statistically significant.
DEM DEM COR RSK RLG GDP ICL GCL HMN UCA PDS ASS RF FUT GDR -82 -73 -45 36 00 -66 -33 42 -39 03 16 36 22
COR
RSK
RLG
GDP
ICL
GCL
HMN
UCA
PDS
ASS
PRF
FUT
86 65 -37 -14 71 41 -54 37 05 -23 -45 -21 61 -37 -19 61 31 -53 34 15 -19 -38 -10 -29 -13 49 57 -05 -05 13 17 24 -21 08 -41 -20 27 -14 00 06 26 14 -12 43 40 -43 -42 43 46 -01 27 -61 55 08 -14 -47 -20 00 -15 -42 25 07 -15 -50 -07 58 76 -06 16 -36 -52 -30 49 64 -07 63 -13 -06
DEM Democracy; COR Corruption; RSK Country risk; RLG Religiousness; GDP Per capita gross domestic product; ICL Institutional collectivism; GCL Group collectivism; HMN Humane orientation; UNA Uncertainty avoidance; PDS Power distance; ASS Assertiveness; PRF Performance orientation; FUT Future orientation; GDR Gender equality. N (all variables but Religiousness) = 58-62; N (Religion) =22; All values > .36 are significant at .005; All values >.45 are significant at .001
Table 2 Correlations among variables
95
Standard coefficient beta Model 1 Step 1 RSK Constant COR Step 2 Constant COR ICL Model 2 Step 1 COR Constant DEM Step 2 Constant DEM REL Step 3 Constant DEM REL GCL Model 3 Step 1 RLG Constant HMN Model 4 Step 1 DEM Constant GCL Model 5 Step 1 ICL Constant PRF Step 2 Constant PRF UCA 0.83 -0.54 0.49 0.66 0.57 0.58 0.26 0.26 0.69 0.34 0.85 0.88 -0.22 0.89
Significance
R-squared
AdjustedR-squared
SE (estimate)
0.67 8.43 2.46 9.28 -2.37
0.50 0.00 0.02 0.00 0.03
0.89
0.79
0.79
1.16
0.92
0.84
0.82
1.04
2.13 7.18 0.08 6.22 3.01 -2.11 5.06 2.43 2.21
0.05 0.00 0.94 0.00 0.00 0.05 0.00 0.03 0.04
0.85
0.72
0.71
19.57
0.91
0.81
0.79
16.51
0.92
0.85
0.83
15.05
-2.01 3.11
0.05 0.00
0.57
0.33
0.29
24.48
-4.89 6.77
0.00 0.00
0.66
0.44
0.43
33.21
2.24 2.55 2.69 3.71 -2.39
0.36 0.19 0.01 0.00 0.03
0.49
0.24
0.21
0.38
0.65
0.42
0.36
0.34
DEM Democracy; COR Corruption; RSK Country risk; RLG Religiousness; ICL Institutional collectivism; GCL Group collectivism; HMN Humane orientation; UNA Uncertainty avoidance; PRF Performance orientation
Table 3 Stepwise regression analysis
Discussion and conclusions

This study empirically demonstrates that culture and institutions have the potential to influence a countrys business environment. More specifically, cultural dimensions such as in-group collectivism, humane orientation, performance orientation, and uncertainty avoidance impact a countrys democracy, religion, and institutions that, in turn, influence a countrys degree of corruption and credit risk. The standard of living in a certain country, as measured by per capita GDP, does not influence its level of corruption nor does it impact its credit risk. This finding provides empirical support to early students of culture argument that culture influences management in organizations [Ajiferuke and Boddewyn (1970a); Farmer 96 and Richman (1965); Haire et al. (1966); Harbison and Myers (1959)].
Humane orientation is a cultural value that urges people to generally be very tolerant of mistakes, kind, generous, and always look towards others with a helping hand. It is not surprising that it correlated highly with religion since in each religion there are ingredients of preaching for tolerance and love. What is unclear about the relationship between those two variables is which one is the cause and which is the result. Further analysis is needed to answer this question. Of the nine different cultural dimensions tested, in-group collectivism has proven to be a very significant determinant of democracy. In-group collectivism urges people to stick to their small group at the expense of the greater public. It encourages people to positively discriminate the

in-group, which practically means discriminating against the out-group. When a countrys social system is comprised of families, tribes, or groups of different ethnic background and denomination, and each sticks to its own, it is hardly surprising that those groups find difficulties in cooperating with each other in building an equal and free democratic society. Both performance orientation and uncertainty avoidance were found to be related to institutional collectivism, which in turn, was found to be related to country risk. The establishment of the variable of institutional collectivism as a moderator between other cultural dimensions and a countrys risk calls for the redefinition of the concept of institutional collectivism. House et al. (2004) have defined it as a cultural value that endorses the prioritization of the groups needs over the individuals needs. The operational variables used to measure the concept include issues such as: the economic system in this society is designed to maximize individual interest (reversed); the pay and bonus system in this organization is designed to maximize individual interest (reversed); and in this organization group members take pride in the individual accomplishment of their group. It is possible that institutional collectivism does not actually measure a value but it is rather a structural measurement that measures the existence of groups institutions. It has been interpreted as such in this study but more inquiry into the nature of institutional collectivism is called upon. GDP per capita was found to be unrelated to a countrys corruption level in Parks (2003) and in Seleim and Bontis (2009) studies. This finding has been replicated here. Moreover, GDP per capita has been significantly correlated with some of the independent variables cultural dimensions such as in-group collectivism, uncertainty avoidance, and future orientation as hypothesized. However, further investigation that used regression analysis procedure has not corroborated the relationship between per capita GDP and the dependent variable of a countrys risk.
between variables the ability to explain causality of those relationships is limited. The model is exploratory by nature and the lack of theory about some aspects of the model makes it difficult to prove that the relationships between variables depicted here are unidirectional. Third, for parsimony reasons the model includes only a selected number of variables and ignores others that might add to its explanatory power. Despite those limitations, the model is a first of its kind in its comprehensiveness and the number of countries analyzed. It involves variables that are at the essence of our social debate such as culture and democracy, and it provides partial answers to questions such as how ready is a country to become democratic? and how could we reduce a countrys credit risk? Moreover, while common answers to these frequently asked questions involved economic, political, and even military measures, the answers provided by this study are societal and cultural. Future research may use the model of the influence of cultural values on democracy, corruption, and institution to calculate a countrys risk by measuring peoples attitudes. This type of work has a great potential for financial institutions that are constantly searching for new and better methods for country credit risk assessment. Moreover, political scientists and government research groups may use the model to estimate the likelihood of a non-democratic country becoming democratic by measuring the strength of societal values such as in-group collectivism that have the potential to hamper efforts for dissemination of democratic principles. Business researchers may replace the dependent variables of corruption and risk by other valuable business measures such as level of entrepreneurship or innovation. Lastly, educators may use the model to assess those countrys values that need to be reformed in order to enhance democracy, minimize corruption, improve institution building, and consequently enhance a countrys access to international credit.
References
In this study, the measure for corruption has correlated so highly with the measure for a countrys risk that it seems that most people who evaluate the risk are mostly influenced, even if unconsciously, by the level of corruption in the country. Since a countrys risk is probably established by perceptions this study has corroborated the relationship between these two variables one more time. The study has some limitations. First, it is based on data that have been complied over a certain number of years and therefore the time line of the model is not crystallized. Some of the data were collected by the GLOBE study between the years 1990 and 2000 and the data for the countries level of per capita GDP, religiousness, and corruption have been aggregated between 2002 and 2006. Yet, there is no reason to believe that most, if not all, of these measures are stable over time. Second, despite the use of regression analyses procedures to analyze the relationships

Alam, M. S., 1995, A theory of limits on corruption and some applications, Kyklos, 48, 419-435 Brademas, J. and F. Heimann, 1998, Tackling international corruption: no longer taboo, Foreign Affairs, September October, 77:5, 17-22 Barro, R. J. and R. M. McCleary, 2003, Religion and economic growth across countries, American Sociological Review, 68:5, 760-781 Borio, C. E. V. and F. Packer, 2004, Assessing new perspectives on country risk, BIS Quarterly Review, December, 47-65 Clements, B., R. Hugounenq, and G. Schwartz, 1995, Government subsidies: concepts, international trends and reform options, IMF Working Papers 95/91, Washington, D.C., International Monetary Fund Child, J., 1981, Culture, contingency and capitalism in the cross national study of organizations, in Cummings, L. L. and B. M. Staw (eds.), Research in organizational behavior, 3, 303-356, London, JAI Dolan, K., B. McKeown, and J. M. Carlson, 1988, Popular conceptions of political corruption: implications for the empirical study of political ethics, Corruption and Reform, 3, 3-24 Dyck, A. and L. Zingales, 2004, Private benefits of control: an international comparison, Journal of Finance, 59:2, 537-600 Erb, C. B., C. R. Harvey, and T. E. Viskanta, 1996, Political risk, economic risk and financial risk, Financial Analyst Journal, November December, 29-46 Farmer, R. N. and B. M. Richman, 1965, Comparative management and economic progress, Homewood, Illinois, Richard D. Irwin, Inc.
97
Freedom in the World, www.freedomhouse.com Gardiner, J. A., 1993, Defining corruption, Corruption and Reform, 7, 11-124 Geertz, C., 1973, The interpretation of cultures, New York, Basic Books Glaser, B. G. and A. L. Strauss, 1967, The discovery of grounded theory: strategies for qualitative research, Chicago, Aldine Publishing Company Glynn, P., S. Kobrin, and M. Naim, 1997, Globalization of corruption, in Elliot, K. A., (ed.), Corruption and the global economy, 7-27, Washington, Institute for International Economics Guiso, L., P. Sapienza, and L. Zingales, 2003, Peoples opium? Religion and economic attitudes, Journal of Monetary Economics, 50, 225-282 Haire, M., E. E. Ghiselli, and L.W. Porter, 1966, Managerial thinking: an international study, New York, Wiley Harbison, F. H. and C. A. Myers, 1959, Management in the industrial world, New York, McGraw Hill Heidenheimer, A. J., M. Johnston, and V. T. Levine, 1987, Terms and definition: an introduction, in Heidenheimer, A. J., M. Johnston, and V. T. Levine, (eds.), Political corruption: a handbook, New Brunswick, Transaction Publisher Hill, C. W. L, 2004, Global business today, Boston, McGraw Hill, Irwin Hofstede, G., 1980, Culture consequences: international differences in work-related values, Beverly Hills, Sage Hofstede, G., 1983a, Motivation, leadership and organization: do American theories apply abroad? Organizational Dynamics, 9, 42-63 Hofstede, G., 1983b, National cultures in four dimensions, International Studies of Management and Organization, 13:1-2, 46-74 Hofstede, G. and M. H. Bond, 1988, The Confucius connection: from cultural roots to economic growth, Organizational Dynamics, 16, 4-21 House, R. J., P. J. Hanges, M. Javidan, P.W. Dorfman, and V. Gupta, 2004, Culture, leadership and organization: the GLOBE study of 62 societies, Thousand Oaks, CA, Sage Publication Husted, B. W., 1999, Wealth, culture and corruption, Journal of International Business Studies, 30:2, 339-360 Kraay, A. and C. Van Rijckeghem, 1995, Employment and wages in the public sector: a cross country study, IMF Working Papers 95/70, Washington, International Monetary Fund Kwok, C. C.Y. and S. Tadesse, 2006, National culture and financial systems, Journal of International Business Studies, 37:2, 227-247 Lambsdorff, J. G., 1998, Corruption in comparative perception, in Jain, A. K., (ed.), Economics of corruption, 81-103, Boston, MA, Kluwer Academic Publication Lancaster, T. and G. Montinola, 1997, Toward a methodology for the comparative study of political corruption, Crime, Law and Social Change, Special Issue on Corruption and Reform, 27:3/4, 185-206 LaPorta, R., F. Lopez-De-Silanes, A. Shleifer, and R.W. Vishny, 1997, Legal determinants of external finance, Journal of Finance, 52, 1131-1150 Leiken, R. S., 1996, Controlling the global corruption epidemic, Foreign Policy, Winter, 105, 555-573 Levine, R. and D. Renelt, 1992, A sensitivity analysis of cross-country growth regressions, American Economic Review, 82:4, 942-963 Mauro, P., 1997, The effects of corruption on growth, investment and government expenditure: a cross-country analysis, in Elliot, K. A., (ed.), Corruption and the global economy, 88-107, Washington, D.C., Institute for International economics McClelland, D. C., 1961, The achieving society, Princeton, NJ, D. Van Nostrand Company, Inc. Merton, F., 1957, Social theory and social structure, Revised Edition, London, Free Press of Glencoe Mulder, M., 1971, Power equalization through participation, Administrative Science Quarterly, 16, 31-38 The Pew Research Center, 2002, http://pewresearch.org Noland, M., 2005, Religion and economic performance, World Development, 33, 8, 1215-1232. Novitzky, I., V. Novitzky, and A. Stone, 1995, Private enterprise in Ukraine: getting down to business, Washington, World Bank Private Sector Division OECD, 2005, http://www.oecd.org/document/49/0,2340, en_2825_495663_1901105_1_1_1_1,00.html Park, H., 2003, Determinants of corruption: a cross national analysis, The Multinational Business Review, 11:2, 29-48 Porter, M., 1990, The competitive advantage of nations, New York, Basic Books Staw, B. M. and E. Szwajkowski, 1975, The scarcity munificence of organizational environments and the commission of illegal acts, Administrative Science Quarterly, 20, 345354 Stulz, M. R. and R. Williamson, 2003, Culture, openness, and finance, Journal of Financial Economics, 70, 313-349. Swamy, A., S. Knack, Y. Lee, and O. Azfar, 1999, Gender and corruption, working paper, IRIS Center, University of Maryland
Tanzi, V., 1998, Corruption around the world: causes, consequences, scope, and cures, IMF Working Papers, WP/98/63, 1-39 Transparency International, 2005, http://ww1.transparency.org Treisman, D., 2000, The causes of corruption: a cross-national study, Journal of Public Economics, 76, 399-457 Triandis, H. C., 1994, Theoretical and methodological approaches to the study of collectivism and individualism, in Kim, U., H.C. Triandis, C. Kagitcibasi, S. Choi and G. Yoon, (eds.), Individualism and collectivism, Vol. 18, Cross- cultural research and methodology series, Thousand Oaks, CA, Sage Publication. Triandis, H. C., R. Bontempo, H. Bentancourt, M. Bond, K. Leung, A. Brenes, J. Georgas, C.H. Hui, G. Marin, B. Setiadi, J. B. P. Sinha, J. Verma, J. Spangenberg, H. Touzard, and G. De Montmollin, 1986, The measurement of etic aspects of individualism and collectivism across cultures, Australian Journal of Psychology, 38, 257- 267 Trompenaars, F. and C. Hampden-Turner, 1998, Riding the waves of culture: understanding diversity in global business, New York, NY, McGraw-Hill Victor, B. and J. B. Cullen, 1988, The organizational bases of ethical work climates, Administrative Science Quarterly, 33:1, 101-125 Weber, M., 1904/1978, Economy and society: an outline of interpretive sociology, Vol. 1, Berkeley, University of California Press Witham, L., 2002, U.S. religiousness tops among worlds industrial nations, Washington Times, Dec 20 World Bank, 1999, World development indicator, Washington, D.C., The World Bank Group Zhao, J. H., S. H. Kim, and J. Du, 2003, The impact of corruption and transparency on foreign direct investment: an empirical analysis, Management International Review, 43:1, 41-62
98
99
Part 2
Markets for CCPs and Regulation: Considering Unintended Consequences What Have We Learned from the 2007-08 Liquidity Crisis? A Survey Making Sense of Asset Prices: A Guide to Required Yield Theory: Part 1 valuing the stock market Our Understanding of Next Generations Target Operating Models The First Line of Defense in Operational Risk Management The Perspective of the Business Line Optimal Bank Planning Under Basel III Regulations A Risk Measure for S-Shaped Assets and Prediction of Investment Performance ILLIX A New Index for Quantifying Illiquidity How Homogeneous Diversification in Balanced Investment Funds Affects Portfolio and Systemic Risk Breaking Through Risk Management, a Derivative for the Leasing Industry A Financial Stress Index for the Analysis of XBRL Data
102
PART 2
Markets for CCPs and Regulation: Considering Unintended Consequences

Serge Wibaut Professor of Finance at Universit Catholique de Louvain (Belgium) and
Partner at Reacfin
D Sykes Wilford Hipp Chair Professor of Business, The Citadel1

Abstract
Central clearing counterparties (CCPs) have become a crucial element of the financial architecture and as such they have lately attracted the attention of prudential authorities at the highest level emphasising their crucial, if not too well understood, role in the financial system, while also reflecting anxiety at imagining the possibility of a main clearing house failing during times of stress. Post Lehman directives and legislation, both in Europe and the U.S., have placed more emphasis on CCPs within the global financial system as a possible fix to derivatives related banking system risk. Many papers have argued for mutualisation of counter party risk, at various degrees, to better stave off another financial crisis, which may be driven, in particular, by derivatives trading. In this paper issues related to regulation of this already heavily regulated industry are discussed in light of the insurance industry literature, focusing on potential issues that may arise from various efforts to solve counterparty problems with greater regulation or legislative directive. Raising concerns about the unintended consequences of particular forms of socialization or regulation of CCPs, the work intends to highlight possible side effects of centralized or monopoly
1 We wish to thank Bluford H. Putnam for his comments on earlier versions of this paper and Richard Levich for a supporting graphic. The views in this paper are solely those of the authors and do not necessarily reflect the views of any affiliated organization.
(either from regulation or fiat) systems for counterparty clearing that may (will likely) lead to greater, not less, systematic risk in ways not foreseen in any potential legislation. To do so, the authors build several simple conceptual structures of markets for CCPs and discuss various profit maximizing structures that may be taken by a CCP under different conditions with allusion to the insurance industry. What becomes clear from the discussion is that the various business models will lead to complexities not easily anticipated and understood due to the interactions of many different economic agents. Regulation, both existing and potential, quickly complicates the various business relationships that allow markets to function and various products and agents to interact at all levels. Simple solutions to the systematic issues raised by the financial crisis via regulation or legislation can have far from simple implications for the stability of the system as a whole.
103
Central clearing counterparties (CCPs) have become a crucial element of the financial architecture and as such they have begun to attract the attention of prudential authorities at the highest level. Regulatory authorities have expressed anxiety at imagining the possibility of a main clearing house failing during times of stress, yet they emphasize the important, if not always well understood, role CCPs play in the financial system. Moreover, post-Lehman directives and legislation, both in Europe and the U.S., have placed more emphasis on CCPs within the global financial system as a possible fix to derivatives related banking system risk. Many papers have argued for mutualization of counterparty risk, at various degrees, to better stave off another financial crisis. In this paper, issues related to regulation of this already heavily regulated industry are discussed in light of the insurance industry literature, focusing on potential issues that may arise from various efforts to solve counterparty problems with more detailed or extensive rules for the industry or by legislative directive. Raising concerns about the unintended consequences of particular forms of socialization or regulation of CCPs, the work intends to highlight possible side effects of centralized or monopoly (either from regulation or fiat) systems for counterparty clearing that may (will likely) lead to greater, not less, systematic risk in ways not foreseen in any potential legislation. To illustrate the potential for unintended regulatory consequences, the approach is to build several simple conceptual structures of markets for CCPs and discuss various profit maximizing structures that may be taken by a CCP under different conditions with allusion to the insurance industry. What becomes clear from the discussion is that the various business models will lead to complexities not easily anticipated and understood due to the interactions of many different economic agents. Regulation, both existing and potential, quickly complicates the various business relationships that allow markets to function and various products and agents to interact at all levels. Simple solutions to the systematic issues raised by the financial crisis via regulation or legislation can have far from simple implications for the stability of the system as a whole.
years [Bernanke (2011)]. But the settlement process has also become considerably more complex due the increased cross-border activity, the financial deregulation that marked the last three decades, and the rise of emerging markets. All these made investors (and regulators) aware of the importance of having a robust clearing and settlement system that could deliver a fast and high-quality service at a reasonable cost. CCPs are heavily connected with securities depositaries, custodians, and the general payment systems of our economies; both in their intrinsic role and in their interaction with the other financial institutions they have become a crucial link in the long chain of financial actors. Figure 1 shows how clearing exchanges are one of the main actors in the web (some would say, the maze) of the global payment and settlement system in the U.S. This figure also shows how this system is fragmented and complex. One can only guess that very few actors fully understand the intricacies and links between the different institutions playing a role in the global payment framework. The settlement and clearing industry has also witnessed during the last decade a sharp rise in consolidation. If consolidation helps increase the efficiency of the clearing and settlement process one cannot ignore the fact that the potential systemic consequences of a central counterpartys failure increase with its size. As such, clearing and settlement cannot but attract the attention of all financial actors, private and public alike. Recent developments in the clearing industry present numerous challenges to regulators and to central banks. The latter have a clear interest in seeing clearing houses and the payment and settlement systems function smoothly so that monetary policy may be implemented in the most efficient manner. As such they may be tempted to encourage the sector to become even more concentrated. At the same time, regulators are concerned with the potential systemic impact of a significant failure within the clearing and settlement infrastructure and they fear that the larger the failing institution is, the larger the systemic risk might be.
Appreciating the regulatory challenges

Central clearing counterparties (CCPs) have become a crucial element of the financial systems architecture. Clearing houses are defined as the location where financial institutions can process their financial obligations and exchange payments. In this process, CCPs may assume counterparty risk, but this is not necessarily so. However, if they do so they greatly simplify the risk management for their participants, but they face the task of managing counterparty risk themselves. The nature of the risk run by CCPs has evolved through time essentially with the exponential use of derivative products. Unlike spot market securities transactions, which are usually cleared and settled in a few days, 104 derivative contracts are often outstanding for months and sometimes Similar to most financial actors, CCPs have followed a trend of technological innovation vastly improving how international users gain access to the markets and their associated clearing systems. These innovations have been calculated such as to reduce the costs of trading and clearing to attract an ever-increasing level of cross-border as well as domestic trading. Like most of these actors, CCPs increase operational efficiency in several ways, but one of their unique features is their credit mitigation role. By becoming counterparties to derivative contracts through novation [Bliss and Papathanassiou (2006)] and by guaranteeing the performance of those contracts for longer periods than do other clearing or settling financial institutions, CCPs have become somehow more than passive spectators in the market for derivative products. In this respect,

Figure 1 U.S. regulatory authority over payment, clearing, and settlement systems
CCPs need to manage their risk as would any financial institution active in those markets. Market and credit risk management can be considered at both the technical level (i.e., what algorithm should be used to compute the price of an option) and at a more conceptual level (i.e., which features should be used to mitigate credit risk). CCPs have at their disposal several instruments to achieve the latter and reduce the probability of their failure: margins, transactions costs, equity, debt, and default funds are the most commonly used instruments considered by CCPs as absorbers or insurance against credit shocks. CCPs have lately attracted the attention of prudential authorities even at the highest level [Trichet (2006); Bernanke (2011); Tucker (2011)], emphasizing their crucial if not too well understood role in the financial system but also reflecting these authorities anxiety at imagining the possibility of a main clearing house failing in times of stress.
Main points of concerns of the public authorities are market structure and risk management in the clearing industry. The two are obviously interwoven. In the past few years, the market structure of the industry has evolved. Vertical and horizontal integration has taken place while simultaneously integrated structures have replaced fragmented clearing activities. In order to facilitate the cross-border clearing of assets (and more specifically of derivative products) and beside innovation, clearing houses have adopted one (or more) of three possible models of integration. The first model consists simply of domestic horizontal consolidation (i.e., merging different derivatives clearing houses in the same jurisdiction, with each of these clearing houses being connected to a different exchange). The merger in 2008 of the CME (already previously merged with 105
The Working Group on Post Trade Services [BIS (2007, 2010)] has exCCP in country 1
tensively examined the level of risk associated with each type of CCP structure.
CCP in country 2
CCP in country31 CCP in country
CCP in country 4
They conclude that a vertical structure offers several risk reduction features. For instance, cost savings arising from economies of scope reduce operational risk and so does the greater level of straight through processing. Vertical integration should also facilitate monitoring and control of participants risk as information should flow between trades and post-trade functions. Supporting this argument, Duffie and Zhu (2009) suggest that a single CCP reduces counterparty exposure relative to a multiple set of CCPs when all derivatives are traded on the one single, overall CCP for derivatives. Heller and Vause (2011) tend to agree with this position.
Exchange Figure 2a Example of international horizontal integration
CCP in country 2
CCPClearing in country 3
Settlement
CCP in country 4
Exchange Clearing Settlement

Figure 2b Example of vertical integration
A CCP within a vertically integrated group may be tempted to distort pricing in order to prevent new entrants in the industry, however. Such a CCP the CBOT), NYMEX, and COMEX2 provides an example of such domestic horizontal consolidation. The second model consists of horizontal consolidation of derivatives clearing houses in different jurisdictions (allowing hence cross-border clearing), very often together with some horizontal consolidation of securities clearing houses in the same jurisdictions (to permit cross-product clearing). An instance of such a merger was the one that took place in 2003 between the London Clearing House and Clearnet SA, which gave rise to LCH.Clearnet. The third model is vertical consolidation with exchanges, clearing houses, and central securities depositories in order to create integrated trading, clearing, and settlement silos. The possible merger of the London Stock Exchange with LCH.Clearnet would be an example of such a vertical consolidation. As observed in Figure 1, all three forms of integration exist for several entities. Quite naturally, the structure of the clearing industry affects the way governance and risk management are organized. A specific structure creates its own specific risks and also specific incentives from its users. For instance, increasing the verticality of a clearing group (i.e., the CCP becomes part of a group that offers a broad set of trade and post-trade services) will entail different kinds of interdependencies than a horizontal structure that brings together markets located in different jurisdiction. In the same way, encouraging concentration in order to benefit from economies of scale and thus to decrease the cost of hedging (lower margins for instance) may give rise to adverse selection and to a too large to fail 106 systemic risk.
2 3 CME stands for the Chicago Mercantile exchange; CBOT, the Chicago Board of Trade, NYMEX, the New York Mercantile Exchange, and COMEX, the Commodity Exchange. It has been suggested that the first mover advantage that LIFFE had in the Bund contract was lost due to the introduction of various factors that moved the liquidity to the DTB. Cantillon and Yin (2008) state, The story the data tells is one of horizontal differentiation and vertical differentiation through liquidity. As a result, DTB attracted a different set of traders than LIFFE, and those traders contributed to the market share reversal.
may also be tempted to enter new activities by subsidizing the costs of these by the revenues generated by the existing business: this could give rise to new and uncontrolled risks. One could possibly conceive that certain CCP characteristics will tend toward naturally occurring monopolies as economies of scale make it difficult for new firms to compete, but not in a traditional sense. Normally one considers economies of scale arguments in the context of efficiencies in production, apropos to a manufacturer. For exchanges and clearing houses the issue is not one of lowering costs but of provision of liquidity and efficiency for the user. Once a provider has achieved a position of providing the most efficient service (perhaps by providing better technology for the user), then that provider is likely to achieve a scale of liquidity that attracts more users. Unless the provider allows a rival to introduce a product that is clearly more efficient, the liquidity in a contract (for an exchange) or size of clearing (for an OTC derivatives clearer) makes the firm most desirable for the user. Thus, in this discussion we use the concept of economies of scale to denote barriers to entry, which most likely will come from best technology to service the client and/or the benefits of being a first mover (attracting a large number of users). This is not to say the benefit of being a first mover or most technologically efficient will prevent competition from unseating a CCP or exchange from its position, but doing so is not easy.3

Also, different products and venues offered by the geographical diversity of a horizontally integrated group may also give rise to other types of risk benefits which supports integration across geography and time zones such as diversification of income streams. This type of growth could, in theory, entail the introduction of other types of risk (noted below). The first and most obvious factor of risk reduction finds its roots in the economies of scale and scope offered by such a group. Scale benefits arise from increased transaction volumes and liquidity and which may benefit from common operations. In the same way, users may benefit from a network effect: the more members are connected to the CCP, the more trades may be netted. Besides, the centralization of trades in one CCP increases also multilateral netting, hence reducing the CCPs own credit exposure. But as in a vertical structure, horizontal integration does not only lead to a decrease in risk. First, the direct consequence of horizontal integration is the possibility that the failure of a large CCP may have some spillover effects onto the markets where it operates. Second, the now larger CCP may face increased exposure and concentration risk if it serves the same users on different geographical markets as would be the natural tendency for any firm (reflecting marketing efficiencies). As such, markets in other geographical locations may be affected by a crisis in another market if the CCP in that market comes under pressure; further, if the initial problem is created by a client that has problems in multiple markets simultaneously the concentration risk may multiply the overall effect. These spill-over effects can create liquidity risk at the firm level which may also surface in a horizontally integrated CCP. If a CCP operates only on a domestic basis, it is bound to be a member of a given payment system or have access to its central bank facilities. These facilities may not be made available if the CCP operates in different jurisdictions, meaning that the CCP may have to rely on the commercial banking systems across boundaries. Liquidity is therefore critical at two levels: the user wants liquidity for trading purposes so an exchange or a clearer needs to be efficient in the provision of that liquidity and the entity needs to be able to provide that liquidity during times of crisis. Thus, liquidity availability at the firm level is also critical. Since liquidity is critical to the netting procedure, regulators must be cognizant of this fact, and regulation which impinges on liquidity, at either level, can quickly create or destroy one competitors advantage or disadvantage; as such regulatory issues (wedges to efficiency) may often drive the competitive model for CCPs. When discussing the competitive model in contrast to the concentrated framework, usual arguments are made. Competition should lead to reduced costs to the users and enhance innovation. Concentration leads to
higher costs and inefficiencies. On the risk side, competition offers more flexibility in the case of the failure of one of the CCPs as an alternative would readily be in place. As such, competition yields less systemic risks but this greater flexibility comes at the cost of lack of homogeneity in procedure and may increase operational risk. Typical to competitive industries operating under increasing economies of scale or scope is, however, predatory pricing. This so-called race to the bottom means that in order to chase away competitors, participants in the industry could bring down prices (i.e., lower margins or transactions costs) and put their operations at greater risk as they seek to gain market share. Since liquidity for a user is critical, pricing will not be the only issue, as we will see, to gaining advantage in the market. Competitive technology in a regulated environment means race to the bottom pricing is not so simple as a way to achieve market dominance. In this paper, we explore some of the issues, pros and cons, of a competitive model for CCPs versus a defined monopolistic model, whether in the form of a government sponsored entity (such as Fannie Mae) or a regulated monopoly. We also address some of the unintended consequences of the different approaches that may arise from a particular model. These unintended consequences that may arise from various regulatory solutions are those that concern us the most. To examine the various approaches we will use as a guide the insurance industry literature where apropos.
Literature
A CCPs risk management relies essentially on two tools: margins deposited by participants and contributions paid by participants to a default fund. Margins rely on defaulter-pays principles whilst contributions to a fund rely on a survivors-pay principle. In case of default by a participant, the margin deposited by the defaulter will constitute the first line of defense (defaulter-pay). If the margin is not large enough to compensate for the loss, the CCP will rely on the default fund to cover any remaining loss (survivor-pay principle). Usually, CCPs use both devices to manage their exposure risk.4 Although generally ignored in the relatively scarce literature on risk management of CCPs, a third buffer still exists in the eventuality of the default fund not being sufficiently funded to cover the losses, i.e., the CCPs equity. This equity is controlled by the holding company or owner of CCPs and, depending on discretionary decisions made by the board of the holding company, may be available to absorb losses incurred by the CCP and/or compensate CCP participants for the risks they are running, essentially operational and counterparty risk.
See Bliss and Papathanassiou (2006) for a thorough discussion of the various tools which can be brought into use under different theoretical situations and legal discussions of the usage of various techniques.
107
Various studies investigate how CCPs should set the margin requirements for their participants and rely on statistical or optimization models or on option theory [Knott and Mills (2002)]. Most of the CCPs use such sophisticated tools. One the best known methodologies used by CCPs is the SPAN system (a model developed by the Chicago Mercantile Ex5
classical insurance-taker. Still, classical results in insurance theory are that it is always optimal for an individual to take some insurance beyond a certain amount of deductible [Arrow (1973)]. The deductible must be such that the marginal expected utility of wealth, when losses are less than the deductible, per dollar increment in expected outlay is equal to the marginal expected utility of wealth, when losses are greater than the deductible, per dollar increment in expected outlay. This condition [derived by Pashigian and al. (1966)] relies thus on the insurance takers utility (and hence his aversion to risk) and on the level of his wealth. If one were to adopt a more modern approach and base rules of deci-
change), but it is far from being the only model used in the industry (see for instance the methodology developed by Eurex Margining AG).6 Very little has been written, however, on the optimal balance that a CCP should strike between margins and size of the default fund. To the best of our knowledge only Haene and Sturm (2009) cover this topic in a quite stylized model with underlying assumptions that may not be robust in a world of competitive CCPs and less than coordinated regulation in multiple jurisdictions. Essentially their assumptions are that only one default may happen per period, the distribution of default is uniform, and financial institutions are risk-neutral. They also assume that there is a fixed cost paid by the users of the exchange intended to recover the operational costs including the desired return on capital. Still, in this simplified framework, some interesting lessons are to be learned. First, the authors conclude that it is always optimal to set up default fund and that in some cases setting up a large fund is sufficient to cover all risks. Second, margin requirements and hence a smaller default fund are indicated if the opportunity cost of depositing some collateral is lower than the probability of a participants default. Besides, a margin policy becomes more attractive if it is associated with some other risk-mitigating measure, i.e., higher margins reduce participants default probability.
7
sions on a Basel 2/Solvency 2 framework, one should probably solve an optimization problem in which one determines the level of margin, default fund, and equity necessary to minimize the probability of default (for the CCP), subject to a set of constraints. They could be a desired return on equity for the CCP as well as constraints describing the way users react to the level of margin and contributions to the default fund. Haene and Sturms assumptions could certainly be used in a first step; however, once again they should be relaxed in order to allow multiple defaults per period to cover systemic risk, as this is much more likely to be the scenario that creates the larger problems. Or put another way, the Haene and Sturm simplified model has assumed away the most severe systematic problems, about which CCPs clients, CCPs, regulators, and central banks should be worried, namely contagion effects that can disrupt whole payment systems.
Building the examples: CCP

Based on the literature noted above, the CCP in a competitive marketplace essentially provides its service to participants who trade between margin costs (deductibles in the insurance literature), transactions costs (fees per transactions), and assurance of performance (credence good).8
Although relaxing Haene and Sturms assumptions is beyond the scope of our contribution, one may assume that allowing for multiple defaults (i.e., systemic risk) may reinforce the case for a default fund. On the contrary, relaxing the risk neutrality assumption and allowing for risk-aversion should reinforce the case for larger margins. Also, allowing for long tail events (i.e., doing away with the uniform distribution of losses) should intuitively plead in favor of a default fund (and equity) large enough to cover extreme events. Haene and Sturms results are, however, more or less in line with the insurance literature. Indeed, one could compare the problem faced by a clearers user to that faced by an insurance taker. The latter faces the choice of subscribing to an insurance contract that does not cover the first losses, i.e., the contract includes a deductible. This is similar, although not identical, to the problem faced in the CCPs case. The margin could be compared to a deductible and the default fund to an insurance main fund. Both cases are naturally non-identical since in the CCPs framework the clearer makes users pay to protect its equity us108 ing defaulters and survivors pay principles, which is not the case for a
5 6 7
See the following link for a report by the CME on SPAN: http://www.cmegroup.com/ clearing/risk-management/span-overview.html. See Risk Based Margining from Eurex (regularly provided brochure) at the following link: http://www.eurexchange.com/download/documents/publications/rbm_final_en.pdf Margin requirements for exchanges and CCPs open up interesting issues that are often ignored. Regulators, not the CCPs, tend to set what types of marginable securities are acceptable. The recent issue of all government bonds in Europe having the same risk weighting for financial institutions inside the European Union can have serious knock-on effects. If a risk management system treats Greek bonds as the same risk as German bonds then clearly these could be a favorite for posting as collateral. But clearly, not all collateral is the same regardless of the credit risk rules set by the regulators. Would the CCP, if left to its own devices to set margins consider them the same or would, perhaps, the corporate bond of a major French, British, or German company be considered more appropriate for collateral than certain government bonds. Since this issue will reach across borders and across currencies the issue of what is a risk free rate for a particular currency comes into play. With exchanges and clearers in multiple locations competing with similar products the currency of denomination of the exchange can have a wide effect on acceptable collateral. The credence good issue is widely understood in the literature and has many parameters for discussion. In its most simplistic form it is apropos to our discussion of provision of CCP products. See Anderson and Philipsen (1998), Dulleck and Kerschbamer (2006), and Nelson (1970), for more general discussions of the concepts applied here.

In our initial example we do not assume legislation or regulation and it is assumed that our company, CCPC1, (competitive market C and 1 for company 1) must decide between the various trade-offs in attracting clients. To simplify matters let us assume that all product offerings are similar, so as not to confuse matters at this point with product differentiation. Thus, as with any market, the profit maximizing position will be where marginal cost of production equals marginal revenues, as the firm is essentially a price taker. Thus, the issue to examine is on the cost side. From the literature above, clearly the marginal consumer will value different capabilities differently; that is, will the firm be around during a crisis to perform (the credence good issue), costs per transactions, and the opportunity cost for having to deploy capital in margins. Thus, in building our CCPC1s approach to each of these conditions, consider the costs of each of the three issues weighed by the consumer while minimizing costs on the margin. 1. Capital of the CCPC1 Ceteris paribus, the per share profit maximizing position is to keep equity capital to a minimum while providing assurance of performance. Debt in the capital structure would logically be considered a positive condition, as long as it was subordinate to performance of the contracts (a position most debt holders would not like, unless fully compensated). Alternatively, CCPC1 could have a second layer of protection for the user, a sinking fund or a set of guarantees provided by the members of the CCP. We may call this member reserves. From the consumers perspective, this set of guarantees, where the members of the CCP shared the losses associated with non-performance, provide a backstop to the minimal equity that the clearer or exchange has at its disposal given non-performance on a contract. Where the guarantee is located in the capital structure (whether senior or junior to equity for a particular loss) will drive equity versus debt (guarantees or sinking fund issues) considerations for the firm and the consumer of the product. 2. Margin requirements CCPC1 can set margin requirements for the contracts it offers to clear. Typically the requirements are based upon some measure of the risk of a price movement of an underlying contract. For our example we are assuming that the contracts themselves, less transactions costs and bid-offer spreads, are zero net present value (not options although our analysis will both utilize and consider options theory). Margin requirements initially will be assumed to include both initial and variation margins without differentiation. Note that for our discussions at this point, we ignore the issue of what securities are marginable and the cross-border currency (numeraire risk free rate) issues. 3. Transactions fees transactions fees are not simply determined in the competitive market by the marginal supplier, since it is only one aspect of the cost to the consumer. The consumer is interested in performance under stress (credence good issue) and total cost per
transaction as well as intangibles that are defined by the business model of an exchange or a clearer. Transactions fees will be valued vis-a-vis the cost associated with margin requirements theoretically; however the actual transactions costs will be driven by factors outside the control of the CCP itself, such as the brokers, banks, or futures commission merchants business models. 4. As such the consumer does not interface directly, in most instances, with the CCP, but may do so through some intermediary party that is setting the final pricing, reflecting margins, capital structure, their business model, and so forth. For our purposes the primary issues can be traded off one for the other in creating the final price to the intermediary and thus the consumer. Each issue will be important to the consumer in calculating the cost that they are paying and what aspect is more important than another in weighing the cost in its entirety. Borrowing from insurance literature [Raviv (1979)], the consumer will trade-off cost and risk or deductible in a casualty insurance product assuming all providers are the same. For life insurance the issue is not quite as clear. Performance at time of death becomes critical, since the deductible trade-off is less obvious (one can be underinsured at time of death and this could be considered a large deductible in comparison, but is not useful for our examination of CCPC1). The assurance of performance is critical, however. Thus, the credence good issue becomes extremely important for the purchase of life insurance, especially so the younger one is. One may argue that credence good issues play a part for major catastrophic insurance issues such as hurricane insurance or earthquake insurance but less so for automobile insurance. Thus, in providing insurance and pricing some providers will choose to present themselves as extremely prudent, secure (high credit rating), able to meet any contingency (the earthquake hits Tokyo or a hurricane crashes New York); others will specialize in providing products to those who are more interested in insuring smaller risks that do not have systemic or extreme (from the point of a call on the insurance) potential losses. Equivalent issues are faced by CCPC1. They are the same choices faced by the insurance provider. Higher costs due to the cost of capital needed to protect against systemic risk versus the lower cost but less secure providers potential competitive pricing. Ceteris paribus, less security implies a lower transactions cost is possible. This trade-off can show up in lower equity for the CCP and lower guarantees by the members of the exchange in case of defaults or operational risks. In Duffie and Zhu (2011), where the trade-off between equity and margins for the CCP is addressed to some extent, the credence good issues were not fully considered. 109
Trade-offs between equity and margins

For our CCP, the earthquake in Tokyo may be equivalent to a financial crisis such as the Lehman default, whereas the property and casualty (car accident) equivalent is a daily occurrence with a predictable volatility. In one case, the tail or extreme event is not that important because it is assumed that it will not affect the participants in the CCP. Or, that the risk of this occurring is so slight that the product user is willing to accept failure or is not willing to pay for the credence good cost (higher equity needed by the CCP which necessitates higher transactions costs or sufficient margin to cover any eventuality), and the shareholder is willing to accept failure of the firm (in an efficient widely held portfolio and a Modigliani Miller (1958) world such a choice for the shareholder is certainly one possibility to maximize expected shareholder value). Thus CCPC1 could choose to keep costs down to the consumer and increasing sales by maintaining little capital or offer assurances by keeping large amounts of capital, either in the form of guarantees or equity. The capital structure will drive costs for the user on the margin. (Race to the bottom pricing is now more complicated a path to dominance even in our simplistic model of the industry.)
in the CCPs capacity to face a credit event if the margin system was wiped away. To the user, the cost of using the CCP is C = Cm+Ct whilst the CCP only receives Ct. The user will maximize his or her expected utility, which generically will present three possible states:

No default occurs and the user only loses C. A default occurs but the margin/capital structure compensates her and she only loses C. A default occurs but she is not entirely compensated for the loss because the capital basis of the CCP is not large enough to cover the credit event.
In a competitive market, the users decision to seek hedging in a particular CCP will depend on the mixture of C and E and on a level of risk aversion. As for the CCP, it will have to choose a particular mixture of E, Ct, and Cm in order to maximize its return on equity knowing that each possible combination will attract some users but turn away some others. Let us denote Q the optimal amount of transactions for the CCP.
The second place that the CCP can make this trade is in setting margin requirements (variation and initial are not differentiated). High margin requirements limit the possibility of losses due to the failure of a contracting party such as Lehman but also entail costs to the consumer. Most of the literature has focused on the mechanism for setting margins, thus the VaR-like mechanisms for calculating the best way to set margins to trade-off ongoing trading with protection against tail events are common. There is no doubt efficiency in setting margins (not requiring more than is needed) is something every CCP must strive to accomplish. This document does not address efficacy of any particular system; instead it is focused on the trade-off of margin requirements as an alternative cost to the consumer versus higher product costs due to needed return on equity (or to pay for guarantees). Higher margin requirements may eliminate the risks of extreme movements but eliminating the extreme loss situation for the CCP via high margins increases the costs to the consumer. Finally, pure transactions costs per trade can be set along a scale. Driving this may be several factors, in particular the return on capital needed to cover operational risks and therefore the return on equity (clearly if all risks are eliminated through high margin requirements then equity levels are smaller and return on equity can be lower with lower transactions costs). In this analysis we assume that operational costs are more than covered by transaction costs and we shall not consider those costs any further. The CCP offers to the user a contract according to which she has to deposit a margin (M) entailing an opportunity cost Cm, a share of the transaction cost Ct covering the risk the of failure (i.e., not the operational 110 costs), and a capital structure E giving the user more or less confidence
9 This is particularly the case for arguing that OTC derivatives migration to a large CCP would require concentration to ensure that sufficient numbers of possible counterparties would be covered so that margins needed to insure against systemic risk would be less than the present system implies.
Considering a competitive market, where Ct is given, the profit maximizing level of production is the Q that equates marginal cost with price. The firm can choose the mix of Cm, Ct, and E that supports profit maximization. To this end, the amount of equity chosen versus the degree to which margins are imposed will depend upon the marginal productivity of capital deployed by the firm versus the opportunity cost of capital for the participants in CCPC1 exchange. Duffie and Zhu (2011) focus on minimizing the need for Cm by creating a large CCP that contains a vast majority of all trades stating that the concentration into one entity of the trading parties for all OTC trades would result in decreased needs for Cm, since there would be sufficient offsets to decrease the margin that would be needed per trade. In the competitive model where CCPC1 trades, Cm is a variable that must be considered against the return on equity desired and/or the overall cost of equity plus guarantees (whether in a sinking fund arrangement or liability of the members or some other default arrangement to protect against tail events).
Making margins more efficient

As already noted, a good deal of the CCP literature is focused on making margins less costly.9 As such it ignores the equity issue and focuses upon the buyer of the clearing capacity whether OTC CCPs or exchanges.

Concentrating all or nearly all contracts traded into a central CCP, it is argued, can lower the cost of margining as trade concentration reduces risk. Duffie and Zhu (2011) argue that with some 85 percent of OTC derivatives concentrated in the major players hands, forcing them to clear in one CCP would reduce costs significantly. Implicit in that analysis is that the CCP already has the most efficient system for measuring risk available. For CCPC1, no matter the path chosen in trading equity versus margin costs it behooves one in a competitive market to make the margining more effective. In a non-competitive market it is not as clear where the concentration will be. Lack of competition can cover many sins in margining efficiency. Sins in margining may come from the regulator with unintended consequences. As noted previously, choice of security which is allowable is often not the decision of the CCP or the distributor of the product (bank for OTC derivatives or the futures commission merchant and brokers for a futures product), but rather the acceptable security is defined by an authority or regulator or with horizontally structured CCPs even multiple regulators. Further, how security placed with a clearer is handled can be driven by regulation: Singapore has regulations on mingling funds that are different from those in the U.S.
obvious that once one is concerned about the probability of ruin, assuming a wrong distribution of volatilities and correlations will entail a wrong mix of margin and capital. Moreover, we shall not recall here how the choice of a normal distribution has all too often meant underestimation of risk for many financial instruments or institutions. The same should go for any CCP. But besides this well-known feature of the normal distribution, and since many risk management system are based on a normal VaR, one should also be fully aware of another feature that comes with it. Indeed it is well known that VaR is a non-subadditive measure of risk with one exception being when the VaR is computed using the Gaussian distribution. If one accepts that VaR based on normal distributions consistently underestimates risk in the post-Lehman environment of risk on - risk off trading [Putnam (2012)], then one may infer that the likelihood of potentially mispricing the amount of margin necessary to protect the CCP is high. Since more and more CCPs are members of vertically or horizontally integrated groups, this means that the risk management process within a CCP should manage risk on a aggregated basis, i.e., looking at a specific counterparty that may be using the CCPs services at different levels of the group. (Mis-)using a VaR measure that is sub-additive or not could make a huge difference in the way the risk is estimated.10
Making efficiency choices

What is an efficient margining system will vary, however, on how one determines efficiency from a profit maximization perspective. Since margining will be determined to a large degree by the assumptions of the distributions of risks faced by the CCP, arguing that there is one and only one efficient system may not be possible. Indeed if CCPC1 decides on being the credence good provider in the system, it may argue for a margining system focused upon tail disruptions (we will use VaR for lack of a better term) and thus define the margin to be set resulting from this type of distributional assumption. It can then choose to handle the earthquake equivalent situation via a higher level of equity or by imposing margins that are larger than its competitors to compensate for that risk. From the insurance literature these customers prefer safety over cost, On the other hand, CCPC1 may choose not to be a best quality provider and be willing to face failure resulting from the earthquake since it is intent on capturing the business that does not value the extra protection provided by higher margins or greater equity. In this case, being the low cost provider through a VaR calculation that assumes that the financial panic of 2008 cannot occur again or that the shareholders are willing to accept failure argues for making normal non-disrupted distributions of risk assumptions. Further, it also argues that this firm would also choose minimum equity. As an important aside and although the technical features of the risk management system are not the immediate topic of this study, let us stress how important the choice of the risk-return distribution is. It is
10 Interestingly, the recent decision by the European Commission to block the tie up of Deutsche Brse and NYSE Euronext will no doubt have implications for horizontal clearing house and exchange structures.
Margin efficiency for the credence good provider

CCPC1 may be willing to compete as a high cost provider if it can segment the market. If there is a demand for services from a provider that is well positioned to function during a significant meltdown in the system, then higher margins could be accepted. In this case, equity decisions held neutral, CCPC1 may choose higher margins to compensate for the assumption of a fatter tailed risk distribution and/or a risk distribution that is not well defined by its two first moments. As such, the cost of margin to the customer may be higher, but the customer may be willing to bear that cost if it is perceived that CCPC1 can survive the most difficult situations.
coverage over potential losses, and as in the insurance literature longer dated instrument users may be more attracted to the provider which is assumed to be the most substantial. Triple A insurance companies traditionally are higher cost providers for the same policies; they also tend to be most competitive with high end, longer term clients (whole life insurance being the example of a product where these issues seem to matter the most).
111
CCPC1 by choosing this route would still, however, be trading off higher cost margins, or a different margining process, for equity. In a competitive market this could be one way of creating some price discrimination through a perception of safety. Alternatively, CCPC1 may choose to hold a higher level of equity and guarantees thereby keeping margin costs for the client down to be competitive but sacrificing a return on capital deployed. Further, CCPC1 may find that being a good citizen is beneficial to their long-term client relationships. For example, the move by the CME Group to set up a fund of U.S.$100 million to help protect farmers against another MF Global type problem in February 2012 is an example of a firm enhancing the safety of its product even though it was not specifically responsible to clients of MF Global.11
Second, assuming regulations are not constraining to eliminate choices for the CCP, intermediaries will seek relationships with the CCPs depending upon their needs with respect to their customers. If their customers are simply price driven, then CCPC3 will look attractive, and so forth. Third, horizontal structures will create potentially different regulatory environments that may negate the credence good value created in one jurisdiction in another, thus the business models of the integrated firm may become country, numeraire, and regulatory specific. Such a factor will affect the ability to efficiently cross-margin or net, imposing a cost on the CCP user, reducing efficiency in all cases.
Suppose a monopoly
Competitive CCPs and choices
We introduce CCPC2 and CCPC3 at this juncture to this market as juxtapositions to CCPC1. We have allowed CCPC1 to have multiple possible choices in the marketplace for trading equity against margins, for choosing to be a credence good provider, or choosing to be a profit maximizing entity which assumes that it may fail in the event of a major meltdown in the financial marketplace. In this market we will assume CCPC2 is the belts and suspender high credence good provider by both having a great deal of capital while also charging higher margins to ensure to its participants longer dated security for their transactions. CCPC3 will take the other extreme. It will attempt to maximize profits by seeking a higher return on capital by holding as little capital as possible and being competitive on margin requirements and by not protecting against the potential of a failure in the system. CCPC2 and CCPC3 have thus mapped the extremes of the trade-offs available for a CCP. One is the low cost, low quality provider and the other the high cost high quality provider. From the insurance literature we know that the consumer will consider both depending upon their needs and their own risk aversion. For the longer dated contract, which are the most likely of concern for those which are risk averse, CCPC2 may obtain clients; and, for the least risk averse clients and most price sensitive, especially for short dated contracts, CCPC3 will attract the business. The question becomes, where CCPC1 will choose to compete. No matter the choice, it still has to make the trade-off decision of equity (equity plus guarantees) to margin. Several observations can be made. First, a regulatory system that treats CCPC1 the same as CCPC2 and CCPC3 clearly defeats competition. The trade-offs become more limited for the CCP. The business model is narrowed, and from the perspective 112 of the marketplace, choice would be limited.
11 See http://www.chicagobusiness.com/article/20120202/NEWS01/120209959/cme-creating100-million-protection-fund-for-farmers-ranchers. 12 The failure of MF Global in November 2011 highlights the impact a clearing member can have on the system. In this case, trading on the exchanges where MF Global was an important factor overall liquidity, as measured by trading activity, declined. For example, the daily volume on the CME was down during the period. Most notable and significant, the failure of MF Global had no systemic spillover effect that could be detected. Various mechanisms protected the clearing houses from significant disruptions while the disposition of client funds were being sorted out. In comparison, the failure of Lehman Brothers had widespread systemic implications. The global nature of the effects were more akin to the concerns raised earlier about horizontal integration of an exchange or CCP and spill-over effects.
Much of the literature noted argues for a concentration of risk into one CCP. This seems to be favored by many regulatory bodies and supported by their research [Haene and Sturm (2009) and Russo et al. (2002)]. From the regulators perspective, the desire to ensure that no CCP fails if one of the members or one of the major financial houses in the OTC market fails is understandable.12 This work is geared to our gaining a better view of the potential for gains for reduced margin costs for a single CCP, which has at its disposal a large percentage of the contracts outstanding, thereby supposedly better insulating the non-failing institutions from the failing ones. Thus, the logic is that the likelihood of a systemic problem developing la 2008 with the default of Lehman Brothers is reduced as the risk is moved to the CCP, which insulates the system. Such a CCP would theoretically be less costly, since the size of the venture provides efficiency through economies of scale with respect to risk information and knowledge of positions and potential netting, especially for longer dated derivatives. Heller and Vause (2011) show that for certain combinations of OTC derivatives markets margining can be reduced by up to 15 percent if both credit default swaps and interest rate swaps are netted by one centralized CCP with the credit risk now being transferred to that CCP. Other studies come to similar conclusions. Essentially, assuming a portfolio effect could be sufficient to argue for such a result, if correlations remain stable (a most tenuous assumption which is often ignored and is unlikely to hold when it matters most a systematic disruption with significant market contagion).

The problem with this logic is that it is assumes such a monopoly will still act as if it were in a competitive market, as per above. Suppose it does not, but rather acts to maximize profit from its monopoly position. The logical conclusion is, as for any monopoly, that the price will be higher than in the competitive market place above and the gains due to size will be taken as abnormal profit. For example CCPC1, which we will now call CCPM1 (M for monopoly), without competition to maximize profits will hold as little equity as possible, thereby increasing the return on capital, while carrying as much debt as is feasible in the capital structure. It would easily be able to compensate lenders or members via a sinking fund by decreasing the quantity of clearing being provided and raising prices to match the demand at that lower position of production. Ceteris paribus with respect to margins, it could keep them similar to that prevailing for CCPC1 in a competitive market but simply raise the fees per transactions so that the marginal revenue equaled marginal costs at the higher price with the lower production of clearing activities. Alternatively, it could keep fees lower, reduce the overall capital protecting the system, and raise margins sufficiently to protect itself. In this case CCPM1 can trade margin protection for equity capital and fees to find the most efficient position for profit maximization and equity price enhancement. Minimal equity, relative to the capital needed for protection, would be a logical starting point. Since the benefit of concentration of all contracts into one CCPM1 has been shown to reduce the size of margins theoretically, ceteris paribus, it is logical that these gains will be surplus profits, even if margins and capital and transactions costs are no more than in the competitive examples above. Thus, the minimum gain in potential efficiency, due to concentration, is well defined. The profit maximizing position, however, suggests higher prices (no matter how it is achieved) for the client and a much lower production of derivative contracts than in the competitive model. One could argue that this should then be a regulated monopoly or that it should act as a Government Sponsored Enterprise (GSE). As such the argument could be that it would position itself according to the regulated monopoly price and quantity point. In this case, one still finds that costs are higher than in the competitive model as per that literature. But, it could be argued that intervention by the regulators may compensate and the monopoly CCPC1 would not seeks gains from its position.
monopolies also behaving in a manner consistent with utility maximization for the management and directors. From the public choice literature, empirical results suggest that government functionaries maximize their own welfare. For the monopoly CCPM1 this may mean supporting political ends instead of systematically providing low cost clearing relative to the competitive market or maintaining an income consistent with that of a monopoly. A possible example of such (mis)behavior could be the temptation to decrease the margins for derivatives on government securities in order to artificially enhance their liquidity and hence their cost. Regulation of bank capital in the Eurozone clearly supported bank purchases of government debt. Is one to assume that the market actually believed that German and Greek debt warranted the same capital allocation for a leveraged entity such as a bank! The unintended consequence of creating a protected monopoly is often not known until much later, after a major bubble or crisis is created. It was certainly logical that BIS rules on capital allocation would reduce bank risk if banks bought AA or AAA credits instead of making local loans; and, by making the rules universal it could make banks around the world work on a level playing field. With competition among credit agencies the credence good argument would have logically tiered the rating agencies as new ones entered the industry with looser rating criteria, la CCPC3 above, with its provision of ratings, while others sought to always be right no matter the situation ( la CCPC2 choosing to maintain an extremely high level of protection against failure). In our regulated world of the past two decades government rules have helped create a semi-monopoly for the rating agency industry through many different regulations set forth by entities from the SEC to the BIS. As such, the unintended consequence was that the profit maximizing behavior was not that of a credence good provider, but rather one which would risk failure to generate fees. Besides, it appeared the agencies failure to address the risks in SiVs, CDOs, and the mortgage-backed market, had very few negative consequences for them. The regulatory mandates created barriers to entry and competition; supply tiers based upon the notion that a competitive market position for the rating agencies would lead to several firms (from the most conservative to the other extreme with some agencies in between), and thus the need to behave as a credence good provider. With a semi-monopoly status created via regulatory fiat the need to be a credence good provider was eliminated. It is logical to expect that the monopoly granted to CCPM1 would lead
Public choice, GSEs, and unintended consequences

Counter to this argument is the evidence from Freddie Mac and Fannie Mae in the housing market, utilizing their semi-government status to grow their balance sheets in manners (reflecting the prevailing legislation to increase housing availability) that would lead to a serious financial disequilibrium. There are many other examples of government created
to behavior that was suboptimal from the perspective of costs and efficiency. And as such, it may be that the monopoly will actually increase, not decrease, systemic risk if history is to be considered. Given that the objective of centralizing CCP activities is to decrease the potential for a systemic problem, the evidence from past efforts suggests that systemic risk may actually be increased. 113
Moral hazards, adverse selection, and the market place

Moral hazard concerns cannot be ignored, as it would appear at two levels if a monopolist solution were to be favored. First, and quite obviously, the too big to fail feature would be become omnipresent in such a framework. Financial history has shown us, repeatedly in the past and again only very recently, how the managers of too large to fail institutions can take excessive risks (see above) and, hence, put the existence of the firm they are running in peril. Besides, the existence of such institutions usually ties the hands of public authorities in times of stress. When dealing with the possible failure of a large and indispensable financial institution especially if the latter spreads its activities over several jurisdictions public authorities are all too often unable to take the decision to let the institution wind down for fear of total disruption of financial markets. Second, by eliminating the risk in the OTC market from derivative trading and transferring it to the monopoly CCP one clearly reduces the need for credit due diligence by the providers of OTC derivatives.13 Understanding why CCPs work in the futures markets or the securities markets and may or may not be effective in the same manner for the OTC derivatives market is critical. The competitive advantage of the OTC derivatives market exists only where credit worthiness is known and accepted. This allows for more one stop solutions for risk management needs for those entities that have established credit. By creating a monopoly CCP the need for the major participants to monitor each other is reduced significantly, nearly to zero. Thus, as in the case of replacing typical C&I loans on the balance sheet of most banks with bundled AAA mortgage backed securities or SIVs, the need for credit knowledge is reduced (at least theoretically and from a regulatory perspective). This loss of institutional knowledge works fine as long as AAA risks are truly low risk. The incentive to the marketplace is (was) to concentrate the risk onto the backs of the rating agencies, as in the case of bank debt. With a centralized CCP, the incentive will be to concentrate the risk on the back of the CCP. And, as a monopoly with only limited competition, why not expect a problem to develop similar to the one that did following BIS rulings on bank capital allocations. Even more at hand is the example of problems with sovereign debt held by the European banks based upon capital allocation rules that treated Greece in the same manner as Germany. The unintended consequence was greater Greek debt held by European Union regulated banks than could be paid back. Combined with that of other debtor countries debt, this has again led to another unintended problem, i.e., bank undercapitalization for stressful periods. Said undercapitalization then raises the fears 114 of another systemic meltdown of the world financial system.
100000 90000 80000 70000 60000 50000 40000 30000 20000 10000 0 Oct-05 Oct-06 Oct-07 Oct-08 Oct-09 Oct-10 Apr-06 Apr-07 Apr-08 Apr-09 Apr-10 Apr-11
CME contracts
Inter-bank forward contracts
Source: Levich (2012)
Figure 3 Daily average trading volume, EUR futures and forwards (U.S.$ millions) (survey months only)
Where there are alternatives, as in a competitive market, for substitute products offered by different CCPs or exchanges consumers switch usage on the margin, thereby trading off among ease of use, safety, cost and so forth. An interesting example can be deduced from data provided by Levich (2012) (Figure 3), illustrating the increased use of FX futures contracts versus FX forward contracts during and subsequent to the 2008 financial crisis, with a continuing effect during the more recent European debt concerns. It is well recognized that the mere existence of a CCP increases the moral hazard due to the loosening of risk discipline by its users and we can only surmise that this problem would become more acute if the CCP were a monopolist. However, this case of moral hazard should also increase the adverse selection inherent to the functioning of CCPs. Indeed the CCPs, whether in a competitive or non-competitive framework, usually apply one-size-fits-all margin and trading fee policies irrespective of the end users credit quality. Being unable to distinguish which client is more at risk, the CCP applies an average pricing policy and by doing so, deters good-quality users from its platform whilst it attracts poorer quality users with a price that undervalues their true riskiness. We can fairly assume that a competitive industry composed of many CCPs would be in a better position, at least theoretically, to counter the effects of this adverse selection situation. CCPs or exchanges for derivatives in the futures and options markets operate in competitive markets and behave according to the rules of competition as discussed above. To be competitive they have to maintain the trade-offs we noted, but also the products are fundamentally different from the OTC market. Homogeneity of product is an essential feature that makes the system work at a low cost. By their very nature OTC derivatives are heterogeneous. It is their heterogeneous nature that makes them useful to
13 See Bliss and Steigerwald (2006) for a good discussion of bilateral clearing and the credit implications versus those for OTC contracts that may be cleared via a CCP.

those whose credit worthiness allows for their use. Could a consequence of a monopoly CCP for OTC derivatives be that the system would work better only if that standardization occurs? Could one unintended consequence be the elimination of competition from (for) the futures exchanges as all products are forced into a standardized bank driven derivatives market?
certain circumstances and for certain types of investors). We can expect that the users of the CCP will be price sensitive but regulatory driven behavior may affect the price sensitivity (and therefore the elasticities) on the margin. Banking systems regulations promote different agendas in different countries or jurisdictions. For these users and thus providers of contracts that would be cleared, the interplay of these regulations will affect sensitivity to price of clearing as conditions (regulatory or market) change. Since the CCP maximizes its profit under a constraint of not failing with a given level of probability, one should also take into account the fact that the higher this probability is, the higher the requirement of equity, fees, and margin will be. Depending upon the relative elasticities to prices and margins and hence how the probability of ruin reacts to changes in the different parameters, the optimal frontier of the monopolist will evolve accordingly, but may not be smooth.
Setting up of the monopoly

The monopolist CCP is free to set the capital structure it wishes as well as the margining and fee policy it deems better in order to maximize profit. As such, and ignoring how it invests the monies from its equity and default funds, its only source of revenue consists of the fraction of the trading fees that exceed operational costs. As a profit maximizer and for a given level of probability of ruin, it should intuitively increase the fees, decrease its own equity, and demand a high level of margin and default fund to guarantee safety at a low cost for the monopoly fund. If the CCP is a monopolist maximizing profits under a constraint of not going bankrupt with a given probability, first-order conditions that margins, transactions costs, and level of equity should meet are such that the ratio of change in profits due to an increase of $1 in each control variables to the change of probability of ruin for an extra $1 of the same control variable must be the same for fees, equity, and margin. The optimal level of production would immediately follow from the vector of control variables satisfying those conditions. The main determinants of the level of optimal trades offered by the CPP will be the users elasticity of trades to fees and the users elasticity of trades to margin (which is an opportunity cost). The closer these elasticities are to zero, the more the monopolist CCPs production will diverge from the competitive models and the more the CCPs price will be higher than its marginal cost. Standard economics tells us that the discrepancy between price and marginal cost is inversely related to the price elasticity. One can fairly assume that the lower the users elasticity to fees, the higher the fees the CCP will apply, but (possibly) simply too the lower the equity that it should be willing to put at work (and this in order to maximize its return on equity) and therefore the higher the margins and possibly the level of default fund. The lower the users elasticity to margins, the higher will the CCP try to raise fees, default fund, and possibly its own equity (depending on the brokers reluctance to increase the default fund). The level of these elasticities surely depends on the users aggregate utility function but it is also largely influenced by external factors such as the magnitude of volatility on the markets (more volatility could mean more hedging activities or more trading by hedge funds) and the regulatory constraints (different jurisdictions impose or forbid the use of derivatives in
Maximizing shareholder value and risk taking by the monopolist

In analyzing risk behavior of the monopolist an alternative approach can be taken. The monopolist cannot likely exist without the support of government granting it monopoly status. Our monopolist example is therefore not one of a natural monopoly, but the monopoly CCP is created to an end. That is, it is most likely granted monopoly status to reduce the probability of systemic risk in the financial system. Doing so would be consistent with the Dodd-Frank legislation in the U.S. and similar efforts in Europe. If the monopoly then exists due to government support, it clearly cannot be allowed to fail. Authorities will enter to bail out the CCP in case of insufficient equity. In effect the CCP shareholders have a put on the authorities, written when the monopoly is created. Like the moral hazards noted above, distortions due to its existence will occur. One then asks, what is the optimal level of risk to take for the monopoly CCP? We will assume that shares are widely held in portfolios thereby allowing portfolio theory to guide us as to the demanded behavior of the CCP. (If the monopoly CCP were closely held the diversification factor would not be in effect thus analysis of the option would be less clear cut14). From a
14 Suppose the shares are not widely held but concentrated in one entity or individual where the monopoly has been granted. In this case the share price per unit of risk taken is a different issue whether the entity is bailed out or not in the case of failure. Failure for a closely held stock would damage the owner and would therefore be guarded against, bailout or no bailout. Seeing monopoly profits, controlling risk of failure would be primary. Risk would not necessarily be sought to an extreme since there would be much to lose. Of course, in reality tax laws and regulations would affect the behavior but the general issues would remain the same. Politically it is almost impossible to see how a monopoly could be granted to one family, closely held firm, and so forth, by a Western government. Thus, it is more likely that a government created monopoly GSE would have its shares widely held and thus the implications for the to big to fail syndrome would promote (truly on the behalf of the shareholders) risky behavior.
115
shareholders perspective, the optimal amount of risk to be taken is one that maximizes the value of the put on the authorities. It follows logically, then, that the shareholder will demand that management take more, not less risk, otherwise the put is not fully valued. Given that this option exists for management it is in their interest to take more risk than would be the case in a competitive market. For example, the capital structure would surely be highly leveraged with risky assets (MF Global, with its speculative positions on European sovereign debt, even though a nonmonopoly entity in the space comes to mind). The existence of the put on the authorities will imply that risk is underpriced, possibly thereby increasing the actual systemic risk potential. Ironically, the desire for a monopoly to counter against systemic risk problems created by the OTC derivatives market could increase the risk of failure of the CCP designed to solve the initial systemic risk concerns. And, such a failure could spread to other financial entities and exchanges intensifying the systemic problem. It should be noted that during the financial crisis of 2008 the futures exchanges continued to function well, even as the OTC derivatives markets dried up (as illustrated by its gain in market share for FX derivative trading). Returning to our methodology above, the constraint of not failing is removed from the CCP in determining the mix of security to be held for contracts whether it is equity, fees, or margin. With this constraint lifted risk becomes less of a concern for the CCP. And for the financial system transference of risk to entities that are not pricing it sufficiently is always a vacuum that will be filled. If regulators push risk out of one place in the financial system it will move to another area. There are many examples of this from the recent crises in Europe, to the pre-euro currency bands, to the implied risk in the yield curves resulting from Fed moves in 2011. The monopoly CCPs profit maximizing position is to value the put on the authorities to its fullest; therefore, it will significantly under-price counterparty risk, providing the space where risks in the market may congregate, creating greater profits for the shareholder, but also concentrating system issues into one place. Thus, in an environment of heightened financial system stress the systems problems get squarely placed on the monopoly CCP. And, we argue that it is likely that the existence of this CCP structure in and of itself will elevate overall market risk (credit risk) in the global system just as centralized rules on credit did so for banks. The consequences of many decisions that led to the banking crises of the last few years were unintended and indeed many of the reasons the regulations were put into place were intended to create the opposite (less systemic risk). It is likely that a monopoly CCP would be created if and only if it were agreed by all parties that its existence would lower systemic risk. We 116 can, however, identify ex-ante (and surely will ex-post) that the actual
result could easily have the unintended consequence of creating more, not less, systemic risk problems in the future. Typical of any intervention into the private exchange of goods and services the unintended consequences are the ones that negate the good of the intervention by an authority.
Conclusion
Constructing a system of CCPs to lower systematic risk via overregulation may introduce many unintended consequences that may lead, in and of themselves, to systemic problems. Although theoretically a central clearing house may lower clearing costs for a region, say for Eurozone banks, how it impinges on the industry as a whole must be considered. From comparison to the insurance industry we find that actors will endeavor to find their competitive niche if left to do so and consumers of clearing services, banks, funds, individuals, and other market participants will align themselves with the clearing or exchange partner that fits their needs. Creating a level playing field via heavy regulation can and will (and we believe does) tend to lessen the competitive alternatives that may be taken in provision of exchange or clearing services. Further, focusing simply on one segment of the financial industry in mutualizing derivatives clearing risk could have spill-over effects onto other parts of the industry, say futures exchanges, that are heavily regulated as well. Experience with GSEs such as Fannie Mae in the U.S. and the standardization in bank risk via Basle I and II suggests that uniformity in rules may lead to consequences that are unpredictable and may increase, not decrease, systemic risk. Thus, we conclude that further regulation of clearing and exchanges must be carefully considered in the light of competitive CCP, in different jurisdictions, with less than coordinated regulators. Simplified models with unrealistic embedded assumptions may be of little use in providing policy guidance. The many unexpected consequences of any one solution to the systemic problems of derivative trading must be explored carefully. Solving one apparent problem will no doubt, based on recent experience, lead to others. In particular, the authors are concerned about moral hazard problems that are created due to mutualization of risk, making lenders, banks, and financial institutions in general less concerned about the risks being taken. Making too big to fail even bigger will, in the authors opinion, lead to bigger failures.

References

Anderson, E. S. and K. Philipsen, 1998, The evolution of credence goods in customer markets: exchanging pigs in pokes, Working Paper Arrow, 1973, Optimal insurance and generalized deductibles, Rand Corporation, R-1108OEO, February Bank for International Settlements, 2007, New developments in clearing and Settlement arrangements for OTC Derivatives, March Bank for International Settlements, 2010, Market structure developments in the clearing industry: implications for market stability, report of the Working Group on Post-Trade Services, November Bank for International Settlements, 2011, Payment, clearing and settlement systems in the CPSS countries, Committee on Payment and Settlement Systems Bernanke, B., 1990, Clearing and settlement during the crash, Review of Financial Studies, 3:1, 133151 Bernanke, B., 2011, Clearinghouses, financial stability, and financial reform, speech given at the Financial Markets Conference sponsored by the Federal Reserve Bank of Atlanta, Stone Mountain, Georgia Bliss, R. and G. G. Kaufman, 2006. Derivatives and systematic risk: netting, collateral and closeout, Journal of Financial Stability, 2, 5570 Bliss, R. and C. Papathanassiou, 2006, Derivatives clearing, central counterparties and novation: the economic implications, European Central bank, working paper Bliss, R. and R. Steigerwald, 2006, Derivatives clearing and settlement: a comparison of central counterparties and alternative structures, Economic Perspectives, Federal Reserve Bank of Chicago, 30:4, 2229 Cantillon, E. and P-L. Yin, 2008, Competition between exchanges: lessons from the battle of the Bund, Working Paper Darby, M. R. and E. Karni, 1973, Free competition and the optimal amount of fraud, Journal of Law and Economics, 16:1, 67-88 Duffie, D. and H. Zhu, 2011, Does a central clearing counterparty reduce counterparty risk? working paper, Graduate School of Business, Stanford University Dulleck, U. and R. Kerschbaumer, 2006, On doctors, mechanics, and computer specialists: the economics of credence goods, Journal of Economic Literature, 44:1, 5-42 EuroCCP, 2010, Recommendations for reducing risks among interoperating CCPs, European Central Counterparty Ltd. European Central Bank, 2009, Credit default swaps and counterparty risk, European Central Bank, Financial Stability and Supervision
European Commission, 2010, Proposal for a regulation of the European Parliament and of the Council on OTC Derivatives, Central Counterparties and Trade Repositories, European Commission, Brussels Haene, P. and A. Sturm, 2009, Optimal central counterparty risk management, Swiss National Bank, Oversight, Berne, Switzerland Heller, D. and N. Vause, 2011, Expansion of central clearing, BIS Quarterly Review, June Knott, R. and A. Mills, 2002, Modelling in central counterparty clearing houses: a review, Bank of England, Financial Stability Review, December Kroszner, R., 2006, Central counterparty clearing history, innovation, and regulation, Speech given at the joint European Central Bank Federal Reserve Bank of Chicago conference, Frankfurt, 3 April Levich, R., 2012, Currency forwards, currency futures, and the impact of the global financial crisis, NYU Stern working paper, February Lo Giudice, S., 2007, Settlement internalization: the production and distribution of services in the (clearing and) settlement industry, Journal of Financial Transformation, 17, 127 - 141 Mayers, D. and C. W. Smith, Jr., 1987, Corporate insurance and the underinvestment problem, Journal of Risk and Insurance, 1, 45-54 Modigliani, F. and M. H. Miller, 1958, The cost of capital, corporate finance and the theory of investment, American Economic Review, 48, 261-97 Nelson, P., 1970, Information and consumer behavior, Journal of Political Economy, 78, 311329 Pashigian, B. P., L. L. Schkade, and G. H. Menefee, 1966, The selection of an optimal deductible for a given insurance policy, Journal of Business, 39, 35-44 Putnam, B. H., 2012, Volatility expectations in an era of dissonance, Review of Futures Markets, March (forthcoming) Raviv, A., 1979, The design of optimal insurance policy, American Economic Review, 69:1, 84-96 Russo D., T. Hart, and A. Schnenberger, 2002, The evolution of clearing and central counterparty services for exchange-traded derivatives in the United States and Europe: a comparison, European Central Bank, Occasional Paper Series, September Trichet, J-C., 2006, Issues related to central counterparty clearing, Speech given at the joint European Central Bank Federal Reserve Bank of Chicago conference, Frankfurt, 4 April Tucker, P., 2011, Clearing houses as system risk managers, speech given at the Bank of England, at the Depository Trust & Clearing Corporation, 1 June
117
118
PART 2
What Have We Learned from the 2007-08 Liquidity Crisis? A Survey

1
Hamid Mohtadi Professor, Department of Economics, University of Wisconsin at Milwaukee

and Visiting Professor, Department of Applied Economics, University of Minnesota
Stefan Ruediger Assistant Professor, School of Business and Economics, University of

Wisconsin at Stevens Point
Abstract
The financial crisis of 2007-2008 was a liquidity crisis. Thus, we must both study the source of the crisis and evaluate the regulatory measures to address it. How was this liquidity crisis, and its associated risk, related to other forms of risk? What was the nature of the vicious cycle that produced the crisis? How did the regulators respond and have continued to respond? What are some quantitative dimensions of liquidity risk measures? We provide a critical survey of the existing literature in an attempt to answer these questions.
1 The initial version of this article was written when Mohtadi served as Quantitative Director (and VP for Risk Management) at U.S. Bank Corp while on leave from the university.
119
The financial crisis that began in late 2007 or early 2008, culminating in the collapse of several large U.S. financial institutions such as Lehman, Washington Mutual, Wachovia, Merrill Lunch, and AIG in 2008, and finally settling down in mid-2009, may have in fact had its roots in institutional failure, market failure, excessive risk taking, etc., as has been argued by numerous pundits and scholars alike. However, much of these discussions have focused on the explanation of what went wrong. As such, the policy implications that arise from these discussions relate quite appropriately to ex-ante circumstances, i.e., to improving the financial and regulatory environment in order to prevent the next crisis. However, expost, one is faced with a different question. That question is this: should the system fail again, what safety tools are in place to limit the scope of the disaster? The emphasis in Basel III proposals on banks building up adequate liquidity can be understood in this ex-post context. Thus referring to the 2007-2008 crisis, the Bank of International Settlements (BIS) writes, the crisis again drove home the importance of liquidity to the proper functioning of financial markets and the banking sector. Prior to the crisis, asset markets were buoyant and funding was readily available at low cost. The rapid reversal in market conditions illustrated how quickly liquidity can evaporate and that illiquidity can last for an extended period of time [BIS (2010)]. This paper is about understanding liquidity and liquidity risk, the first in terms of its ex-post properties in attenuating the impact of a severe financial downturn, and the second, in terms of the interaction between different dimensions of liquidity risk, notably funding and balance sheet liquidity risk among banks and other financial entities on one hand, and market liquidity risk among traders, on the other. Understanding this interconnection by bank managers is as important in being prepared for another financial crisis, as having a large amount of liquidity at hand.
1.150.000.000.000 1.100.000.000.000 1.050.000.000.000 1.000.000.000.000 950.000.000.000 900.000.000.000 850.000.000.000
Figure 1 Trends in tier-1 capital for all FDIC insured banks (2007-2009)
12,00 11,50 11,00 10,50 10,00 9,50 9,00 8,50
07
08
9 00 /2 30 12 /3
00
00
00
00
00
00
00
20
20
00
/2
/2
/2
/2
/2
/2
/2
1/
1/
/2
31
30
30
31
30
30
31
/3
3/
6/
9/
3/
6/
9/
/3
3/
6/
30
12
Source: www.fdic.gov and www.bankregdata.com
Figure 2 Trends in tier-1 capital relative to risk-weighted assets of all FDIC insured banks (2007-2009)
for all the FDIC insured banks in the U.S. for the period that just covers the crisis. A large dip occurred from the beginning of the second quarter of 2008 to the end of the first quarter of 2009. While this decline does not bode well for the promise of countercyclicality and thus immunity of tier-1 capital from the decline in assets prices and the rise in market risk, it should be noted that during this period a few banks exited the system. Furthermore, the institutions varied greatly in size over the period. For example, even among the top 10 banks, assets varied by a factor of over 10 (from U.S.$160 billion to over U.S.$1.8 trillion over this period). Thus, to glean a better sense of the actual changes in liquidity from the values of tier-1 capital, one should consider tier-1 capital relative to the size of riskweighted assets. In this way, liquidity is measured relative to bank size. In addition, banks exiting the system would create a drop in both the numerator and denominator, so that the ratio is less sensitive to this factor. The results are depicted in Figure 2. The decline in liquidity in this relative sense is certainly noticeable in Figure 2 as well. In short, to the extent that liquidity of banks was still somewhat tied to markets some procyclicality remained and arguably contributed to the spread of the crisis and the credit freeze that characterized the severest period of the crisis. Basel II had tried to address this question to some extent. For example, the Final Rule (Basel II) stated, A bank should incorporate liquidity risk into
Banks liquidity and the crisis: an overview

The focus on requiring banks to build adequate liquidity in order to deal with a potential crisis is not new. However, the intensity of the attention to this issue and the emphasis on its importance is new. The participants in this discourse include national and international regulators, banking experts, and academics, but the stakeholders include everyone. One fundamental question is whether a healthier and larger amount of liquidity on the part of the banking system would be able to reduce the severest impacts of another Great Recession, should there be one, by breaking the chain that constituted the vicious cycle of illiquidity-fire sale-illiquidity [Brunnermeier and Pedersen (2009)]? This question is relevant in evaluating the high bar that Basel III has set for liquidity requirement for banks, and is relevant for the European Central Banks current difficulties (at the time of this writing) in addressing the sovereign debt crisis. To the extent that the emphasis in banking liquidity has always been on tier-1 capital, it would be useful to examine the performance of tier-1 capital during the height of the financial crisis and to see whether it remained relatively iso120 lated from the deep market downturn. Figure 1 demonstrates this trend
12
9/
1/
20
09
3/ 31 /2 6/ 007 30 /2 9/ 007 30 / 12 200 /3 7 1/ 2 3/ 007 31 /2 6/ 008 30 /2 9/ 008 30 / 12 200 /3 8 1/ 2 3/ 008 31 /2 6/ 009 30 /2 9/ 009 30 / 12 200 /3 9 1/ 20 09 3/ 31 /2 01 0

the assessment of its capital adequacy. A bank should evaluate whether capital is adequate given its own funding liquidity profile and given the liquidity of the markets in which it operates [Federal Register (2008)]. However, several factors conspired to reduce the effectiveness of regulatory mechanisms in preventing a liquidity crisis. First, the implementation of the Basel II accord in the U.S. banking system was long lagging behind Europe and the first U.S. bank that began to fully implement the Final Rule (JP Morgan) did not begin to do so until January 2009.
2
be drawn upon when needed, and that under a cash flow approach in which the firm attempts to match cash outflows against contractual cash inflows across a variety of near-term maturity buckets. While insurance companies tend to follow the former approach, banks tend to follow the latter [BIS (2006)]. A key issue is the extent to which it is the parent versus the subsidiary units that provide liquidity to affiliates that are under stress and the implications of cross-affiliate transfers of funds and collateral across national borders, sectors, and currencies. Liquidity support from parents is less common in the insurance sector than in the banking sector, where the parent firm is a source of strength and ensures that there are resources at both the parent and subsidiary levels to address liquidity events as they arise. This is consistent with the increased centralization of liquidity risk management in banks [BIS (2006)]. To what extent was the decentralized approach towards liquidity risk management in the insurance industry linked to the downfall of AIG as a whole and thus the entire liquidity crises that ensued? To get an insight to this question, consider that the CDO and CDS agreements. These agreements were issued by a subsidiary of AIG, AIG Financial Products (AIGFP), using the AAA rating of AIG to provide ratings for the CDO and CDS agreements. While only some CDO and CDS agreements had to be paid because of the default of the underlying asset, this clearly highlights the existence of an agency problem where one entity was responsible for the liquidity risk of another entity that actually carried out the risky operations [Whitehead (2010)]. Moreover, AIG did have to shore up far more money as it lost its own AAA rating due to the above process. More specifically, the OTC transactions from AIG lacked regulatory oversight, for three reasons: 1) the Commodity Futures Modernization Act of 2000 essentially eliminated regulation for the OTC CDS and CDO transactions of AIG; 2) the subsidiary of AIG, AIGFP, was subject to regulation and oversight by the New York State Insurance Department, which specifically excluded the kind of OTC transactions AIGFP was engaged in from regulatory oversight and thus AIG was able to avoid having to fulfill reserve requirements for these transactions; and 3) the fact that AIG was not required to fulfill reserve requirements for its OTC transactions, essentially allowed them to offer CDS at lower costs than its competitors [Cunningham and Zaring (2009); Hazen (2009); Werrett (2009); and Whitehead (2010)]. At the core of the collapse of AIG were two issues: 1) AIG applied inappropriate risk models which predicted that AIG had a 99.85% chance to never having to make a CDS payment [Whitehead (2010)]. The risk models also failed to account for the possibility of systemic events and occurrence of large economic shocks. 2) AIG only paid attention to a CDS payout situation caused by the default of the underlying asset, loan, or bond, completely ignoring that in the case of a downgrade of AIG, it
Second, under Basel II the emphasis on stress testing came about late in the game and in fact in many U.S. banks the examination and implementation of stress testing did not begin until well into the crisis period. Third, while stress testing in Basel II focused on credit risk, the emphasis on the adequacy of liquidity under stress conditions was not sufficiently addressed at that time. This task was in fact postponed and relegated to Basel III, which has yet to be implemented in the US. To complement these principles [of sound liquidity management], the Committee has further strengthened its liquidity framework by developing two minimum standards for funding liquidity. These standards have been developed to achieve two separate but complementary objectives. The first objective is to promote short-term resilience of a banks liquidity risk profile by ensuring that it has sufficient high-quality liquid assets to survive a significant stress scenario lasting for one month. The Committee developed the Liquidity Coverage Ratio (LCR) to achieve this objective. The second objective is to promote resilience over a longer time horizon by creating additional incentives for banks to fund their activities with more stable sources of funding on an ongoing basis. The Net Stable Funding Ratio (NSFR) has a time horizon of one year and has been developed to provide a sustainable maturity structure of assets and liabilities [BIS (2011); bracket added by the authors]. We will discuss these proposed measures in some detail below. Of course, to the extent that high quality assets qualifying for liquidity include sovereigns [BIS (2010)], recent events regarding European sovereign debt raise concerns about the level risk that is associated even with this type of assets. Time will only tell what the outcome will be.
Industry practice and the liquidity crisis

What was the actual practice of the financial sector, broadly defined, in managing liquidity risk and how was that practice responsible, if at all, for the financial crisis of 2007-2008? This question is of importance because the entire financial sector is not subject to Basel regulations (i.e., insurance companies are exempt) and yet the interaction between non-banks and banks is of critical importance for both bank liquidity and systemic risk considerations. No example better serves to illustrate this point than the collapse of AIG and its central role in securitization and credit default swaps and its interconnections to the rest of financial sector. At a formal level, one might distinguish between liquidity under an asset approach in which the firm maintains liquid instruments on its balance sheet that can
Banks would enter a parallel run period during which banks own methods and the advanced IRB methods were utilized. Once accepted by the regulators, the banks would exit the parallel run and become fully Basel II compliant.
121
would have to post billions of dollars in collateral as the guarantor to the obligations of AIGFP, thus making another downgrade more likely and exacerbating the situation [Haldane (2009); Whitehead (2010)]. Industry-wide, one of the most common sources of firm-specific liquidity risk and liquidity stress across all sectors was downgrades or other negative news, leading to a loss of market confidence in a firm [BIS (2006)]. In addition to the centralized approach, liquidity stress testing was also conducted at the subsidiary level, or by geographical regions, when it was conducted at the group level. This was especially widespread in the insurance sector, but was also seen among some banks. This is especially relevant when the nature of stress scenarios differ between jurisdictions. Firms that did not conduct stress testing at the subsidiary level essentially assumed that all liquidity risks reside in the main corporate unit; these risks were thus analyzed by a group-level test. Subsidiary-level stress testing results were also reviewed at the group level, regardless of centralization, but in this case, subsidiaries had greater autonomy to devise scenarios that were specific to them. We have seen how granting such autonomy towards risk centers may have been at the root of some of the most risk-prone behavior in the insurance industry as was the case with AIG. In the final sections of the paper, we will also examine why the regulatory enforcement of the practice of stress testing failed to equip the banking sector with sufficient liquidity to provide some cushion for the impact of the financial crisis. But prior to that, we will present what is the key section of this survey, namely, the underlying mechanism of the vicious cycle that produced the crisis.
1/
Adjusted closing prices

90 80 70 60 50 40 30 20 10 0
7 7 7 7 7 8 8 8 8 07 8 00 00 00 00 00 00 00 00 00 20 00 /2 /2 /2 /2 /2 /2 /2 /2 /2 7/ /2 17 17 17 17 17 17 17 17 17 /1 17 /1 7/ 20 08
SPDR Dow Jones REIT ETF Financial Select Sector SPDR Fund Industrial Select Sector SPDR Fund Technology Select Sector SPDR Fund Consumer Discretionary Select Sector SPDR Fund
3/
5/
7/
9/
1/
3/
5/
7/
11
9/
Source: finance.yahoo.com
Figure 3 Index prices in four major U.S. sectors
Brunnermeier (2009) extends the above analysis by identifying four distinct economic mechanisms that played a role in the liquidity and credit crunch. First, the effects of bad loan write-downs on borrowers balance sheets caused two liquidity spirals. As asset prices dropped, financial institutions had (i) less capital and (ii) less access to borrowing, because of tightened lending standards. The two spirals forced financial institutions to shed assets and reduce leverage, leading to fire sales, lower prices, thus even tighter funding, amplifying and generalizing the crisis beyond the mortgage market. Second, concern about future access to capital markets led banks to refrain from lending further. Third, runs on financial institutions (Bear Stearns, Lehman Brothers, and others), following the mortgage crisis, eroded bank capital further. Finally, the mortgage crisis was amplified and became systemic through network effects, which can arise when financial institutions are lenders and borrowers at the same time. Because each party has to hold additional funds out of concern about counterparties credit, liquidity gridlock can result.3 Garcia (2009) provides an overview of the liquidity spiral hypothesis introduced by Brunnermeier and Pederson (2009). This takes several stages: i) a decline in asset value erodes investor net worth more than his/her gross worth and borrowing limit. ii) This results in investor forced-selling to reduce overall position and maintain leverage ratio. iii) The sales depress the price further, inducing more selling and so on. iv) A margin spiral reinforces the loss spiral. As margins rise, the investor has to sell even more to reduce his leverage ratio. Margins spike in times of large price drops, leading to a general tightening of lending.
Liquidity risk and systemic propagation: a vicious cycle

Several key research papers in economics have pointed to the possibility of a vicious cycle between different forms of liquidity on one land, and the rise of systemic risk, on the other. A path breaking paper in this area is by Brunnermeier and Pedersen (2009). By linking trading liquidity with funding liquidity this paper shows their mutually reinforcing possibilities and consequently the resulting downward vicious spirals. Thus they write: We provide a model that links an assets market liquidity (i.e., the ease with which it is traded) and traders funding liquidity (i.e., the ease with which they can obtain funding). Traders provide market liquidity, and their ability to do so depends on their availability of funding. Conversely, traders funding, i.e., their capital and margin requirements, depends on the assets market liquidity. We show that, under certain conditions, margins are destabilizing and market liquidity and funding liquidity are mutually reinforcing, leading to liquidity spirals. The model explains the empirically documented features that market liquidity (i) can suddenly dry up, (ii) has commonality across securities, (iii) is related to volatility, (iv) is subject to flight to quality, and (v) co-moves with the market. The model provides new testable predictions, including that 122 speculators capital is a driver of market liquidity and risk premiums.
Network credit risk problems can be overcome if a central authority or regulator knows who owes what to whom, Brunnermeier notes. Then the system can stabilize. But the opaque web of obligations in the financial system that is currently characteristic of securitization tends to be destabilizing, leading to heightened liquidity and credit problems.
11

Since in this model the decline is triggered by bad loans and write-downs before spreading to the other sectors, and since it is well known that such loans were concentrated in real estate, it would be of interest to examine the implications of this model by considering the decline in major sectors of the U.S. economy. This is depicted in Figure 3. As can be seen, the real estate downturn preceded that of other sectors. In fact, it was not until August of 2008 that the impact of this downturn began to be felt in the financial sector. Thus, evidence seems to lend support to the common understanding of the leading role of real estate and of the bad loans and write-downs that triggered the downturn. Support for Brunnermeier and Pedersens liquidity spiral mechanism is found, among others, by Adrian and Shin (2010) who document strong procyclical behavior of marked-to-market leverage among financial intermediaries. Viewing balance sheets of financial intermediaries as a measure of aggregate liquidity (by showing that balance sheets effects spread through changes in the repo markets) the authors are able to establish that changes in dealers repos are actually capable of predicting changes in financial market risk as measured by the innovations in the VIX (Chicago Board Options Exchange Volatility Index) [Adrian and Shin (2010)]. Repo markets are also at the heart of an analysis provided by Gorton and Metrick (2011), who identify a run on the sale and repurchase market, as the main cause of the financial crisis 2007-2008. Gorton and Matrick show how concerns about the liquidity of markets for bonds that were used as collateral led to increases in repo haircuts, leading to a further decrease in liquidity. Using the spread between the LIBOR and the OIS (overnight index swap rates) as a proxy for the systemic risk, they demonstrate that securitization banking was a core factor in creating such a systemic event. Longstaff (2010) tests for the existence of financial contagion by using subprime asset-backed collateralized debt obligations (CDOs) (ABX indexes). Longstaff finds that financial contagion effects exist and spread primarily through liquidity channels [Brunnermeier and Pedersen (2009)] as well as risk-premium channels [Acharya and Pedersen (2005)], rather than through information channels [Dornbusch et al. (2000), thus providing evidence of cross-market linkages. Hameed et al. (2010) also attempt to test the theoretical predictions of Brunnermeier and Pedersen (2009). Using portfolio information from investment banks, securities brokers, and dealers listed on NYSE, weekly changes in aggregate repo, and weekly spread in commercial paper (CP) they show that negative market returns decrease stock liquidity, especially during times of tightness in the funding market. They further show that these adverse liquidity effects spill over into other sectors and attribute this to capital constraints in the market making sector. An empirical investigation of contagion effects in hedge funds is provided by Boyson et al. (2010). Basing their analysis of hedge fund contagion on the liquidity spiral model of Brunnermeier and Pedersen (2009), they
find that large adverse shocks to funding and asset liquidity strongly increase the probability of contagion. Using monthly hedge fund index return data they are able to link hedge fund contagion to liquidity shocks. Another study by Fontaine and Garcia (2007) finds that liquidity covaries with changes in aggregate uncertainty, as measured by the volatility implied in S&P500 options, and with changes in monetary stance, as measured by bank reserves and monetary aggregates. Finally, Acharya et al.. (2009) focus on the contribution of banks to systemic risk. Measuring systemic risk of financial institutions as systemic expected shortfall (SES), i.e., its propensity to be undercapitalized when the system as a whole is undercapitalized, they show that leverage (measured as assets to common equity ratio) in the form of short-term repos and short-term asset-back commercial paper agreements play a large role in determining the systemic risks of financial institutions. They show that the SES measure would have been able to predict the distress in market liquidity, the drop in asset prices, and the overall increase in credit risk during the recent financial crisis. In short, there is now a well-established and empirically verified research on the conceptual explanations of the mechanism that produced the financial crisis of 2007-2008. We will investigate whether this rich new literature has begun to influence regulatory policy and if so, how. Before that discussion, however, we will spend the next section on an important component of liquidity risk, namely market liquidity risk and its measurements.
Measuring market liquidity risk

The increasing importance of understanding liquidity risk in corporate banking calls for a better understanding of other forms of liquidity risk, especially market liquidity risk. For example, the BIS writes in the Basel III regulations: Information such as equity prices and credit spreads are readily available. However, the accurate interpretation of such information is important. For instance, the same CDS spread in numerical terms may not necessarily imply the same risk across markets due to marketspecific conditions such as low market liquidity. Also, when considering the liquidity impact of changes in certain data points, the relation of other market participants to such information can be different, as various liquidity providers may emphasize different types of data [BIS (2010)]. Given this background, the need to understand different types of liquidity, especially market liquidity, is of utmost importance. One question is how should one incorporate an actual measure of liquidity risk into a banks quantitative methodology? The basic approach has been to incorporate the cost associated with illiquidity into a market risk-based Value at Risk (VaR) calculation. For example, Bervas (2006) defines an Exogenous Liquidity Cost (ELC) as the worst expected half of the bid-ask spread at a particular confidence level and adds this to the value of VaR. 123
Basically when calculating a market-based VaR, the underlying assumptions are that (a) the positions can be liquidated or hedged within a fixed and fairly short timeframe (one day or ten days), (b) that the liquidation of positions will have no impact on the market, and (c) that the bid-ask spread is not affected by the size of the position. As a result, the price referred to is the mid-price or the last known market price. However, as Bervas (2006) argues, the quoted market price cannot be used as a basis for valuating a portfolio that is being sold on a less than perfectly liquid market: for example, in practice, we must take account of the orderly liquidation value or even its distress liquidation value. The standard VaR model is not a reliable guide because it neglects the risk to which a portfolio is exposed during its liquidation. It is nonetheless possible to adjust VaR measures. This is done by essentially recalculating the distribution of asset returns using not the market value but the liquidation value in normal times or in times of stress. In general, the value that is realized from reselling assets is not equivalent to their theoretical market price because a liquidity cost, represented by the half-spread, puts a strain on the sale price. The parametric method for calculating VaR, applied to an asset, is based on the assumption of the normal distribution of returns of this asset. Let be the expected value of the distribution and its standard deviation. The lowest return expected at date t, at a confidence threshold of 99%, is: Rt* = ln(Pt*/Pt) = 2.33 (1)
available, (b) it assumes spreads and prices are normally distributed, leading to an underestimation of the risk as both tend to have long tails [Mohtadi and Ruediger (2011)], and (c) that summing up the price risk (standard VaR model) and the exogenous liquidity cost amounts to assuming that these two components are perfectly correlated (i.e., that high variability of the mid-price is associated with the high variability of the bid-ask spread itself), leading to a possible overestimation of the risk. Thus, effects (b) and (c) may counter one another. The data limitation issue may be addressed by a simple alternative method known as Roll [Roll (1984)], which estimates the implied spread using only observed market price series. However, the assumptions used to arrive at this estimate are highly limiting. The normality assumption associated with liquidity risk under the treatment by Bervas (2006) is perhaps the most limiting of all of its assumptions, since in general liquidity events are highly unusual tail events. Bangia et al. (1999), who also advocate adjusting VaR with a liquidity risk measure, do not rely on this normality assumption, even though they preceded Bervas. Bangia et al. state that: We are interested not in average circumstances but in unusual, tail-event circumstances, albeit due to overall market conditions (exogenous liquidity risk) rather than as a result of an individual trader attempting to dispose of an unusually large position (endogenous liquidity risk). We define exogenous liquidity risk measurement in terms of a confidence interval or a tail probability. While we have begun to move away from the assumption of normality in finance, as for example discussed by Mohtadi and Ruedigers (2011) survey of the literature on the rise of heavy tailed distributions in finance, nonetheless the normality assumption unfortunately still dominates much of the practice. The basic generalization offered by Bangia et al. is that the normal 99% quintile of 2.33 for liquidity risk in Bervas (2006) is replaced by a scaling factor a, ELC = Pt( + a) (5)
where Pt is the asset price (i.e., midpoint) at time t and Pt* is the worst price expected at a confidence threshold of 99% (hence coefficient 2.33 under the normal distribution). The value of VaR at date t, measuring the highest potential loss at a confidence threshold of 99% is by definition, VaR = Pt - Pt* = Pt(1 - eR*t ) = Pt(1 -e 2.33) (2)
If we assume that the relative bid-ask spread is also taken from a normal distribution (see below for relaxing this assumption), the worst relative spread at the threshold of 99% is + 2.33 , where is the expected relative spread and is its standard deviation. In that case, the exogenous liquidity cost (ELC) at a one-day horizon is the worst half-spread at the threshold of 99%: ELC = Pt( + 2.33 ) (3)
such that a 99% VaR would result. Their estimate of a is in the range of 2 to 4.5, depending on the product and the market. Thus, while some market and products may exhibit near normal liquidity behavior (a near 2.3), others exhibit a distribution that has a much longer tail than normal (a near 4). Compare, for example, the daily spreads in Japanese yen versus the Indian rupee (figures below). It is clear from this comparison that the liquidity risk implied by the bid-ask spread is widely different for the two currencies, with the latter showing a dramatic departure from Gaussian. It follows that a measure of liquidity risk based on Bangia et al. (1999) is superior especially in emerging markets and other less liquid environments.
So that the liquidity adjusted VaR, which we will call VaRL, is given by: VaRL = VaR + ELC = Pt(1 -e 2.33) + ( + 2.33 ) (4)
Several drawbacks of this methodology are that (a) it requires large 124 samples of daily or even intra-day trading data, which are not always

S(t,x) in equation (6). To do this the authors first assume the usual Geometric Brownian Motion for the competitive flat-supply S(t, 0) process: dSi(t,0) = i(t)Si(t,0)dt + i(t)Si(t,0)dWi(t) (7)
Let us suppose that percent of the equity is to be liquidated at time T when a crisis state occurs. The liquidity cost is then given by, LT = -XT[S(T,-XT) S(T,0)]
Figure 4 Daily spread distribution for Japanese Yen (5/95 5/97)
(8)
where the liquidation price is less than normal price S(T,-XT) < S(T,0) if shares are sold (XT > 0) so that LT > 0. This equation then allows for the ultimate estimation of the value of S in crisis times from which various measures of risk including VaR can be derived under the liquidation state. Following a number of algebraic steps one obtains the following equation, VL = XTS(T,0) LT = VT[1 - C2XT] < VT T (9)
Source: Bangia et al. (1999)
As can be seen from this equation, by simply replacing the securitys price with a liquidity-adjusted price, this paper goes deeper than others and rederives the securitys liquidity adjusted price as the basis for VaR calculation or any other risk measure in the first place. By modeling this
Figure 5 Daily spread distribution for Indian Rupee (5/95 5/97)
Perhaps, however, the most rigorous and fundamental treatment of how to measure liquidity risk comes from Jarion and Potter (2005), who hypothesize the existence of a stochastic supply curve for a securitys price as a function of transaction size. Consider a security supply curve in which the price S at time t depends on the aggregate holding of a stock x and is given by, S(t, x) = S(t,0)[1 + c 1cx + n(1 - 1c)x] (6)
aspect, this paper is able to provide an adjustment that is based on a Monte Carlo engine and does not rely on historical prices. Finally, an interesting measure that contains the promise of linking the VaR-based liquidity measures discussed above to systemic risk is provided by Adrian and Brunnermeier (2009) who propose to use a CoVaR analysis. This is supposed to capture the VaR of financial institutions conditional on other institutions being in distress.
The first term on the right in this supply curve is the classical asset values process S(t, 0), as might be available from an option pricing approach such as the Black-Scholes model and is characterized by the usual geometric Brownian motion. The second and third term capture the quantity impact x on the assets price which is assumed to be linear, with a different slope coefficient in normal times n versus a crisis scenario c where 1c is an indicator variable taking on values of 0 and 1 only. With the additional assumption that c > n the quantity impact on equity prices is assumed to be larger in crisis times than in normal times.
Regulatory response
How has regulatory policy changed in response to the liquidity crisis of 2008? In this section we try to address this question in some detail. We divide the discussion into two segments. The first question is, how should regulators respond to a potential liquidity risk? This question naturally surveys the normative literature in this area. The second part of the discussion concerns the actual regulatory policy on liquidity risk management and its evaluation. The literature on optimum liquidity risk policy is relatively recent and
What makes this approach fundamental and basic is that while other papers adjust the VaR by introducing a liquidity risk additively, this paper goes further and rederives the securitys liquidity adjusted price, in the first place, as the basis for VaR calculation or any other risk measure. The key to the model by Jarion and Potter is how to estimate the slope of
therefore somewhat sparse. But the few papers that are available offer important insights and general risk management tools. One interesting paper in this area is by Jan Willem End (2008) of the Central Bank of the Netherlands (DNB). This paper applies data from all of the major banks in the Netherlands in a Monte Carol simulation model to both develop 125
a liquidity risk management tool and to simulate the bank response to a liquidity crisis, including the modeling of the funding of liquidity risk. The paper takes into account the key drivers of liquidity stress, such as on- and off-balance sheet contingencies, feedback effects induced by collective reactions of heterogeneous banks, and idiosyncratic reputation effects. The model (and the tool) also contributes to the understanding of contagion and aggregate effects (externalities) across banks in response to liquidity crises. The outcome supports enhancing liquidity buffers by the regulators as well as liquidity risk management. As stated, the paper provides a useful tool for central banks to evaluate the importance of various factors to banks liquidity positions. Somewhat contrastingly, however, a paper by Cao and Illing (2010), which provides a baseline game-theoretic model for regulatory analysis of systemic liquidity shocks, argues that narrow banking and imposing equity requirements as a buffer are inferior mechanisms for coping with systemic liquidity risk. The paper shows that banks may have an incentive to invest excessively in illiquid long-term projects. In the prevailing mixedstrategy equilibrium, the allocation is inferior from the investors point of view since some banks free ride on the liquidity provision due to their limited liability. The paper compares different regulatory mechanisms to cope with the externalities and shows that a combination of liquidity regulation ex-ante and lender of last resort policy ex-post can maximize investor payoff. An interesting and factual paper by Goldsmith-Pinkham and Yorulmazer (2010) studies the 2007 bank-run on Northern Rock, the fifth largest mortgage lender in the U.K., by analyzing both the investor response and the government guarantee response. Of relevance to our survey is the observed effect of rational investor response to market news regarding the liability side of banks balance sheets: banks that relied on funding from wholesale markets were significantly affected, a result consistent with the drying up of liquidity in wholesale markets and the record-high levels of the London Interbank Offered Rate (LIBOR) during the crisis. The second part of our discussion focuses on the actual record of regulation in this area. The BIS seems to have understood the potential link between liquidity risk and systemic risk that was discussed earlier. Thus, in its publication [BIS (2009a)] it stresses the point that as risk increases markets become less liquid, in turn affecting the risk profile of the banks. It shows how market liquidity has a strong influence on banks exposure to market risk and credit risk, which in turn affect market liquidity, thus completing a vicious cycle. Yet, the regulatory practice of stress testing the banks balance sheets failed to equip the banking sector and the financial sector in general with sufficient liquidity to provide some cushion for the impact of financial crisis. Why was that so? Haldane (2009) identifies the major weaknesses of stress-testing based on Basel II as disaster myopia, the notion that the subjective probability attached to events decreases over time. Moreover, the stress-tests that were performed dur126 ing the last few years by banks only relied on data from good economic
times, thus making the time-periods used for stress-tests inadequate. A more general criticism by Brunnermeier et al. (2009), Kretzschmar et al. (2009), Haldane (2009), and Moosa (2010) has concluded that stresstesting guidelines allowed for too much autonomy for banks own internal approaches to stress-testing that in most cases relied on too short a time frames, instead of using a stress-testing scenario provided through the regulators. Additional criticisms of the regulatory policy can be classified into three effects: network effects, procyclicality of capital, and leverage. Atik (2010), Brunnermeier et al. (2009), Haldane (2009), and Kretzschmar et al. (2009) emphasize that Basel II failed to account for network effects, the correlations of the balance sheet of one institution with the balance sheets of other institutions; Brunnermeier (2009), Kretzschmar et al. (2009) and Moosa (2011) point out that Basel II regulations generally failed to address issues of leverage or generally the extent to which Basel II regulations ignored or even encouraged off-balance sheet activities; and finally, Andersen (2011), Ayuso et al. (2004), Brunnermeier et al. (2009), Brunnermeier and Pedersen (2009), Jokipii and Milne (2008), and Moosa (2011) stress the procyclicality of Basel II regulations. Responding to the criticism in the literature and its own analysis of the failure of its stress-testing procedures, BIS has identified the major weaknesses of the stress-testing guidelines of the Basel II rules (BIS, 2009b). This self-criticism has revolved around four factors: the short time horizon of crisis scenarios, the underestimation of the network effects, the system-wide interactions and the feedback effects, and finally the lack of applicability of historical default rates. Thus, in 2009 the BIS released new guidelines for sound stress-testing (BIS 2009b). The BIS has formalized, crystallized and strengthened these earlier publications in its Basel III document, which is essentially a morphing of these earlier publications into Basel III. The grand stated goal of Basel III is reducing the risk of spillover from the financial sector to the real economy [BIS (2010)]. Basel III also develops two specific measures of liquidity that are built on these conceptual underpinnings; a short-run (30 days), the Liquidity Coverage Ratio (LCR), and a longer-run (one year) measure of liquidity, the Net Stable Funding Ratio (NSFR), which addresses the structure of assets and liabilities of banks. Under LCR banks are to hold sufficient amount of high-quality assets that can be converted into cash even under severe liquidity stress. These high-quality assets should still be able to provide liquidity, under stress conditions, without little or no loss of value, and ideally, be central bank eligible [BIS (2010)]. The measure is defined relative to the cash flow (as opposed to balance sheet) as, Stock of high quality assets net cash outflows over next 30 days 100%

BIS stresses that high-quality liquid assets should retain their value even in fire-sale situations and should not be used in any form for collateral, securitization, hedging, or credit-enhancements. The LCR is an extension of coverage ratios commonly used by banks for internal risk analyses. The BIS envisions that the new LCR will help banks get through or lead to an orderly resolve of the bank in a combined idiosyncratic and market-wide stress scenario in which the bank faces cash outflows. The NSFR is designed to complement the LCR by creating a longer-term planning horizon for banks. The goal is to limit the need of banks to rely on short-term wholesale funding during times of buoyant market liquidity [BIS (2010)]. This is an acknowledgement of the liquidity spirals described by Brunnermeier and Pedersen (2009). NSFR standard is defined as, available stable funding required stable funding 100%
Conclusion
We have studied liquidity risk in several ways in this survey. Following an overview of the financial crisis of 2007-2008, we inquired into the causes of the crisis and presented the state of knowledge and research on the vicious cycle of illiquidity-fire-sale-balance-sheet-effects-illiquidity. This approach established an explanation of systemic risk and the spiral interaction between various forces that underline the emergence of a crisis. Next, we covered the industry practice of liquidity risk management from an organizational point of view and asked whether this factor was also at work as one source of the liquidity crisis. We then covered the incorporation of liquidity risk into the traditional measure of market risk. This issue has direct implications for banks quantitative and advanced IRB approach. In the final section, we studied the state of regulatory policy and provided an evaluation of its strength and weaknesses. It is hoped that in presenting this critical survey, driven by a focus on what went wrong, we have provided a much needed and unique perspective that has been lacking in the literature. We have learned, not only that firms, markets, and regulators were all to blame, but how liquidity risk actually travelled, how it is measured, and how regulators have responded. Time will only tell whether we have learned the right lessons from the disaster that we have just experienced.
The NSFR builds on traditional net liquid asset and cash capital methodologies used widely by internationally active banking organizations, bank analysts and rating agencies (BIS, December 2010).4 The goal of this is to provide stability to a bank in the case of a firm-specific event such as a significant decline in profitability or solvency arising from heightened credit risk, market risk or operational risk and/or other risk exposures; a potential downgrade in a debt, counterparty credit or deposit rating by any nationally recognized credit rating organization; and/ or a material event that calls into question the reputation or credit quality of the institution [BIS (2010)]. To summarize, Basel III introduces new capital buffers to address the procyclicality of Basel II regulations; introduces the requirement to account for counterparty credit risk, thereby again responding to impacts of systemic events; and stresses the need to hold more high quality liquid assets by introducing new liquidity measurements into its regulations, thus responding to the need to reduce liquidity risk. It also aims to limit leverage, as well as counterparty risk. In short, Basel III considers a greater role for macroprudential regulation in addition to its already existing microprudential regulation, thus responding to the need to better account for systemic risk caused by globally interconnected financial institutions. Despite these many regulatory improvements, Basel III has not remained immune to criticism. For example, it has been argued that Basel III rules may be insufficient to ensure against extreme risk (such as what characterized the Great Depression) by focusing mainly on higher quality assets and on stress periods less than one year [Varotto (2011)]. The inadequacy of covering the risk of tail risk is also shared by Gordon et al. (2011).5 Time will tell if the measures introduced in Basel III will be able to prevent another crisis or not.
References

Acharya, V., L. Pedersen, T. Philippon, and M. Richardson, 2010: Measuring systemic risk, NYU Working Paper Adrian, T. and H. S. Shin, 2010, Liquidity and leverage, Journal of Financial Intermediation, 19:3, 418437 Andersen, H., 2011. Procyclical implications of Basel II: can the cyclicality of capital requirements be contained? Journal of Financial Stability, 7:3, 138-154 Atik, J., 2010, Basel II and extreme risk analysis, Loyola-LA Legal Studies Paper No. 2010-40 Ayuso, J., D. Prez, and J. Saurina, 2004, Are capital buffers pro-cyclical? Evidence from Spanish panel data, Journal of Financial Intermediation 13, 249264 Bangia, A., F. X. Diebold, T. Schuermann, and J. D. Stroughair, 1999, Modeling liquidity risk, with implications for traditional market risk measurement and management, Wharton School Working Paper 99-06 Bank of International Settlements (BIS), 2006, The Joint Forum: the management of liquidity risk in financial groups. May Bank for International Settlements (BIS), 2009a, Findings on the interaction of market and credit risk, Basel Committee on Banking Supervision Working Paper No 16 Bank for International Settlements (BIS), 2009b, Principles for sound stress testing practices and supervision, Bank for International Settlements Consultative Document, January Bank of International Settlements (BIS), 2010, Basel III: international framework for liquidity risk measurement, standards and monitoring, Basel Committee on Banking Supervision, December Bervas, A., 2006, Market liquidity and its incorporation into risk management, Financial Stability Review No. 8 May 2006 (Financial Stability Co-ordination Directorate Financial Stability and Market Research Division Banque de France)
Stable funding is defined as capital; preferred stock with maturity of equal to or greater than one year; liabilities with effective maturities of one year or greater; that portion of nonmaturity deposits and/or term deposits with maturities of less than one year that would be expected to stay with the institution for an extended period in an idiosyncratic stress event; and the portion of wholesale funding with maturities of less than a year that is expected to stay with the institution for an extended period in an idiosyncratic stress event [BIS (2010)]. See Mohtadi and Ruediger (2011) for examination of the state of knowledge on heavy tail statistics in finance.
127
Billio, M., M. Getmansky, A. W. Lo, and L. Pelizzon, 2010, Econometric measures of systemic risk in the finance and insurance sectors, NBER Working Paper No. 16223 Boyson, N. M., C. W. Stahel, and R. M. Stulz, 2010, Hedge fund contagion and liquidity shocks, Journal of Finance, 65, 17891816 Brunnermeier, M., 2009, Deciphering the liquidity and credit crunch 2007-08, Journal of Economic Perspectives, 23:1, 77100 (also NBER Working Paper No. 14612, December 2008) Brunnermeier, M. and L. H. Pedersen, 2009, Market liquidity and funding liquidity, Review of Financial Studies, 22:6, 2201-2238 Brunnermeier, M., A. Crockett, C. Goodhart, A. Persaud, and H. Shin., 2009, The fundamental principles of financial regulation, Centre for Economic Policy Research. Available at http:// www.cepr.org/pubs/books/P197.asp. Cunningham, L. A. and D. Zaring, 2009, The three or four approaches to financial regulation: a cautionary analysis against exuberance in crisis response, The George Washington Law Review, 78:1, 39-113, 55 n.60 Dornbusch, R., Y. Park, and S. Claessens, 2000, Contagion: understanding how it spreads, The World Bank Research Observer15, 177197. Federal Register, 2008, Rules and Regulations, Vol. 73, No. 148 Fontaine, J-S. and R. Garcia, 2007, Bond liquidity premia, Working Paper 2007, Universite de Montreal, CIRANO and CIREQ Garcia, R., 2009, Liquidity risk: what is it? How to measure it? EDHEC Business School, CIRANO, Cirano, Montreal Goldsmith-Pinkham, P. and T. Yorulmazer, 2010, Liquidity, bank runs, and bailouts: spillover effects during the Northern Rock episode, Journal of Financial Services Research, 37:2-3, 83-98 Gordon, J. A., A. M. Baptista, and S. Yan, 2011, A comparison of the original and revised Basel market risk frameworks for regulating bank capital, University of Minnesota Working Paper Gorton, G. and A. Matrick, 2011, Securitized banking and the run on repo, Journal of Financial Economics, Available online 25 March 2011 Haldane, A. G., 2009, Why banks failed the stress test, speech given at the 2009 MarcusEvans Conference on Stress-Testing. Bank of England Publications Hameed, A., W. Kang, and S. Viswanathan, 2010, Stock market declines and liquidity, Journal of Finance, 65, 257293 Hazen, T. L., 2009, Filling a regulatory gap: it is time to regulate over-the-counter derivatives, North Carolina Banking Institute, 123:13, 123135 Jarrow, R. and P. Protter, 2005, Liquidity risk and risk measure computation, Cornell University Working paper Jin, C. and I. Gerhard, 2010, Regulation of systemic liquidity risk, Financial Mark Portfolio Management, 24, 3148 Jokipii, T. and A. Milne, 2008, The cyclical behaviour of European bank capital buffers, Journal of Banking and Finance, 32, 14401451 Kretzschmar, G. L., A. J. McNeil, and A. Kirchner, 2009, Integrated capital adequacy principles for institutional, asset and economic risk factor stress testing, (March 13, 2009). Available at SSRN: http://ssrn.com/abstract=1318264 or doi:10.2139/ssrn.1318264 Longstaff, F. A., 2010, The subprime credit crisis and contagion in financial markets, Journal of Financial Economics, 97:3, 436-450 Mohtadi, H. and S. Ruediger, 2011, The heavy tail in finance: a survey, forthcoming in Econometrics: New Developments (Nova Science Publications) Moosa, I. A., 2010, Basel II as a casualty of the global financial crisis, Journal of Banking Regulation, 11:2, 95-114 Roll, R., 1984, A simple implicit measure of the effective bid-ask spread in an efficient market, Journal of Finance, 39:4, 11271139 Varotto, S., 2011, Stress testing credit risk: The Great Depression scenario, Journal of Banking and Finance, forthcoming Werrett, J., 2009, Achieving meaningful mortgage reform, Connecticut Law Review, 42:1, 319-364 Whitehead, C. K, 2010, Reforming financial regulation, Boston University Law Review, 90:1, 16-36
128
PART 2
Making Sense of Asset Prices: A Guide to Required Yield Theory, Part 1 Valuing the Stock Market
Christophe Faugre Professor of Finance, Bordeaux School of Management (BEM) and Center for Institutional Investment Management, State University of New York Albany School of Business1
Abstract
The bad news is that there is widespread confusion about how asset prices are determined. The good news is that this series of short essays about Required Yield Theory aims at establishing a clear understanding of the underlying mechanisms behind asset prices. Installment # 1 covers the stock market. I survey the current state of knowledge regarding stock valuation and showcase the logical and empirical effectiveness of Required Yield Theory as it explains how the S&P 500 is valued. Based on the tenets and results of this new theory, I derive a list of key insights for investing in the S&P 500.
1 I would like to thank David Lebris and the editor for useful comments.
129
The current state of confusion regarding stock (and asset) valuation

The challenge, of course, is the calculation of intrinsic value. Present that task to Charlie and me separately, and you will get two different answers. Precision just isnt possible. To eliminate subjectivity, we therefore use an understated proxy for intrinsic-value book value when measuring our performance. Warren Buffett. Letter to the Shareholders, February 26, 2011.
other traders irrationality is likely to fall prey to psychological biases as well, even though these might be of a different kind. Financial markets are swayed by a mixture of these patterns, and we just do not know which bias is likely to prevail at any particular moment.2 Even Warren Buffetts contrarian dictum: be greedy when others are fearful and fearful when others are greedy does not necessarily have traction in the world of behavioral finance. I view the empirical success of RYT as a new and strong piece of evi-
So, where do I begin? There is so much confusion out there amongst academics, practitioners, and the investing community regarding how to value stocks and other asset classes such as treasuries, gold and real estate that seems quite tragic, bordering on comedic. In this series of essays, I am showcasing Required Yield Theory; a new asset pricing theory developed jointly with J. Van Erlach in the mid-2000s. What sets Required Yield Theory (hereafter RYT) apart from the existing body of knowledge is that it provides a clear and intuitive mechanism by which assets are priced, and it works! This is a bold claim to make, but it is supported by the fact that RYT has performed better in terms of tracking error than any other approach (known to us) when applied to valuing the S&P 500, Treasuries and gold [Faugre and Van Erlach (2005, 2009)]. The foundation for RYT rests on well known principles of rational economic behavior. One such principle is the so-called Fisher (1896) effect. Three decades ago, most economists would have agreed that this and other cornerstone principles should describe the way actual financial markets work. However, to the dismay of those engaged in this pursuit, the innumerable tests conducted in the last thirty years have produced little to no empirical support for the Fisher hypothesis. What sets RYT apart and why is it able to bridge the gap between financial theory and reality? The crude answer is that we were just lucky and stumbled on a combination of these standard principles plus new ones, which worked. For instance, the basis for our valuation of the S&P 500 is rational economic behavior. But we add another layer to the model related to basic investor psychology by quantifying the impact of fear on the market. Even though many academic bastions are still defending this view, the rational approach to asset pricing has fallen into disfavor mostly due to the lack of empirical success mentioned above. From being an underdog in academic finance two decades ago, the counterrevolution known as behavioral finance has taken a dominant position in the field. According to that school of thought, stock prices do not behave randomly as the efficient market hypothesis presumes. Non-random patterns in stock returns can emerge and be persistent but not necessarily predictable or exploitable. The standard explanation for many of these market or behavioral anomalies is that investors make cognitive mistakes. Even if it is pointed out to them what these mistakes are, it is of little use. They cannot break out of these behavioral patterns because that is just how the brain 130 is designed to apprehend reality. A smart trader trying to capitalize on
dence in support of the hypothesis that investors behavior is rooted in rational economic calculus. Nevertheless, it is also true that psychological biases do indeed throw investors off the rational path occasionally and may lead to significant discounts or premia found in asset prices. RYT is able to separate out and quantify some of these basic psychological biases.3 I invite the readers to judge whether or not you find the logic and evidence presented here compelling. We certainly do. In this first essay, I survey the state of current knowledge regarding how the stock market is priced. My focus is on explaining the behavior of broad equity indexes, in particular the S&P 500. Of course, it is impossible to explain how to value a stock index without covering the valuation of individual stocks. I do a cursory examination of several techniques favored by practitioners. But my true goal is to cover the important milestones in academic research, as well as some original approaches you may or may not have heard of. I am well aware that this could turn into a tedious exposition. Consequently, I try to avoid arcane discussions and minimize mathematical exposition. I discuss when financial models are based on assumptions far removed from reality or when internal logic is lacking.4 Then, I turn to RYT, and show how it helps making sense of how the price of a stock index such as the S&P 500 is determined.
The elephant in the room: managing assets without understanding how asset prices are determined
A particular puzzle has been haunting me for a while. How can equity investment managers and so-called gurus make it in that business, and sometimes make it big, when they do not know how to value stocks? I do not mean to be facetious or irreverent towards the profession. But, the pervasive lack of clear understanding and use of unscientific techniques is a serious issue, albeit not necessarily one that the industry
Even though there are examples of sustained anomalies, such as the value premium puzzle (value stocks historically getting higher returns than growth stocks), which are still not wellunderstood. Much work remains to be done to integrate key psychological patterns in the current version of the theory. Without oversimplifying human psychology, another often decried characteristic, greed, is currently in the process of being worked into the theory. This aspect may have important implications for understanding asset bubbles. I am not alone in critically surveying academic asset pricing theories. Notable is Shojai and Feiger (2009)s survey, which tackles mainly static modern portfolio theory, in particular CAPM, and option pricing theory.

wants to dwell upon. For example, it is well understood that most equity analysts do attempt to value stocks. This is just part of their job. Generally these attempts do not produce exact or even reliable results, which I will discuss later. At any rate, the above inquiry led me to gather a series of observations about the industry: 1. The various themes popularized today in the financial media and books are that stock prices are determined by supply and demand; that they follow random walks; or alternately that markets are driven by fickle human psychology, by momentum, or even by societal (Elliott) wave patterns. Somewhat surprisingly, many ex-physicists turned finance gurus turn their back on what they were trained to do: that is, look for repeatable patterns and general laws. Instead, they embrace the notion that there are no unifying set of scientific principles for accurately determining stock prices. This is because the research community has resigned itself to a narrow field of possibilities regarding what we can understand in finance. Generations of economists including myself, have heard the criticism that economics and finance cannot be like Physics, that human behavior cannot be reduced to a set of equations, etc. Of course, these arguments will have saliency as long as the quest for finding these principles is unsuccessful. 2. The fact that investment professionals have a limited understanding of any principles that govern stock prices does not constitute a problem (for them) as long as everyone else shares the same limited knowledge. We are all chickens running around with our heads cut off. 3. Trading is a skill in its own right. Traders are action-takers. They are like infantry captains making decisions under fire, quickly adjusting their tactics in response to the enemys actions, often as a result of gut intuition. By contrast, equity strategists are generals with a synthetic view of the battlefield. They are supposed to have deep understanding of the theories of warfare, but could not necessarily be as effective as their captains on the field.5 The key lesson is that being a successful trader does not require full understanding of the underlying principles of asset valuation. This is rather obvious, given that there are successful traders out there, and they, as much as we, share the same quandary about how stock prices are determined. Rather, traders anticipate price movements to a great extent based on their intuitive grasp of market psychology. Of course, algorithmic trading may change all that. But that discussion is for another time and place. 4. Let us clarify a point about market psychology. If there are laws governing asset prices, they obviously originate consciously or unconsciously from the human mind and are manifested in trading behaviors. I differentiate between the thought processes associated with what I call rational cost-benefit-risk analysis, which is what traditional economics focuses on, versus the thought patterns that overlap and sometimes interfere with that type of analysis, which is the subject matter of behavioral finance. Rational cost-benefit-risk analysis works simply by comparing the expected material side
7 5 6
of gains and costs of undertaking an action. Once all the objective costs, benefits, and risks are accounted for and clear to everyone, which can get complicated, any person can logically get and agree with the conclusions of such analysis. On the other hand, when overtaken by various emotional states such as overconfidence or fear for example, individuals get tunnel vision and often deviate from the recommendations of this type of cost-benefit logic.6 5. Because financial professionals have a limited understanding of how stock prices, and by extension, how returns are determined, they shift their attention to capturing still important but peripheral gains for their clients. The tendency in the asset management industry is towards market segmentation and product diversification. The industry promotes choices between cap and style categories, an array of sector and asset allocations, and tax-sheltered investments. This happens for two reasons: i) strategies that produce absolute abnormal profits (net of transaction costs) cannot be sustained because they are either duplicated, countered, or eventually reach the point of diminishing returns to AUM size; and ii) as managers like to state in their fund prospectuses past performance is not a guarantor of future performance. In other words, abnormally good performance does not persist, which leads investors to select products that first and foremost match their risk/gambling versus return profiles. 6. The relative performance of active versus passive management is a perfect example of market segmentation. It is interesting that active management has clearly come out as the loser in terms of average documented performance, and yet, mysteriously, it endures. Either active managements performance has not been truly appreciated, maybe because the best performance is hidden from view, or as I believe is more likely, active management taps into a segment of the investing clientele with a higher willingness to take on risk. When selecting an actively managed fund, investors are consciously choosing to partake in a lottery.7 7. Salesmanship and marketing are critical skills as well. For example, the ability to inspire trust even in the face of less than stellar performance helps managers maximize money inflows and limit outflows, which ensures a steady stream of revenues from management fees. In fact, these managers can certainly make the same case to their cli-
Another and possibly better analogy is that of athletes and their coaches. Of course, there are athletes who are also their own coaches, i.e., tennis superstar Roger Federer. Of course, economists model the satisfaction of gains and losses in what is called a utility function, which should in principle be able to reflect what we know about typical psychological and emotional states. But, it is often assumed that utility functions satisfy some rationality postulates. On the other hand, I do not mean to say that trade-off and risk analyses conducted by economists are the gold standard. In particular, psychological costs do matter and should be incorporated in cost-benefit analyses whenever relevant. An interesting quick study would be to examine how skewed the distribution of returns is for actively managed funds. I had an eye-opening conversation recently with an asset manager for high net worth individuals. Most of his clients truly expect high returns with zero downside risk. He was adamant that this wishful thinking trait was embedded in investors psychology.
131
ents as done in points 2), 3), and 5) above: Alright, I did not perform well this past year, but understand that no manager can have long series of winning streaks. I believe our strategy is sound, sometimes you just get unlucky. Hence, clients are driven to invest in funds that match their worldview and incomplete understanding of financial markets, which could also be influenced by salesmanship, and do not invest in products which may objectively be better for them, given their risk-aversion profile. 8. While investment skill is certainly important in theory, success in the investment industry can in many instances be attributed to a combination of luck and skill, where luck seems to play the larger role. Some managers are able to outperform the rest because their specific mental skills and strategies happen to do well in a particular market environment. But when the environment changes, another set of managers picks up the torch. In theory, and going a bit against my point #5 above, one cannot rule out that some managers use complex algorithms that enable them to reliably pick out market inefficiencies, and those algorithms are not easily copied or understood. This, for example, is the approach reportedly followed by James Harris Simons Renaissance Technologies fund, which has sustained better than average performance since the mid-1980s. However, as proven by Long-Term Capital Managements debacle in 1998, such algorithms are not always bullet proof. 9. The Warren Buffett way: i.e., value investing. Buying a distressed asset at a discount that is deep enough implies that you do not have to worry about fully understanding and accurately measuring that assets true worth. The key is to ensure that the discount you get is greater than the risk you bear. In other words, limit the downside and maximize the upside. The art of picking truly undervalued stocks has to do with exploiting circumstantial barriers to market efficiency, such as illiquid or regulated markets, which force sellers to severely underprice their asset. 10. Academia and high profile professionals have been helping to perpetuate these notions. I do not think there is anything wrong with the financial industry trying to make money in any way they can (as long as they do so ethically). And it is certainly possible that the practices listed above are just another constant of life, i.e., a manifestation of the cross-section of human capital skills. But, like many others, I have signed up and joined the quest for understanding how financial markets work. I am an optimist at heart and I view finance as a new science with much of its underlying true principles still undiscovered. Think back to the medical science of 16th century Europe, when bleeding patients was considered the end all and be all for medical treatments, but certainly was not very effective against most ailments. Today the science of asset valuation is not very effective. These are exciting times, with lots to be done. And lucky for me not as 132 dreadful as 16th century Europe!
The poor operational performance of stock valuation models

There is today a crisis of confidence regarding the tools of asset valuation. However, let me be clear that there are well-understood and indisputably correct foundational principles in finance. But, the lack of success in the application of these principles has led many to adopt a nihilistic view of finance/investment as a science. To start off, the root principle of asset valuation is that the intrinsic value of an asset is equal to the weighted sum of its expected future cash flows.8 Intrinsic means the objective, normal, or sensible value of an asset, based on the assets cash flow and risk characteristics. The weights get smaller the further in time the cash flow is; i.e., they get discounted using the so-called discount rate. This is because people are innately impatient and prefer receiving cash flows today than in the far future. This approach is the well-known Present Value (PV) principle, which, for instance, has been successfully applied to pricing bonds. By extension, valuing stocks should be relatively easy. Stocks are analogous to bonds but with two added features: 1) their cash flows often tend to grow and also fluctuate; and 2) they contain risks unique to equity ownership. However, the application of the PV principle to stocks has been fraught with problems. Essentially, there are two categories of unresolved issues: what valuation formula should be used? And, what are the correct input values? Regarding the first point, any valuation formula must assume something about the cash-flow lifecycle of the company that is to be valued. There are at least a half a dozen different views on how to go about modeling this lifecycle. I am not exploring these views here. The second issue can be further decomposed into: what are the appropriate expected future cash flows? And, what is the value of the weight or discount rate to be applied to each cash flow?9 Most of the current confusion about stock valuation centers on these last two questions. For example, sell-side analysts must report price targets and use these methods to make the case for buy/hold or sell recommendations. However, these and other practitioners simply go through the motions of valuing stocks without having much conviction about their estimates. This is because there is great uncertainty regarding which inputs and formulas to use. But before moving on to address these last two questions, let me tackle the issue of what a good price estimate should look like.
This is broadly accepted as correct, while actually not quite exact. In particular, the intrinsic value of a stock must at least be equal to the liquidation value of its assets in place minus liabilities. It is true that the issue of identifying future cash flows may be tied to the choice of formula for evaluating these cash flows. In particular, the field often uses a fading rate assumption of the cash flow growth to a normal level. For the sake of not crowding out the presentation, I choose to bypass this issue here.

What does a good price estimate look like?

A price target is a forecast based on the current intrinsic value of the stock and a projection of that value in the future. The relationship between the actual market price and the target can be established as follows: Current market price = intrinsic value + noise0 Where, intrinsic value = PV (expected cash flows) + explainable psychological value premia or discounts; noise0 = current [random economic shocks + unexplainable psychological value premia/discounts] Future market price = forecast (intrinsic value) + noise1 Where, noise1 = future [random economic shocks + unexplainable psychological premia/discounts] Here I am modifying the standard view of intrinsic value a little bit, to include any effect related to investor psychology that can be explained and quantified. The key word is explained. A purely random phenomenon while quantifiable is not explainable in the sense that, even if there exists an underlying deterministic mechanism that controls it, it is computationally too costly to figure out. Hence a probability distribution is typically assigned to its outcomes. Even though intrinsic value purports to be derived from objective principles, it does not have to be dissociated from market psychology. For example, a widespread drop in confidence regarding the economy may impact the discount rate used by investors to value a particular stock.10 In other words, the discount rate and the intrinsic value can be affected by psychological biases. It is nevertheless a fine line to walk. The whims of demand and supply can create underinflated prices as exemplified by the valuation of other asset classes such as mortgage-backed securities during the 2008 financial crisis.11 Sometimes fear can lead to fire sale prices, which by any stretch of the imagination do not correspond to a fairly assessed value for these assets. Psychological biases can be lasting and it would be a pity to ignore them and blindly and erroneously accept that reversion to objective fundamentals must happen fast.
surprise analysts with better-than-expected earnings are often rewarded with a ho-hum increase if any. However, the market is punishing stocks even more than usual for earnings disappointments... Part of the problem is fear of the valuation levels that many stocks have reached. With the market at these levels, if stocks are slightly down (in terms of unexpected earnings), they get severely punished. It is fairly well accepted in the profession that financial markets incorporate news very fast [Robertson et al. (2006)]. This is what academics often refer to as the efficient market hypothesis. However, it is the content and interpretation of the news which matter. Variations will arise in peoples assessment of how big and lasting the impact of the news will be on the companys profits. Investors may be able to assess the immediate impact of bad news, but it is harder to know how long the rough patch will last. If your view is more optimistic than the rest of the market, your estimate will come out higher than the market price, and vice versa if your view is more pessimistic. Let me briefly comment on the phrase the market is overreacting. In my opinion it is a misnomer. The reason is that contrary to what is commonly thought, the market is not a collective assessment of what an asset is worth. There is no such thing as a market, rather there is a marginal trader. Overreaction, in this case, is not a characteristic of the entire market. In my example, the marginal trader belongs to the most pessimistic segment of the investor population. Thus, she brings down the price to a level that actually corresponds to her assessment, i.e., one extreme of the distribution of beliefs about this company. This situation will persist as long as the optimists sit on the sideline and do not to outbid the pessimists. Certainly, the spread of disagreements amongst all investors may signal to the optimists that they are isolated in their belief, which obviously is not very useful for any type of Bayesian updating and confirmation. This uncertainty is not necessarily reflected in the discount rate that they use. In that case, they do not view the asset as being worth less than the pessimists (otherwise they would be the pessimists). It might just be a question of finding the right time to invest in the stock or that there is another stock that has similar features but less news uncertainty. In other words, this can be viewed as an instance of market illiquidity. If the disagreement is clearly polarized as in a bimodal distribution, then
Once these factors are taken into account, a possible rule of thumb for the intrinsic value estimate to be viable is that the value gap (current market price minus intrinsic value) should be a function of the disagreement regarding how distressed or fast-growing the firm is. Disagreement between you and the market, but more importantly amongst all investors. Let me explain with an example. Assume that a stock is currently undervalued (value gap negative) because the market is actually overreacting to bad news. The finance literature has documented that negative surprises tend to impact more disproportionately stocks that are already highly valued. Wall-Street Journal columnist Deborah Lohse (WSJ, November 12, 1996 issue, p. C1) notes that: Analysts say that stocks that
the optimists should in principle act fast and bid up the stock. That is because they realize there is a consensus amongst like-minded investors and their view is reinforced. On the other hand, some optimists may
10 I am referring to the market risk premium here, even though the literature has not separated out risk due to economic uncertainty versus psychological fear. 11 A testimonial by the then OTS Acting Director Scott Polakoff about the oversight of AIG in March 2009 in front of the senate banking committee chaired by Chris Dodd is enlightening. He stated that the cash flows to super senior AAA MBSs were essentially sound as of September 2008, and that the reason for the decline in values were linked to the decline in the values of the real estate collaterals. The testimonial can be viewed at http://www.cspanvideo.org/program/Interventiona&start=1598.
133
decide that their belief is so strong that it does not need confirmation by outsiders. Indeed, they may intentionally pursue a contrarian strategy. However in this case, as long as no additional informational content can be extracted from the given news, these investors are really going out on a limb and gamble. Hence, to find out if your stock price estimate is good, consider the following process. If your estimate is say less than 5% off and the firm is not going through any major changes, and there is agreement among investors about the prospects of the firm, you may have found a good estimate and the valuation result could be deemed satisfactory. On the other hand, a gap of 10% or more may be a clue that you need to pay close attention to the latest news about the company and find concurring evidence of significant changes. If the news has a broad range of interpretations, you may decide to use a conservative estimate. However, even this logic can be deceiving. Who says that accu12
to truly know what the price will be in the future we would have to know what the expectations of investors will be in the future regarding the firms profits, growth, and risk. I call this forecasting the forecasters. This is not an impossible task, but errors may compound. In practice, analysts take the current valuation estimate and grow it at the same rate as expected growth in earnings or dividends. This is not ideal, but there is no obvious first best method. In the approach laid out above, I also assume that the forecast is a consensus forecast, and that this forecast is identical to the intrinsic value of the stock in the future, which may not be correct. Adding insult to injury, trying to forecast any systematic bias in market psychology may not be a joyride. Of course, some market anomalies like the value premium puzzle do appear to be persistent, but may not actually be exploitable.13 Other anomalies like the so-called January effect or weekend effect have largely disappeared soon after they were discovered. In the end, practitioners often prefer resorting to simpler and less scientific methods such as relative valuation, which apparently have less uncertainty associated with their inputs. Relative valuation methods such as using benchmark P/E, or enterprise value/EBITDA ratios to value a firms stock can give as good or even better results than present value methods.14 Relative valuation methods are often able to capture information contained in the unknown intrinsic value. The firm might be going through a phase that is hard to model using PV methods, while the peer group may be going through a similar phase, maybe in terms of market share expansion and profit growth. The bottom line is that the present value method is still the correct method, but practitioners hide the issue of uncertainty under the carpet. They do not necessarily feel the pressure to address the problem because everyone is in the same boat, and still, professionals want to project an aura of competence, confidence, and expertise. A line I have often heard and plead guilty of having abused is that valuation is more of an art than it is a science, so that subjective input choices and fudge factors are easier to justify.
racy is better when your estimate is close to actual? As discussed above, it may be the case that a stock price is actually way off from its true intrinsic value because investors are sitting on the sideline, leaving room only for the most pessimistic traders. Another instance is that investors may not be paying attention to that stock, or are selling it to reallocate cash to other more attractive or glamorous stocks. This certainly was the case during the tech bubble period of the mid-1990s when staple companies were quite undervalued (Procter and Gamble, for example) due to a mass migration of money into tech companies. The question then becomes, should you believe that the market price will converge to your estimate quickly and therefore base your trading strategy on that gap narrowing? I think the answer is that it is dangerous to do so, again because news keeps arriving which may invalidate your intrinsic value estimate and that the bias in the market may last longer than you expect. The lesson here is that it is virtually impossible to tell whether valuation results have a very high level of accuracy, because, even if your estimate is close to actual, the true intrinsic value is unknown and there could be deviations due to random noise. When the level of agreement about the news content is high, then you are doing fine. However, with a large amount of disagreement amongst all market participants, there is no alignment regarding the expected direction of prices, and the market price can contain a large random component. In that case, your estimate can be way off, and there is not much you can do about getting more accuracy. On the other hand, doing an inventory of the news surrounding the company and of the degree of disagreement about the content of news between you and the rest of the market can help establish when you are far off the mark. In particular, being able to gauge the psychological state of the market and quantifying it may help you figure out whether your estimate of intrinsic value is a good estimate. The second step for determining the price target can be even more fraught with problems. Now we really are adding another layer of uncer134 tainty to the procedure, as this step is about forecasting. For example,
12 These thresholds are meant to be indicative, not scientifically precise. 13 Phalippou (2008) shows that the majority of investors may not be able to exploit the so-called value premium. The value premium shows up in stocks with low levels of institutional ownership (representing 7% of overall stock market cap). Portfolios sorted on increasing institutional ownership have lower value premia. There are larger arbitrage costs for low institutional ownership stocks. 14 Of course, we are back to the question of how to gauge what a good estimate is.
Cash flows and discount rates

Let us get back to the two thorniest issues, which professionals are still grappling with today. What cash flows and which discount rate(s) should be applied in stock valuation models? Let me start with the simplest issue: The cash flows When teaching the standard stock valuation model, most textbooks begin with the statement that stock prices are equal to the

present value of future dividends. That is not a good start. My students eyes glaze over when I open a lecture on stock valuation with that statement. In their experience, they see that many companies they know do not pay dividends: Google, Starbucks, etc. Typically these are high growth companies. How do we value these companies stocks then? Many educated people believe that the price of a stock should equate the present value of future earnings. I reason out with my students that it is incorrect (unless dividends and earnings are the same). The truth of the matter is that some earnings are reinvested to produce future earnings. So just summing up, discounted earnings amount to doing some double-counting. The key element to understand is that this fundamental valuation principle is correct but that using actual dividends (or forecast based on actual) to compute a stock price is not necessarily what is meant here. Actual dividends and projected past trends may not accurately reflect the future pattern of cash flow payments that will accrue to investors. Especially if these cash flows are disproportionately coming from capital gains. Capital gains essentially reflect expected jumps in future cash flows or growth trends. Economists have realized that when the actual dividends are small, the present value of projected dividends based on actual does not come anywhere near what the price of a stock is. In my experience, when a corporation has a payout ratio of less than 45%, dividends are useless for valuing the companys stock. An alternative is to use the concept of free cash flows to equity.15 This concept measures the potential dividends that the investors would receive if the company decided to pay these as cash and not hoard them. This might be a better measure, but there is a chance that these cash flows are illusory and will never be paid out, at least in the amounts and sequence believed. This is of concern. In fact, companies can be sitting on a pile of cash waiting for the right opportunity to reinvest or acquire other businesses. Thus, projecting future free cash flows using how much is currently in the pile as baseline to estimate future cash flows can lead to an overestimated value. Sometimes, free cash flows to equity can be distorted by debt that is been used to finance dividends or by sudden fluctuations in capital expenditures. Because there is great measurement uncertainty associated with estimating future cash flows, finance professionals may again turn to relative valuation techniques for valuing non-dividend paying stocks. The biggest (academic) blow to the present value (PV) of dividends model came in 1981 when Robert Shiller of Yale tested the ability of the PV model to replicate market volatility. Of course, most practitioners knew by then that using the PV approach did not work well. Shiller pushed another big nail in the coffin of the PV method applied to stock valuation by arguing that ex-post dividends are not volatile enough to account for price volatility. The result was fairly obvious in retrospect, given that corporations have a clear incentive to smooth-out dividends, and if one assumes that ex-post dividends are 100% responsible for price volatility.
Shiller (1981) notes that price volatility might be due to changes in the discount rate. But as he states in the paper: we cannot observe discount rates directly (p. 430), and hence, he does not pursue the analysis.16 Taking a small hiatus from this exposition, I would like to expand on Shillers position regarding the state of finance. This is as good a place as any to do this. Shiller has been one of the harshest critics of the rational approach to understanding asset prices. He argues that economic models used to value financial assets are broken because they are based to the premise that economic agents are rational. The models assume that people optimize their own happiness, typically measured as lifetime consumption, without being influenced by psychological biases. Kahneman and Tverskys (1979) piece was a breakthrough article formalizing the general finding that human beings view the same outcome differently when it is framed in terms of losses rather than gains. One may interpret this approach and subsequent research in behavioral finance as a call to abandon the paradigm of the rational homo economicus in favor of a more psychological approach (homo psychomicus). Behavioral finance aficionados have shifted their attention to incorporating heuristic and judgment biases and studying the impact of irrational decisions on financial markets. But in doing so, they pay attention to the peculiarities of individual trees and lose sight of the driving forces and common patterns that shape the forest. Bringing these aspects of human behavior into focus was a great achievement. However, I respectfully disagree with the notion that because we have not found the correct rational model to explain investors behavior, we should be giving up on the whole enterprise. Similar to how the famous black swan example teaches us about the inductive process, absence of evidence is not evidence of absence. Alternately, one of the tenets of Required Yield Theory is that prices really are the present value of expected dividends, with a twist! I will show later that the volatility predicted by Required Yield Theory is in line with actual market volatility. The discount rate By far, the biggest amount of confusion today surrounds the issue of the determination of the discount rate. I will summarize some of the main ideas out there and shed light on the logical and intuitive inconsistencies of some of the modern views in this field of research. I will divide the discussion into two parts. The first part is a discussion of the way economists conceive of the discount rate (for equity) or what they term the required return or expected return. The second part is about one key input in the discount rate: the so-called equity risk premium, the bane of all financial planners existence!
15 There are other models that can work when dividends are too small. For example, the residual income model of Ohlson (1995) or even for when earnings are negative [Faugre and Shawky (2005)]. 16 More on why changes in the discount rate are the key to the whole thing later.
135
The various (unsuccessful) attempts for determining the discount rate and asset prices
Let me first begin by stating that I have a lot of respect for the great economists and thinkers who have worked hard to advance our knowledge of asset pricing theory, and even though I am presenting a gloomy picture of the state of current knowledge, many of the underlying ideas remain valid, and are gems for us to use in furthering the science.
(the mean) is determined by the desire of investors to be compensated for bearing systematic risk. In other words, the path of stock prices is guided in a statistically measurable way by tracking the path of investors expectations and compensation for risk. To some extent, prices must be related to what investors focus on, whichever it may be: the firms fundamentals or market fads. Could this turn the concept of random walk on its head? No, not quite, because the arrival of new information may force the (log of) stock prices to appear random after all. The process of news arrival keeps yanking expectations around.19 Notwithstanding, a more pertinent question might be: is the information generating process ergodic? Ergodic means that once information is in a particular region there is a positive probability that new information will return in that region. In that case, a region can be defined as the collection of news having similar content. For example, one may ask if a stream of news about a particular corporation has overlapping news content, or is it random in the sense that any event is as likely to be reported about this particular corporation as any other. I think the answer clearly is that in the very short term a stream of news is informationally correlated. A firm operates in an industry. It has a given management and executive team and sells products with specific characteristics. It has a unique narrative associated with it. This will make some pieces of news more likely to appear than others. An absurd example is that a manufacturer of golf balls will not be ascribed a piece of news stating that a consumer got salmonella from ingesting a golf ball. On the other hand, it is within the realm of likely news that a famous golf pro may be endorsing that brand of golf products. The implication is that the pattern of return is not necessarily random, given the ergodic property of news in the short term. By contrast, in the mediumterm news can be expected to arrive more randomly and be weakly correlated with past news. In the long term, with low frequency data, the picture may again shift back to where current news are correlated with past news because the business that survived has progressed along the industrys lifecycle, and again there is a clear narrative associated with it.20 But, let us get back to our discussion of CAPM. Given that investors determine the mean of the distribution of an individual stocks return via a
The static Capital Asset Pricing Model and beta

The Capital Asset Pricing Model (CAPM) is a product of the 1960s. Advances in linear optimization had led Markowitz to come up with his portfolio diversification approach in the 1950s, and CAPM followed suit. Sharpe (1964), Lintner (1965), and Mossin (1966) brought us CAPM. It was a huge leap of understanding. Not all risks are created equal in the equation of financial life, and that (more risk)
(greater expected return). Only sys-
tematic risk, or risk associated with the system, was proven to matter. The underlying idea is that if investors rationally choose to diversify, they will end up holding a portion of the market portfolio (which contains all risky assets, not just stocks but everything that is tradable, including real estate).17 In fact, the analysis rules out the notion that some investors might invest in fewer stocks than the collection making up the whole market! That is, no corner solutions are allowed. Because any security tautologically belongs to the market portfolio, that security should be priced only in relation to its contribution to market risk, i.e., the so-called non-diversifiable risk. Given the scope and ambition of the model, it is only fair to wonder
18
whether CAPM is a good description of reality, or is the model still an embryonic and too idealized a version of the securities markets? There were high hopes for the model when it came out in the 1960s. Following a string of persistent anomalies documented throughout the mid-1970s and 80s, a case was slowly building against CAPM, and in 1992 Fama and French buried CAPM but then brought it back in its resurrected form: the Fama-French Three Factor model (1993). What Fama and French had found in 1992 is that by looking at a cross-section of securities, the beta (which measures the intensity of systematic risk) was not explaining adequately subsequent mean ex-post mean returns. Their strategy was to enhance CAPM so that the new model would predict these empirical anomalies. Still, what can we learn from CAPMs insights? CAPM is a statement about how the average or mean return on a security is related to its standard deviation, the market standard deviation, and the correlation between that securitys return and the markets. CAPM establishes a causal link between investors behavior, i.e., their rational expectations of return and the mean of the statistical distribution of returns for that security. An oversimplistic interpretation is that investors know about the stocks characteristics and rationally anticipate the mean. A more insightful inter136 pretation is that the first moment of the statistical distribution of returns
17 Some approaches also consider non-traded assets, for example Mayers (1973) and Jagannathan and Wang (1996). 18 It is interesting to note that in these approaches the market portfolio is not mathematically characterized to be well diversified. It is implicitly assumed that way, but the CAPM proof does not seem to need that assumption. For instance, during the 2008 financial crisis, the market portfolio was not well diversified: all asset classes returns became highly positively correlated with each other due to financial contagion (except gold). 19 In fact, this is not a new idea as the fundamental price (cum dividends) is often referred to as a martingale. Martingales are less restrictive than random walks. See the brilliant survey by Leroy (1989) on the topic. This is an important feature for RYT as it applies to valuing the S&P 500. 20 This may explain why some researchers find that stock returns appear to mean-revert in the long term. An interesting research project would be to determine a measure of informational content and establish its dynamic statistical properties. A related question is how ergodic the process of good versus bad news announcements is?

rational expectation process, one may wonder whether they are determining the volatility (standard deviation) via the same mechanism. CAPMs implicit answer is that investors do not believe they can, as the standard deviation is taken as exogenous in their optimization procedure. Thus, no inference can be made regarding the optimal shape of volatility. By contrast, the path-breaking research of Engle (1982) on the so-called ARCH modeling of the stock market clearly shows that the standard deviation (or variance) has an autocorrelated structure overtime. Market volatility exhibits a form of momentum. Highly volatile stocks tend to remain that way for a while, and vice versa. Thus, there might be explainable and optimal volatility patterns that CAPM does not account for. Let us continue exploring CAPM and the time dimension. As a static model the standard CAPM is by default an unconditional model; i.e., expected returns are not based on an evolving information set. This assumes that investors are unsophisticated statisticians and do not learn. Investors do not have any underlying dynamic model to use for valuing stocks. They do not try to infer the behavior of returns from the evolution of fundamentals over time, which is a bit preposterous to assume, given that analysts do that all the time. For example, to some degree, the mean stock return should also be related to the return on book value of equity for the firm. Another basic observation is that the arithmetic mean return is not necessarily the best measure of expected return especially when historical returns are negatively serially correlated, meaning that positive returns tend to follow negative returns. In that case, the average compounded return is a better measure. Investors expectations might very well be about average compounded returns, which the CAPM analysis would also be missing, although some adjustments can be made to transform a mean return into a compounded average.21 Even if the statistical mean were the right concept, the sample mean return may differ from investors rational expectations, because it is an ex-post measure computed by averaging returns over time. Ex-post return measures are often contaminated by new information that was not available at the time when rational expectations were formed.22 To address this, some authors have suggested using the conditional form of CAPM. That is conditioning expectations on the relevant information set available. On the other hand, observed returns might be drawn from distributions with changing moments, as seen with ARCH models. Possibly, these changing moments might also be attributable to the firm moving through the various phases of its lifecycle.23 For that reason, empirical tests based on time-series sample mean returns may not even be valid for confirming or refuting the theory.24 Hence, while CAPM has been put into question and replaced by improved models, empirical tests have not necessarily been designed appropriately to provide the final coup de grace. For example, recently Bali and Engle (2010) claim that their enhanced version of the conditional CAPM works well and encompasses the predictions of the Fama-French
three factor model! But more to the point, there is not much confidence attached to using the CAPM beta in stock valuation methods. This is because CAPM has no proven track record of consistent and realistic valuation estimates. But, this might also be because other inputs interfere, even though beta is the correct input. On the other hand, I am yet to meet an analyst who has replaced the CAPM beta with Fama-French betas. It is still not clear that the Fama-French approach can deliver better discount rates estimates than CAPM. The Fama-French factors are quite volatile month-to-month and based on very limited trials. I found that they generate more instability in the discount rate and price estimation. Overall, the lack of positive results when using CAPM over the last fortyfive years and yet-to-see working alternatives has generated a climate of confusion and skepticism in the field of stock valuation.
The first generation of dynamic asset pricing models

A general trend in economics has been to complexify models, often because mathematical knowledge was being disseminated from physics into economics (with about a 20 years lag), and new generations of economists were eager to challenge their peers with their newfound high-tech abilities. This trend has in certain circumstances led to good outcomes, while in other times the results have not been that useful in enhancing our understanding of reality. In the 1960s and 70s, the trend was to extend models from static to stochastic dynamics. Dynamics means the passage of time. Stochastic means adding statistical uncertainty to the model. Obviously, time plays a key role in finance. In many instances, individual households make time-based choices. They decide whether it is worth putting money aside for rainy days or a major purchase down the road, versus enjoying life now, given their current and prospective income. There was a big push in the 1970s to incorporate what was then referred to as micro foundations of macroeconomics and finance. For finance, this meant that any prediction about the behavior of asset prices and markets had to be based on the root motivator of human behavior: maximizing happiness or satisfaction.25
21 Estrada (2010) attempts to reframe CAPM using geometric means. 22 A purer measure of expected return is, for example, Value Line and First Call analysts expectations [Brav et al. (2005)]. 23 Another interesting research topic would be to link the skewness of the firms return distribution to its lifecycle. 24 Another well-known reason why empirical tests of CAPM may not be correct is the so-called Roll (1977) critique, which simply states that the market portfolio is an elusive concept, since it must contain all traded risky assets. While progress has been made to broaden the market portfolio due to faster computing technology, the problem is that using baskets closer to the true market portfolio may never be close enough. 25 This is understood in economics as maximizing the expected utility derived from consumption. Of course, there are many other things human beings may try to maximize that are related to happiness and not to consumption: creative potential and selfexpression, harmony in relationships, the pleasure of viewing a beautiful landscape, being of service to others sometimes through their religious communities, also psychic rewards such as peer recognition and public awards, or simply being accepted by their community and many other possible things. I am not taking issue with the standard macro model, as it is flexible enough to accommodate many arguments. For example, economists have long incorporated things like leisure, human capital, and/or pollution in their formulations.
137
Merton (1973), Rubinstein (1976), Lucas (1978), and Breeden (1979) were at the forefront of building these micro foundations. In particular, Lucas is a master model builder. The trick in his 1978 paper is to assume that the economy is based on fruit-giving trees. The fruits are the dividends. People can invest in trees and in bonds. Essentially, Lucas (1978) derives the famous canonical asset pricing equation, which under some technical simplifying assumptions relates the real expected rate of return on bonds and risky assets (the trees) to the rate of real per-capita consumption (and income) growth and some other parameters describing the mental make-up of the average economic agents, i.e., his/her degree of impatience and aversion to risk. This is the central result of the paper and it also constitutes one of the pillars of RYT. In particular, the concept of required yield itself is tied to income/capita growth. Unfortunately, the empirical tests of Lucas and subsequent models did not mesh with reality, for some reason consumption growth was not volatile enough to account for market price volatility. I am exploring these and related issues next.
consistent with microeconomic behavior. These estimates were drawn from the micro and labor economics literature, and were judged to be ironclad. Thaler (2001) gives an example of the consequences of having a level of risk aversion required to make the model work: If I proposed to you a gamble in which you have a 50 percent chance that your wealth will double and a 50 percent chance that your wealth will fall by half, how much would you pay to avoid the chance that you will lose half your wealth? If you have a coefficient of relative risk aversion equal to 30 [at least three times bigger than consensus estimates], you would pay 49 percent of your wealth to avoid a chance of losing half your wealth, which is ridiculous. This was the puzzle: the equity premium, which compensates individuals for bearing systematic risk, was too big, in view of how relatively insensitive people are to consumption risk in the economy. Of course, these assertions were heavily model-dependent. Without going into detail, there has been a plethora of attempts to solve the puzzle, each installment more complicated than the last, but not necessarily more realistic or more successful.
The equity premium mess

There were high hopes in the early 1980s, because finally, a complete rational micro foundation for investors behavior had been laid out by Lucas (1978) masterpiece. But these hopes were again dashed, as the predictions of the models were not jiving with what macroeconomists understood about peoples aversion to risk and impatience from other studies.
The recent attempts desperately seeking common sense

The attempt to solve the equity premium puzzle has lasted a quarter of a century, with many failures and a few self-proclaimed successes [Mehra (2003, 2008)]. An open-minded view of this process is that it was necessary to go through these iterations of failures to find out what was not working and redirect the researchers efforts towards promising new directions. However, it has been more than 30 years since Lucass model appeared on the scene, and I do not think other disciplines such as modern engineering or medicine could justify efforts that would take that long, without delivering a product that works. It is rather incongruous that no model, either a modified version of Lucas (1978) or other frameworks, has been able to tie a measure of risk to the equity premium in a sensible manner.26 For example, one might find reasonable the idea that the differential in risk (standard deviation) between stocks and bonds should be correlated with the gap in returns. Well, as Cliff Asness (2000) finds out empirically, the answer is not a resounding yes. Recently, McGrattan and Prescott (2000) claim to have finally solved the puzzle: it was due to tax changes. The run-up in the stock market was because the average tax rate on dividends fell dramatically between 1962 and 2000. But the profession has not embraced Prescott and MacGrattans claim as the definitive solution.27 In 2004, Bansal and Yaron came out with an all-gunsblazing model that combined many of the modeling innovations accumulated in the last quarter century. It appeared that the puzzle had been finally cracked. Their approach has since become the Cadillac-model for
The initial puzzle

In 1985, Rajnish Mehra and Edward Prescott recast Lucass (1978) model and expressed the asset pricing equation as a function of the equity risk premium, i.e., the difference between the return on the S&P 500 and a riskless return. The equity risk premium is one of the most widely used quantums in finance. It is an input in many capital budgeting decisions and has been at the center of social security reform debates and actuarial estimates for future pension fund liabilities. Mehra and Prescott (M&P) showed that the equity premium is primarily a function of the covariance between consumption growth and the return on risky assets. In other words, if the covariance is positive, lower than expected future consumption growth is associated with a lower than expected return on equity. This constitutes an increase in systematic risk, and hence the premium rises. The interpretation is that a negative shock to your future income and consumption, for example due to an unforeseen recession, is associated with lower returns on your stock investments. This is a double whammy for you. M&P had been trying to publish this article for about six years, and it was increasingly clear during that time that the Lucas (1978) model was not explaining asset prices well. Their stroke of genius was to couch this failure into a puzzle and get the rest of the finance field excited about resolving it. M&P found that the level of risk aversion needed to 138 make the model work was at least three times larger than the estimates
26 Of course, I would be remiss if I did not mention that we have our own solution for the puzzle. One version is in Faugre and Van Erlach (2009) and another version links the premium to a portfolio insurance motive and the premium on a put option based on the S&P 500 as the underlying asset [Faugre and Van Erlach (2006)]. 27 By the way, Required Yield Theory also gives tax rates a front-seat role in asset valuation.

taking on many outstanding puzzles of finance. Their solution in the 2004 paper? Investors are sensitive to long-run consumption risk. Shocking! It is worth taking a moment to look at their approach. Among other model enhancements, Bansal and Yaron inject a small but highly persistent component in the consumption and dividend growth processes that guide the dynamics of the model. In the simplest of terms, this means, for example, that if long-run real consumption/capita growth is about 1.8 percent in the U.S., which it has been over a 50 year period prior to 2006, investors are worried about whether it could be 1.799 percent or 1.801 percent looking into the future 80 to 100 years from now. These minute changes then affect the riskiness of owning stocks. This is because the equity premium is a function of the covariance between observed equity returns and consumption growth. But, as Sargent (2007) points out: The tenuous part of this argument is how the representative consumer can come to be sure about the presence of these long-run risks when they are so difficult to detect statistically. This is the polite way of saying that there is no way consumers can even dream of grasping these minute changes taking place in the distant future. Beeler and Campbell (2009) make the same point in a salient deconstruction of Bansal and Yarons approach. An even more recent and stranger literature integrates the principle of uncertainty or ambiguity in decision making. Ambiguity refers to situations where investors do not rely on a single probability distribution to describe the relevant random variables. Ambiguity aversion means that investors dislike ambiguity about the data generating process of asset returns, not simply that the return prospects have greater variance. To quote Epstein and Schneider (2008): The agents information consists of (i) past prices and dividends and (ii) an additional, ambiguous signal that is informative about future dividends. Our setup thus distinguishes between tangible information (here, dividends) that lends itself to econometric analysis, and intangible information (such as news reports) that is hard to quantify yet important for market participants decisions. We assume that intangible information is ambiguous while tangible information is not. This approach generates several properties of asset prices that are hard to explain otherwise. In markets with ambiguous information, expected excess returns decrease with future information quality. Indeed, ambiguity-averse investors require compensation for holding an asset simply because low quality information about that asset is expected to arrive. This new approach, based on Knightian (1921) uncertainty, is making some headway in resolving the equity premium puzzle, because the added uncertainty risk is now priced in equities [Leippold et al. (2004)]. Al-Najjar and Weinstein (2009) make the point that in laboratory experiments ambiguity aversion could essentially be indistinguishable from situations where subjects misapply heuristics that serve them well in real-world situations. Samuelson (2001) discusses in detail how infrequently encountered situations such as laboratory settings lead to
inappropriately triggered analogies, which can account for framing effects and other behavioral anomalies in experiments. More to the point though, while an interesting idea, just determining by how much the equity prices must be further discounted, necessitates a knowable model of the risk associated with managing the multiple probability distribution assessments, and what those multiple probability assessments might be. Once again it is doubtful that investors would actually go through such steps consciously or unconsciously. In other words, how can the reliability aspect of the signal occupy a separate box in investors brains, distinct from the signal itself? Rather, it would seem more reasonable that investors directly use their own assessment of the reliability of information conveyed in the signal as it may apply to valuing the prospects of the firm, and update the return distribution accordingly, by making it more risky. The ambiguity framework can lead to the absurd result that a signal, say about a company sponsoring a golf tournament, which is economically inconsequential and at the same time has an extreme level of ambiguity, leads investors to discount the price of that stock. On the other hand, when simply taken at face value, such a signal with little reliability may just lead to no action taken.28 So essentially we are back to square one.
Trying to make sense of the stock market: Required Yield Theory

The core of Required Yield Theory as applied to valuing the S&P 500 is a notion that was first introduced in a formal way by Irving Fisher in 1896. The idea is very simple and is often presented as the foundation for the determination of interest rates in money and banking courses. Investors seek to earn a real return after expected inflation. The real return is itself determined by the 1) productivity of capital, which includes technological progress, and 2) by the rate of impatience of people in the economy. For Fisher it was clear that the real return had to be positive. Eighty years later, Darby (1975) and Feldstein (1976) were pointing out that personal marginal tax rates can also act as value depreciators and thus raised expected returns. Our version of the Fisher hypothesis is that investors desire to earn a real return after expected inflation and marginal taxes on dividends and capital gains. In our paper [Faugre and Van Erlach (2009)], we establish that this real return must be equal to long-run GDP/ capita growth, i.e., about 2% in the U.S. when investors invest in a broad index such as the S&P 500. We call this real return the Required Yield. In the rest of this section I go through the steps we used in our (2009) article to demonstrate that point.
28 Interestingly, Required Yield Theory accommodates for the presence of fear in the pricing of an equity index. The notion of the equity fear premium that I will introduce later is flexible enough to include fear due to Knightian uncertainty. The approach I use is to quantify the effect of the fear on the equity risk premium. I do not try to determine the cause of fear, or the mechanism by which it enters the mind. I am also careful to distinguish between the equity premium and the fear premium, which the literature confounds.
139
The role of marginal tax rates in stock market valuation

A good place to start is to establish that marginal tax rates matter in stock market valuation. For the sake of illustration, I use a simple experiment and examine the impact of the Bush tax cuts of 2003 on the S&P 500. TheJobs and Growth Tax Relief Reconciliation Act of 2003was passed by Congresson May 23, 2003 and signed into law byPresidentBushon May 28, 2003. The top capital gains rate of 20% was reduced to 15%. The top rate on dividend income was reduced from 35% to 15%. A basic question is whether this Act had an effect on the S&P 500 inverse P/E ratio or not? The expectation is that the forward E/P should fall around announcement and/or after enactment dates. That is, consistent with the after-tax Fisher effect, and everything else constant, investors are willing to earn lower expected return on their equity investments when marginal tax rates are expected to drop. Here I assume that the forward earnings yield (inverse P/E) is a good proxy for expected returns. It is clear from Figure 1 that the effect was very pronounced around these dates, as the percentage decline in the earnings yield was large relative to surrounding days.29
16%
5% 4% 3% 2% 1% 0%
11/30 12/10 12/20 12/30 1/9 1/19 1/29 2/8 2/18 2/28 3/10 3/20 3/30 4/9 4/19 4/29 5/9 5/19 5/29 6/8 6/18
-1% -2% -3% -4% -5% -6% 1/7/03 Bush announces tax plan 3/19/03 Bush launches Iraq invasion 5/28/03 Tax plan becomes law
Figure 1 Bush tax cuts announcements in 2002-2003 percentage change in S&P 500 E/P ratio (five-day moving average)
Correlation
14%
SP500 E/P
12%
1yearEIn ation GDPDe
SP500 E/P 1 year inflation GDPDefl
SP500 E/P 1 year inflation 100.00% 84.69% 100.00% 69.21% 79.91%
10%
8%
6%
4%
The impact of (expected) inflation on equity returns

While it has been well documented, the interpretation of the correlation between the trailing E/P and inflation has been controversial [Sharpe (1999, 2001)]. The common explanation in the literature is that the E/P ratio is a real return [Ritter and Warr (2002); Siegel (2002)] and that investors suffer from inflation illusion [Modigliani and Cohn (1979)]. In our view, later expanded in this essay, the correct explanation is that the Fisher Effect holds for equity returns. That is, the forward E/P ratio is an expected nominal yield and hence it is affected by expected inflation. Figure 2 shows that the S&P 500 forward earnings yield has been highly correlated with measures of inflation and expected inflation over a 50year period.30 The respective Spearman correlations of the forward E/P is 69% with actual GDP deflator and 85% with expected inflation. This is not a proof that our explanation is more correct than the inflation illusion explanation. Only that there is a simpler and equally valid explanation: that of the after-tax Fisher effect.31
2%
0% J-52
J-54
F-56
M-58
M-60
A-62
M-64
M-66
J-68
J-70
J-72
A-74
A-76
S-78
O-80
O-82
N-84
D-86
D-88
J-91
J-93
F-95
M-97
M-99
A-01
M-03
M-05
J-07
Figure 2 S&P 500 forward earnings yield (inverse P/E) versus actual GDP deflator and one-year expected inflation (quarterly 1954-2006)
expected dividends collapses to an expression that only depends on next periods expected earnings per share and expected return. In that case, the expected return is also called the capitalization rate. The implications of this result are far-reaching, as it makes it possible to side step all the confusion surrounding the fact that dividends are not volatile enough to explain prices because 1) earnings are more volatile than dividends (albeit not much more), and 2) at any point in time, each price corresponds uniquely to one discount rate, which implies that the volatility of prices can be explained by the volatility of point-to-point discount rates. A useful feature of the stock market (S&P 500) that makes it easier to
Why valuing the market might be easier than valuing individual stocks
Growth opportunities play a key role in stock valuation. Simply stated, they represent the value of activities that the firm undertakes and which produce a greater internal return than investors require from the business. When growth opportunities are positive, businesses tends to grow fast and generate abnormally high profits, and vice versa, when growth opportunities are negative. The knife-edge case is when the present value of growth opportunities is zero. This means that a business cannot find any investment for which they can outpace their own investors expected return. An important result in finance is that when the present value of 140 growth opportunities is zero, then the intrinsic value or present value of
value than individual stocks is the mean-reversion of growth opportunities to zero at the macro-corporate level. The corporate sector cannot in
29 A shortcoming of Figure 1 is that does not show if these effects were reversed later on. Of course, a drop by x% would have to be made up by a gain of more than x% to revert back to previous levels. At any rate, reversal is hard to confirm because other macro variables such as corporate earnings growth start interacting with the valuation in the intermediate term. Sialm (2009) is an attempt to integrate and measure the effect of taxes on equities, but with limited applicability to stock valuation methods. 30 Faugre and Van Erlach (2009) show that over that period, the S&P 500 earnings yield is stationary around a constant term at a 5% confidence level. Inflation and expected inflation are also stationary over the sample (respectively at the 5% and 10% levels). Hence, using Spearman correlations is meaningful. 31 Of course, Brennan (1970) was one of the first to integrate the effect of taxes in required returns by analyzing an after-tax version of CAPM.

aggregate outpace the economy. Firms belonging to a mature cross-section of the economy cannot generate abnormal high profits sustainably, nor can they be lagging the economy for long periods of time.32 Hence, at any point in time, the value of the market index cannot stray very far from the capitalization rate times expected earnings. This is great news for us, because this result makes valuing the S&P 500 a simpler affair, as long as we can determine the capitalization rate independently.
investment opportunities elsewhere. It is up to you.33 The government realizes that removing these free cash flows would not impede internal firm growth. It could easily decide to confiscate them as taxes on realized equity returns. What would happen if the government wanted to confiscate more than dividends (as taxes) is that investors would buy and hold stocks, thus postpone capital gains. Firms would stop paying dividends, so that these would accumulate as cash. The government would lose tax revenues for the foreseeable future. On the other hand, the best way for the government to seize these revenues directly would be to increase the corporate tax rate. But, by confiscating more than dividends, the government would impede corporate internal growth and threaten the ability to raise future tax revenues. This is assuming that the government is wise enough to anticipate the detrimental effects of predatory taxes. Accepting this premise, from whatever direction taxes are coming, a portion of dividends must be left in the pocket of investors after all taxes have been taken out. Another way to characterize this political equilibrium is to say that the dividend payout ratio is always greater than the blended marginal tax rate, assuming that earnings per share are a fairly good proxy for (dividends +capital gains). We find in the 2009 paper, that this pattern has held empirically from 1953-2006 for the S&P 500. The conclusion from the above string of arguments is that the after-tax real expected return will be equal to at least 2% for a long-term investor. Nevertheless, the full proof that 2% is the required return is not complete. The next step is to find out what happens to the short-term investor. In particular, how is it that 2% is a real after-tax return point-to-point? And how is the equity risk premium incorporated in the analysis?
The required yield: 2% as an absolute minimum after-tax real equity return?

In the 2009 paper, we first demonstrate that at the macro level a longterm investor in the S&P 500 will earn at least 2% real return after-tax. This is so because 1) aggregate corporate earnings of a representative group of mature firms that constitute a relatively fixed portion of the economy will grow at the same rate as GDP; and 2) the growth in share ownership must be closely tied to population growth in the long term. This implies that earnings-per-share will grow at the same rate as GDP/ capita in the long run. Because the market P/E ratio is also constant, the capital gains rate is equal to earnings per share growth or GDP/capita growth, which has been about 2% in real terms in the U.S. over 19292006. Thus, a long-term investor using a buy and hold strategy can earn at least a real 2% compound return on a pre-tax basis. To obtain this result on an after-tax basis we need to address the interplay between government tax policies and corporate dividend policy. In our 2009 article, we assert that after-tax dollar dividends are always large enough to pay for taxes on capital gains. Put differently, dividends themselves are more than enough to meet the tax burden on total dollar equity returns. This is not an easy case to make. To step back a little, we need to assume that the political equilibrium has evolved in the U.S. to produce the following outcome: the corporate and personal tax system (on capital gains and dividend income) must be designed not to cannibalize economic growth. It makes sense that it would be so, but this is not a standard argument in economics these days. Within the current established tax system, investors can always pursue a buy and hold strategy, so they do not pay capital gains taxes until they realize the gains in the far future. In addition, U.S. section 1014 of the Internal Revenue Code of 1954 allows for a stepped up basis of unrealized capital gains upon transfer of stock ownership to descendants. Thus, investors can pass along their investment from one generation to the next, without being taxed on unrealized gains. But, the government could have had a different view about equity income taxation. Let me explain. Dividends are free cash flows that firms choose not to reinvest because these would not boost internal growth in a meaningful way. Firms in that position are sending the following message to their shareholders: we cannot earn the returns you are expecting from us if we reinvest these cash flows. We are giving them to you because you may find better
The equity premium, business cycle risk and the required yield
At some point in the research that culminated in our 2009 article, we came to a crossroad. It looked like the asset pricing literature treated Irving Fishers view of interest rate determination and the CAPM approach as schizophrenic views of how stock returns are determined. Maybe a more accurate characterization is that the literature on equity pricing did not really attempt to integrate the Fisher effect in the standard asset pricing model [Two partial attempts are found in Boudoukh and Richardson (1993) and Boudoukh et al. (1994)]. One of our many ah-ha moments was the realization that they are both correct views of reality, but in a different
32 The fact that the S&P500 is a survivor biased index helps in that respect. The argument goes further in our (2009) paper, in which we explain that dividend policy enhances this phenomenon. 33 At the macro level, it is possible that cash dividends are reinvested in other ventures, which may contribute to raising average economic growth, only if the new ventures are cash-starved companies with successful growth opportunities. I am assuming that these cash starved companies are raising the bulk of their equity capital via the standard IPO or SEO market. Also as mentioned before, corporations tend to smooth dividends out, and sometimes use debt to make up for earnings shortfall. However, in aggregate for the S&P 500, dividends are paid out of earnings. Furthermore, in our model we do not have to assume that dividends are smoothed out to achieve our main result.
141
way than shown in modern expositions. Let me explain. Think back to that money and banking course and recall that the so-called Fisher effect is that interest rate = real interest + expected inflation. It is common in modern textbooks to find an apology for Fishers oversights. He did not mention the inflation risk premium nor did he make provision for marginal tax rates and additional risks due to the nature of the asset (risky bonds or equity) in his formula.34 The modern Fisher effect is often presented as: Interest rate = real interest + expected inflation + inflation risk premium (if needed) + extra risk premia (default, etc.) We can also assume that the interest rate is after tax, to complete the model. In that sense, there is no contradiction between the insight of CAPM and Fishers equation as the equity risk premium can be accommodated easily. But, by an interesting series of back and forth between theory and empirics, we were forced to take a different route. Empirically, we were finding that no matter how we sliced it, adding a sustained equity premium to the 2 percent real return was making any preliminary model based of the logic of RYT worse in terms of fitting the data. This truly was the inductive phase of our research. We decided to trust what the data was trying to tell us, i.e., that 2 percent is a real after-tax real return earned by investors even in the presence of the equity premium. By contrast, it is clear that Fisher (1896) thought that the real interest rate varies for various horizons or even individual by individual. But, it is constant in the long run (see his 1896 article, p. 91). The question we contemplated was: how could we theoretically determine that the 2 percent was an absolute benchmark for the stock market real after-tax return? Again, an obvious candidate was the long-term constant return that is also the focus of standard dynamic asset pricing models, except that again the existing models include an added risk premium. As seen previously, the long-term investor can get 2 percent after-tax real at minimum, if she stays in. The 2 percent value is a return that is already net of the corporate failure rate in the economy; hence risk-free in the long run. Of course, if a long-term investor buys at the peak of a market bubble, her return may fall slightly short of 2 percent in the long run. So the outcome of getting 2 percent per year is not independent of the initial bid and entry point. However, it will be less and less dependent on the initial bid when the investment horizon is extended to thirty years or beyond. Long-term investors compete amongst themselves and against shortterm investors for buying the S&P 500. Hence, the highest bidder must be willing to accept the lowest possible expected return next period, which is 2 percent. Why? First, as explained above, the value of the S&P 500 index at time t has the useful property that it is determined to a great extent by the capitalization rate and expected earnings at t+1, because of the mean-reversion of growth opportunities to zero. It is also logical that long-term investors would not bid the index to get less than 2 percent 142 next period because then they could not guarantee that this loss would be
made-up in the future. If a short-term investor were to add a risk premium in excess of the 2 percent real after-tax return, he/she could never be the highest bidder, as long-term investors can outbid the index to get exactly 2 percent. On the other hand, short-term investors can always set out to replicate long-term investors behavior. Thus, short-term investors do not have any incentives to bid higher and receive less than 2 percent after inflation and taxes, given that in addition, they face short-term risk. In each period, the new highest bidder behaves accordingly under the belief that future bidders will do the same. What anchors that belief is that nominal growth opportunities are quickly mean reverting at the stock market level. Given these bidding behaviors, we set the equity premium as: Equity risk premium (after-tax real) = 2 percent minus the after-tax real yield on a given Treasury (the latter net of inflation risk premium) As we considered the above concept, it appeared to us that there was an inherent source of ambiguity in the literature regarding the equity risk premium: what Treasury yield should be used to measure the premium? This is not a trivial issue. For instance, why is the premium larger for short-term bonds, than it is for long-term bonds? The standard answer in the literature is that the term structure of interest rates makes it so. We thought about it differently. What if the equity premium varied according to the investors hedging horizon? Maybe investors have preferred habitats in terms of the maturity of Treasuries they use to pull out of risky assets. This inquiry led us into considering the notion of business cycle risk. We conjectured that investors trying to hedge over a short time horizon, must principally worry that productivity growth (as a source of returns) may not rise back to where it is forecasted to be over that span of time. This led us to create this next definition: Business cycle risk = risk that mean-reversion of productivity growth may not happen over relevant hedging horizon We had to show the empirical relevancy of this concept. We computed the business cycle premium variable as the difference between a productivity growth target that assumes mean-reversion and an estimate of expected book value per share growth for the S&P 500 over the relevant horizon. The horizon for this estimate corresponds to a given Treasury maturity. In the 2009 paper, we use the 1-year, 10-year, and 30-year Treasuries. We find that the actual yield spreads between the longer and shorter maturity bonds are empirically explainable by this new variable. In
34 Fisher (1896) was not oblivious to risk. In his concluding section, he states (p. 88) that appreciation or depreciation of money as introducing a third element into the rate of interest. This element is to be added to or subtracted from the sum of the other two elements, which are a payment for capital (or the rate of interest proper) and a payment for insurance, [emphasis added]. And in a footnote on page 92, we cannot strike out the insurance element as a mere additive term with which the theory of interest proper has no concern. A complete theory has yet to be written.

the long run, the S&P 500 and the 30-year Treasury appear to deliver a 2 percent return after-tax and inflation, point-to-point. Hence the business cycle premium is zero in the long run. This makes sense because a longterm investor should realize that there is zero risk that mean-reversion of productivity growth would not happen in the long run. Henceforth, this business cycle risk premium variable appears to logically and empirically account for the bulk of the equity risk premium on an after-tax and real basis. This is because the equity risk premium on an after tax and real basis is the difference between 2 percent (the after-tax real expected return on the S&P 500) and the after-tax real yield on shorter-terms Treasuries, which is equal to the business cycle risk variable. This was a nice breakthrough. We then decided to take a look at periods when the after-tax real 30-year Treasury yield actually fell short of the 2 percent mark. Why did this happen? A natural explanation was about flight-to-safety, in other words a manifestation of widespread fear on the market.
1 0,5 5,5 5
VIX Close
4,5
TFP
4
3,5
2,5
1,5
A-04 J-04 S-04 D-04 F-05 M-05 J-05 S-05 D-05 F-06 M-06 J-06 O-06 D-06 M-07 M-07 A-07 O-07 J-08 M-08 J-08 A-08 O-08 J-09 M-09 J-09 A-09 N-09 J-10 A-10
A behavioral twist? The equity fear premium

One of the novel aspects of the 2009 article is that we find a way to incorporate fear in the valuation analysis. Investor fears is captured via flight-to-safety behavior. In particular, we assert that fear is present in the market when the 30-year Treasury yield falls below its normal level, which we say is 2 percent real after-tax. Investors scared of holding risky securities dump them and buy long-term Treasuries. They are willing to sacrifice some yield on the Treasury side in order to hedge risk. At the margin, when the influx into Treasury ceases, investors must be indifferent between holding long-term Treasuries and equity. In other words, the amount by which the yield has risen on the equity side must balance out the loss of yield on the Treasury side. In the 2009 paper, we assume that these yield movements are symmetrical on an after-tax and real basis. We then take the shortfall gap and attach it to the after-tax real required return on the equity side to compute the discount rate. In my follow-up article [Faugre (2010)], I complement this approach by adding a flightto-liquidity premium, which is based on the shortfall of the target federal funds rate from the one-month T-bill yield. The total fear premium is then couched as the sum of the flight-to-safety and flight-to-liquidity premia.35 For example, in that paper [Faugre (2010)], I find that during the 2008 financial crisis the Spearman correlation between the VIX and the equity fear premium is 71 percent during the 2007-2010 period, versus 14 percent over the 2004-2007 period, on a daily basis. The VIX is considered by many as the investors fear gauge [Whaley (2008)]. Figure 3 illustrates how in synch the two indexes became in the midst of the crisis. It turns out that the two main advantages of using the fear premium over the VIX are that 1) the fear premium is determined independently from the S&P 500, whereas the VIX is not. This means that the VIX cannot help to value the S&P 500 concurrently. 2) The fear premium accounts for downside volatility, whereas the VIX does not distinguish between upside or downside, although clearly very high values of the VIX are associated with downside risk [Ambrosio and Kinniry (2009)].
Figure 3 Total fear premium (TFP) versus VIX 2004-2010 daily observations (index base =1 in July 27, 2004)
Treasuries arbitrages and the RYT valuation formula

As shown in previous sections, the required yield of 2 percent is simply the minimum real after-tax return that investors can earn by investing in the S&P 500 for a one-year horizon. The required yield already contains the market risk premium associated with the business cycle. As such it should in principle be greater than the risk-free rate. What is the risk-free rate? Recall that our analysis is done on a real after-tax basis. As such, real Treasury yields are not risk free. For example, even the one-year Treasury yield that is risk free on a nominal basis is subject to uncertainty regarding the value of future inflation and taxes. The key aspect of the required yield is that it is a minimum yield. Investors will arbitrage between the asset classes that provide the highest minimum real aftertax return for the same level of risk. For the S&P 500, the required yield is also subject to same uncertainty regarding future inflation and taxes. Hence, the risks are the same for the index and the 1-year Treasury. In the 2009 article, we also include the 10-year Treasury in the analysis, because it is consistent with a measure of expected one-year yield, which accounts for mean reversion of productivity.36 If the required yield were always greater than the 1-year and 10-year Treasury yields, the issue of arbitrage would be moot. However, there might be instances where these T-yields in fact outpace the required yield. There are three possible case scenarios: 1) tax rates on interest income are dropping significantly relative to equity tax rates ceteris paribus; 2) the 1-year real after-tax Treasury yield rises because of yield curve inversions or drop in short-term inflation expectations; and 3) the 10-year real after-tax Treasury yield rises because of a boost in the long-term inflation risk premium.
35 For details regarding the construction of these variables, see this note available at http:// www.albany.edu/~faugere/DataforRYT.pdf. Daily estimates for the S&P 500 are also posted at @SP500Forecast. 36 Even though it is subject to price risk.
143
In Faugre and Van Erlach (2009), we mathematically derive the key valuation formula as a function of the forward earnings yield, based on the arguments and assumptions covered above: eE /Pt t+1 = [Max{RE , r1 , r10(t+1)}] t+1 (t+1) [(1 t+1)[1 + (1 c t+1) AEGE ]] i,t+1
The valuation formula above is fully consistent with the after-tax Fisher effect [Darby (1975); Feldstein (1976)], as it expresses the forward earnings yield as a function of a real return plus expected inflation, on a before-tax basis. More importantly, this formula is strictly based on the PV of dividends principle, with the caveat that PVGOs mean revert to zero quickly.
Where RE = g + E + Ft+1 is the nominal required yield plus the fear t+1 t+1 premium. The nominal required yield is the sum of g (long-term real GDP/ the next year and the fear-based premium Ft+1 > 0. The variables r1 and (t+1) yields. The parameters c t+1 and t+1 respectively are the marginal capital gains tax rate and the marginal blended (income and capital gains) tax rate. The variable AEGE is the capitalized expected abnormal earnings growth i,t+1 component, which can be positive or negative, depending on whether the reversion of growth opportunities to the mean occurs from above or below. It measures the deviation of the before-tax required return away from
37
Empirical success of Required Yield Theory for valuing the S&P 500
In the 2009 paper, we compare the performance of the RYT model with a version of the Fed model [Lander et al. (1997)] popularized by Yardeni (2002): eE /Pt = 10-year Treasury yield. It turns out that the Fed model t+1 was one of the most successful models to mimic the behavior of the S&P 500, from 1979 up until the end of 2002.38 In the article, we test that the forward earnings yield is stationary.39 We find that on a quarterly basis, the adjusted R-squared is 74 percent for the Fed model versus 88percent for RYT over the period Q4 1953- Q3 2006. Given that expected earnings data for the S&P 500 became available after 1979, we also find the adjusted R-squared is 87 percent for the Fed model versus 94 percent for RYT over the period Q1 1979- Q3 2006. Figure 4 gives a good visual of the differences between the two models. We concur with Shiller (1981) that ex-post dividend fluctuations do not play a major role in explaining market price fluctuations. Rather, at low frequency (quarterly data) we find that S&P 500 prices fluctuate mostly in response to changes in expected earnings, inflation rates, and marginal tax rates. The latter two variables directly impact the required return on equity via the Fisher/Darby/Feldstein effect. Because investors appear to price the index based on wanting to earn a real return equal to a constant real long-term GDP/capita growth after tax, we do not need to appeal to permanent changes in dividend growth as an explanation for stock market volatility. Based on our model, the calculated yearly standard deviation of S&P 500 total returns over the period is 26 percent versus 27percent actual. More recently in Faugre (2010), I tackle the issue of expanding the fear premium concept and tracking its value on a daily
capita growth rate at 2 percent) plus E the expected inflation rate over t+1
r10(t+1) respectively are the real after-tax yields on the 1-year and 10-year T-
the expected sustainable return on equity (ROE) for the index.
16% 15% 14% 13% 12% 11% 10% 9% 8% 7% 6% 5% 4% 3% 2%

S&P 500 fwd EY Fed Model RYT assuming 2.21% LT growth
D-53 A-55 S-56 J-58 M-59 O-60 F-62 J-63 N-64 A-66 A-67 D-68 M-70 S-71 F-73 J-74 N-75 M-77 J-78 D-79 A-81 S-82 J-84 J-85 O-86 F-88 J-89 N-90 A-92 A-93 J-95 M-96 S-97 F-99 J-00 N-01 M-03 A-04 D-05 A-07
Figure 4 S&P 500 forward earnings yield RYT formula versus Fed model (quarterly observations 1954-2006)
basis. Using interpolated daily forward (operating) earnings estimates, I am able to construct daily estimates going back to July 2004.40 Because I include flight to liquidity behavior in the fear premium, I am able
1700 1650 1600 1550 1500 1450 1400 1350 1300 1250 1200 1150 1100 1050 1000 950 900 850 800 750 700 650
RYTEstimate SP500
J-04 A-04 A-04 N-04 F-05 M-05 S-05 D-05 M-06 J-06 O-06 J-07 A-07 A-07 N-07 F-08 J-08 S-08 D-08 M-09 J-09 O-09 J-10 M-10 A-10 N-10 F-11 J-11 S-11 D-11 A-12
144
Figure 5 RYT estimates for the S&P 500 versus actual (daily from July 2004 to November 2011)
37 Hence the two regimes I =1, 2. The proof of the RYT valuation formula appears in Faugre and Van Erlach (2009). There are related valuation methods that assume some reversion of earnings growth, such as Ohlson (1995); Dechow et al. (1998) Lee et al. (1999), Lee and Swaminathan (1999), Bakshi and Chen (2005) and Ohlson and Juettner-Nauroth (2005). 38 The Fed model started misbehaving after that, for reasons unknown to the literature. RYT actually explains that when interest rates drop beneath the required yield, the required yield then becomes the basis for the discount rate to apply to the equity index. 39 Essentially, we use after-tax real variables. In the 2009 paper, our tests reject nonstationarity of these time series at the 1% level for most variables and 5% level in a couple of instances. 40 This initial months significance is because I need data on the 20-year TIPS, which became available only after that date.

to track closely the behavior of the S&P 500 during the 2008 financial crisis. Figure 5 shows the performance of the RYT model to explain the S&P 500 price on a daily basis from July 2004 till November 2011. The mean absolute percentage deviation on a daily basis from July 27, 2004 till November 4th 2011 was 5.8 percent.
(2001); Siegel (2002); Asness (2003); and Campbell and Vuolteenaho (2004)] generally believe that the (forward) earnings yield is a real return.42 However, it is a well-known result [Reilly and Brown (2006)] that the forward earnings yield is equal to the total return when the present value of growth opportunities is zero. Our result indicates that we can view the earnings yield as the total nominal return, had the S&P 500 not produced these abnormal positive returns due to falling tax rates and inflation rates. Hence, the size of the ex-post equity premium (difference between expost market return and a Treasury yield) is also explainable in light of the drivers of the large historical ex-post returns: downward trending taxes in absolute and relative terms (favoring dividends), and a steady decline of actual and expected inflation.
Why are ex-post market returns and the equity premium so large?
Some readers may still be puzzled as to why we argue that the stock market has essentially returned a real 2 percent after tax, given that expost real after-tax returns appear to have vastly exceeded 2 percent. In fact, the constant for the after-tax real total S&P 500 return is estimated in our 2009 paper at a value of 5.23 percent over Q4 1953 - Q3 2006. However, the explanation is consistent with RYT. Using the RYT formula above, the before-tax ex-post market total return can be rewritten as:41 (1 t+1)(g + E )] 1, where Dt+1 is the dividend next period, and all otht+2 er notations are as before. This last equation shows how it is possible to have witnessed such a large average ex-post equity return over the last fifty years. Essentially, the second term on the right hand side indicates that for a given rate of EPS earnings growth close to GDP/capita growth, the historical combination of declining blended tax rates and declining inflation expectations is the main reason that ex-post capital gains were abnormally high. Furthermore, the dividend yield was also affected by tax trends. The relative dividend income tax rate declined in comparison to the capital gains tax rate by about 1.96 percent per year on average, over the period Q4 1953 - Q3 2006. This effectively boosted the dividend yield by that same amount to its historical average of 3.73 percent (calculated based on one-year forward expected earnings). Firms were still able to increase their nominal dividend payouts without sacrificing sustainable growth. These effects on ex-post capital gains rates and dividend yields may contribute to the empirical findings that ex-post returns have been found to vary inversely with changes in expected inflation rates [Bodie (1976); Nelson (1976)], since trends in inflation and blended tax rates (as well as dividend tax rates relative to capital gains) were all correlated over the period. Notwithstanding, in the absence of such trends the average stock market return would have been capped by the (adjusted dividend yield) + EPS growth. The adjusted dividend yield is (3.73% - 1.96%). EPS growth equals long-term nominal GDP/capita growth at about 5.61% = 2.03% + 3.58% (inflation). In other words, the average return would have been about (3.73% - 1.96%) + 5.61% = 7.4% versus the 11.6% compounded average that was observed over the period Q4 1953 - Q3 2006. Interestingly, this adjusted estimate is very close to the average forward earnings yield over the period, which was 7.7 percent. The fact that the ex-post total compound return minus inflation turns out to be 8.0 percent, close to the forward earnings yield, is the main reason why observers [Ritter RE = Dt+1/Pt + (Pt+1- Pt)/Pt = Dt+1/Pt + [e E / e E [(1 t+2)(g + E ) t+1 t+2 t+1 t+1
Insights and takeaways for investors

Here I try to distill the essence of RYT as it relates to trading the S&P 500.
The inverse P/E ratio and the effects of inflation and taxes on the S&P 500 index
First of all, RYT confirms that the relevant variable to keep an eye on is the inverse forward P/E ratio, or forward earnings yield, based on S&Ps operating earnings; a measure which suppresses the effect of one-time extraordinary charges. The earnings yield is akin to an expected return, and takes into account a multitude of factors: 1) the impact of expected inflation and tax rates; 2) the reversion of growth of corporate book value to average economic growth; 3) the possibility of yield curve inversions and high long-term yields as floor expected returns for investors; and 4) the psychological effects of fear on the market. RYT teaches us that when inflation expectations rise or tax rates increase, the earnings yield also rises. This means that the market price will drop, given earnings. When interest rates are low, like they are in the current environment, the behavior of the earnings yield is mainly dictated by what happens to the required yield (in nominal terms and including a psychological premium). The best time to invest is on the tail-end of a high inflation period, prior to the Fed pursuing a restrictive monetary policy, similarly, when taxes are high and are expected to be reformed downward.43 Essentially, this constitutes a transfer of wealth from existing investors to new investors, above and beyond the capacity of the market to produce the required yield. This was, for example, the situation during the bull market of the 1980s till 2000.
41 This is assuming that PVGOs are zero. 42 We present a formal counter argument to the inflation illusion in the (2009) paper, based on a simple manipulation of the Gordon (1962) growth model. 43 However, the caveat is that this chain of events must happen in a way that there must be some residual uncertainty. Otherwise, these reforms would instantaneously be reflected in higher prices, with no chance for new investors to capture these gains.
145
Main drivers day-to-day and the impact of Fed policy on the stock market
It is clear that day-to-day the main drivers of the S&P 500s movements are the inflation risk premium and the 30-year Treasury, which reflects the fear premium. In fact, the daily fluctuations in these variables are enough to produce the necessary price volatility. A drop in the inflation risk premium or the fear premium lowers the required earnings yield and thus constitutes an upward movement for the market. In an environment where the nominal yields on the 1-year, 10-year, and 30-year Treasuries are all below the nominal before tax required yield, movements of the 30-year Treasury yields are inversely related to movements in the index. This is due to the fact that the fear premium impacts the 30-year Treasury yield. On the other hand, as the federal funds target is effectively 0 percent nowadays, the flight to liquidity premium is nonexistent. But, if the Fed decided to raise short-term interest rates, we might witness a flight to liquidity which would have an adverse effect on the S&P 500 index. The Feds reducing short-term interests does not have any significant effects on the stock market, unless short-term rates are greater than the nominal before-tax required yield. So for example, if the Fed decides to avert an oncoming recession based on the understanding that an inverted yield curve is a precursor of such event, the stock market will not grow in proportion of the interest rate drop. This is because the required yield will become a binding constraint, which will level the market off, as long as the other risk premia are not moving. The Fed pursuing a policy of lowering long-term rates (quantitative easing) may impact the market in ways that investors and the Fed might be puzzled about, because they are missing the linkage between the fear premium and the S&P 500. For example, if investors are fearful, but the 30-year Treasury is overpriced because of the Feds excess demand for these instruments, investors will turn to other shelter investments such as gold. On the other hand, when the price of gold is bumped up to higher levels, investors are unable to cheaply hedge against risk and will discount the price of the S&P 500 even more. Another case scenario is when investors are willing and able to pay higher prices than the Fed, then quantitative easing will be ineffective, in the sense that the Fed should not spend taxpayer money, because long-term interest rates will drop anyways. This is due to investors being on the same side of the market as the Fed, but the Fed not fully factoring in that flight-to-safety is taking place, and that investors are setting the equilibrium price. On the other hand, even though investors are fearful, if they expect quantitative easing to end at a set date, speculators will short Treasuries, which will sterilize the Feds policy. Hence, long-term rates can appear to remain stuck, even though the Fed is attempting to drive the rates down. The fear premium might be present, but is not reflected in the gap between the required yield and long-term rates anymore.44 Bottom line: the market may drop further than anticipated by investors. The Fed may think that this is 146 because the rates are still too high, but that would be a misunderstanding
on their part. Of course, they may decide to alter the terms of their quantitative easing program, with more detriment to financial markets, if this effect keeps being misunderstood. What if the RYT formula concludes that the market is undervalued? Should you buy? No, not necessarily. The gap between the estimate and actual does not have to narrow over time. Again, the simple reason is that news arrive which throw investors expectations off kilter. However, based on my experience running the model, rightly or wrongly, I can say that inflation expectations and growth expectations seem to be more stable than inflation and fear risk premia. But that might just be the result of how the model has been implemented so far. Expected earnings fluctuate, but all in all they are very clearly trending upward. My view is that even though macroeconomic forecast revisions (surprises) about employment or capacity utilization or some other economic indicators do impact the market, they do so in a way that often can only be justified by unrealistic revisions in expected earnings or growth (maybe less so for the latter). The problem with revised macro data is that investors trust in the forecasts is shattered, as they realize they were misled.45 This puts in question not just what they believed but also how they approach future macroeconomic releases. A sense of uncertainty arises which makes them overreact. Often, the market comes back to previous or new highs fairly rapidly in a matter of weeks. In the context of the data, the S&P earnings forecasts issued about every week are usually fairly insensitive to these revisions. On the whole, in my experience, it turns out that RYT S&P 500 price estimates converge again to actual based on these insensitive earnings forecasts, which confirms that some temporary market overreaction was happening. But let us not make that a general rule. Should you follow Buffetts wisdom and buy when others are fearful? The premise of this view is that fear is transient, and that once people come to their senses, valuation levels will pick right back up. But as the financial crisis exemplifies, the fear premium has been high since fall 2008, and has been sustainably high at this point for more than three years. RYT helps to recognize when low P/E ratios do not indicate that fear is present in the market. If you are a long-term investor, however, it makes sense to buy, because in the long term fear will subside. So buying for the long term when the fear premium is really high, is indeed a good idea.
What RYT does and does not do

RYT gives a snapshot of why the market is valued the way it is, using concurrent inputs (i.e., expectations captured as of now). RYT is not a
44 This is a limitation of RYT. 45 The Fed and Treasury do not realize the great cost they inflict on the economy by being way too often wrong in their forecasts, and having a cavalier attitude about revisions.

predictive theory, only an explanatory one. To become a predictive model, we need to find good forecasts for all inputs, which is a different ball of wax altogether and completely unrelated to how good the model is as an explanatory theory.
help or destroy the welfare of others as that of a physician. It is about not getting sucked into the lure of ego and arrogance and yes, even the cynicism associated with the race to get name recognition and acceptance by peers. It is about finding what your authentic drive is, where contribution may lie, so you do not become a robot, blindly applying techniques that your teachers taught you, when they made you feel that you had to be worthy to be let in the inner sanctum. The bottom line: get clear about what your calling is. Finally, a problem that plagues the profession, known for years, is the amount of sheer nonsense that gets published in top economic and finance journals. Many practitioners have long understood that reality. They deride academics for producing a vast amount of useless knowledge. Listen to your inner voice/intuition when it tells you that even though you are awestruck as you receive instruction from famous professors, sometimes the emperor truly has no clothes. You should not surrender your brain and willpower to their worldview without critical appraisal and more importantly, you should be asking questions and thinking about problems that matter to you and the world. Take a stand. How can I help resolve these pressing and relevant issues? Very few individuals, some with famous names have followed that calling. They inspire me. Does this research by an ex-classmate, teacher, or colleague make sense? Do they try to justify the lack of applicability of their result by dazzling you with their technical prowess? Do you base your research on common sense or arcane models with no basis in reality? The time is upon you to engage in the process of creating value for society.
What should you expect to earn over the next 40 years?

Barring any major political or economic upheaval and given that Western central banks are committed to stable long-term inflation targets of around 2 percent, and hoping that tax schedules are kept fairly unchanged, we are looking at a nominal expected return on the S&P 500 of around 5% = (2.03%+2%)/(1-0.194), where 19.4 percent is the average blended (dividends and capital gains) marginal tax rate over the period 2004-2011. This value is quite a bit higher than current long-term Treasury yields at about 3 percent.
The future of RYT and a call to freshly graduated economics and finance Ph.Ds
It should be clear to the reader by now that I strongly believe that there are unifying principles behind the behavior of asset prices. RYT is a new approach that rests on well-known and accepted economic principles. By contrast with the mainstream literature, we have been able to reconcile these principles with the actual behavior of the S&P 500. Notwithstanding, RYT is not a panacea. It is a burgeoning theory showing some striking empirical successes and which can and should be improved upon. One aspect that needs improvement is explaining market bubbles. In that respect, the approach followed in Faugre (2010) might be promising. The approach is to extract premia from markets for specific financial instruments. These premia are defined as gaps between equilibrium prices and off-equilibrium points, purely explainable by psychological biases. In particular, finding where a premium for greed might be hiding remains an exciting challenge. We are also still working at merging RYT with the standard asset pricing economic model la Lucas (1978) with heterogeneous agents. This is not a futile endeavor, as we believe that RYT falls within the framework of the consumption asset pricing model. But it is far from straightforward. Another direction of research is to apply this approach to other G7 stock markets, to show that the logic of a required yield tied to real GDP per capita growth extends to these economies, with varied equity risk premia and return mean-reversion properties. Lastly, a few words if I may, to some of you who are budding economists or just graduated finance Ph.Ds. Two key phrases for you to think about: desire to make a difference and listening to your own intuition or common sense. First, I believe that not enough academic economists set out professionally to make a difference in the world. Whether one actually ends up achieving that, is a different matter. However, striving for that goal does provide a particular perspective. It is not about getting a job or publishing in A-level journals, although these are important. It is about thinking of yourself as a professional with the same capacity to
References

Al-Najjar, N. I. and J. Weinstein, 2009, The ambiguity aversion literature: a critical assessment, Economics and Philosophy, 25, 249-284 Ambrosio, F. J. and F. M. Kinniry, 2009, Stock market volatility measures in perspective, Vanguard Investment Counseling and Research Asness, C. S., 2000, Stocks versus bonds: explaining the equity risk premium, Financial Analysts Journal, 56:2, 96-113 Asness, C. S., 2003, Fight the fed model, Journal of Portfolio Management, Fall, 11-24 Bakshi, G. S. and Z. Chen, 2005, Stock valuation in dynamic economies, Journal of Financial Markets, 8:2, 115-151 Bali, T. G. and R. F. Engle, 2010, Resurrecting the conditional CAPM with dynamic conditional correlations, Working Paper, NYU School of Business Bansal, R. and A. Yaron, 2004, Risks for the long-run: a potential resolution of asset pricing puzzles, The Journal of Finance, 59:4, 1481-1509 Beeler, J. and J. Y. Campbell, 2009, The long-run risks model and aggregate asset prices: an empirical assessment, NBER Working Paper # 14788 Bodie, Z., 1976, Common stocks as a hedge against inflation, Journal of finance, 31:2, 459470 Boudoukh, J. and M. Richardson, 1993, Stock returns and inflation: a long-horizon perspective, American Economic Review, 83:5, 1346-1355 Boudoukh, J., M. Richardson, and R. F. Whitelaw, 1994, Industry returns and the Fisher Effect, Journal of Finance, 49:5, 1595-1615 Brav, A., R. Lehavy, and R. Michaely, 2005, Using Expectations to Test Asset Pricing Models, Financial Management, 34:3, 31-64 Breeden, D. T., 1979, An intertemporal asset pricing model with stochastic consumption and investment opportunities, Journal of Financial Economics, 7, 265-296 Brennan, M., 1970, Taxes, market valuation and corporate financial policy, National Tax Journal, 23, 417-427 Campbell, J. Y. and T. Vuolteenaho, 2004, Inflation illusion and stock prices, American
147
148
Economic Review, 94:2, 19-23 Darby, M. R., 1975, The financial and tax effects of monetary policy on interest rates, Economic Inquiry, 13, 266-276 Dechow, P., A. Hutton, and R. Sloan, 1997, An empirical assessment of the residual income valuation model, Working Paper, University of Michigan and Harvard University Engle, R. F., 1982, Autoregressive conditional heteroskedasticity with estimates of the variance of U.K. Inflation, Econometrica, 50, 987-1008 Epstein, L. G. and M. Schneider, 2008, Ambiguity, information quality, and asset pricing, Journal of Finance, 63, 197-228 Estrada, J., 2010, Geometric mean maximization: an overlooked portfolio approach? Journal of Investing, 19:4, 134-147 Fama, E. F. and K. R. French, 1992, The cross-section of expected returns, Journal of Finance, 47:2, 427-465 Fama, E. F. and K. R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics, 33, 3-56 Faugre, C., 2010, The equity fear premium and daily comovements of the S&P 500 E/P ratio and Treasury yields before and during the 2007 financial crisis, UAlbany Working Paper Faugre, C. and H. A. Shawky, 2005, A valuation formula for high technology firms that are in the early stage of their lifecycle, Advances in Financial Planning and Forecasting, New series, 1: 101-130 Faugre, C. and J. Van Erlach, 2005, The price of gold: a global Required Yield Theory, Journal of Investing, 14:1, 99-111 Faugre, C. and J. Van Erlach, 2006, The equity premium: consistent with GDP growth and portfolio insurance, Financial Review, 41:4, 547-564 Faugre, C. and J. Van Erlach, 2009, A Required Yield Theory of stock market valuation and treasury yield determination, Financial Markets, Institutions and Instruments, 18:1, 27-88 Feldstein, M., 1976, Inflation, income taxes and the rate of interest: a theoretical analysis, American Economic Review, 66, 809-820 Fisher, I., 1896, Appreciation and interest, New York: Macmillan for the American Economic Association, as reprinted in Fisher (1997), vol. 1 Gordon, M., 1962, The investment, financing and valuation of the corporation, Homewood, IL: Irwin Hein, S. E. and J. M. Mercer, 2003, Are Treasury inflation protected securities really tax disadvantaged? Working Paper 2003-9a, Federal Reserve Bank of Atlanta Jagannathan, R. and Z. Wang, 1996, The conditional CAPM and the cross-section of expected returns, The Journal of finance, 51, 3-54 Kahneman, D. and A. Tversky, 1979, Prospect Theory: an analysis of decision under risk,Econometrica, 47:2, 263-292 Knight, F. H., 1921, Risk, uncertainty, and profit, Boston and New York: Houghton Mifflin Lander, J., A. Orphanides, and M. Douvogiannis, 1997, Earnings forecasts and the predictability of stock returns: evidence from trading the S&P, Journal of Portfolio Management, 23, 24-35 Lee, C. M. C. and B. Swaminathan, 1999, Valuing the Dow: a bottom-up approach, Financial Analysts Journal, 55:5, 4-23 Lee, C. M. C., J. Myers, and B. Swaminathan, 1999, What is the intrinsic value of the Dow: a bottom-up approach, Journal of Finance, 54:5, 1693-1741 Leippold, M., F. Trojani, and P. Vanini, 2004, Learning and asset prices under ambiguous information, Working Paper # 249, National Centre of Competence in Research Financial Valuation and Risk Management LeRoy, S. F., 1989, Efficient capital markets and martingale, Journal of Economic Literature, 27, 1583-1621 Lucas, R. E., 1978, Asset prices in an exchange economy, Econometrica, 46: 1426-1446 Lintner, J., 1965, The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets, Review of Economics and Statistics, 47:1, 13-37 Mayers, D., 1973, Nonmarketable assets and the determination of capital asset prices in the absence of a riskless asset, Journal of Business, April, 46, 258-267 McGrattan, E. R. and E. C. Prescott, 2000, Is the stock market overvalued? Federal Reserve Bank of Minneapolis Quarterly Review, 24, 20-40 Mehra, R., 2003, The equity premium: why is it still a puzzle? Financial Analysts Journal, 59:1, 54-69 Mehra, R., 2008, Handbook of the equity risk premium, Elsevier, Amsterdam, The Netherlands Mehra, R. and E. C. Prescott, 1985, The equity premium: a puzzle, Journal of Monetary Economics, 15:2, 145-161 Merton, R., 1973, An intertemporal asset pricing model, Econometrica, 41, 867-887 Modigliani, F. and R. A. Cohn, 1979, Inflation, rational valuation and the market, Financial Analysts Journal, 35, 24-44 Mossin, J., 1966, Equilibrium in a capital asset market, Econometrica, 34, 768-783 Nelson, C. R., 1976, Inflation, and rates of return on common stocks, Journal of Finance, 31:2, 471-483
Ohlson, J. A., 1995, Earnings, book values, and dividends in security valuation, Contemporary Accounting Research, 11, 661-687 Ohlson, J. A. and B. E. Juettner-Nauroth, 2005, Expected EPS and EPS growth as determinants of value, Review of Accounting Studies, 10:2-3, 349-365 Phalippou, L., 2008, Where is the value premium? Financial Analysts Journal, 64:2, 41-48 Reilly, F. K. and K. C. Brown, 2006, Investment analysis and portfolio management, 8th ed. Mason, OH: Thomson-Southwestern Ritter, J. R., 2001, The biggest mistakes that we teach, Journal of Financial Research, 25, 159-168 Ritter, J. R. and R. S. Warr, 2002, The decline of inflation and the bull market of 1982 to 1999, Journal of Financial and Quantitative Analysis, 37, 29-62 Robertson, C., S. Geva, and R. Wolff, 2006, What types of events provide the strongest evidence that the stock market is affected by company specific news? Proceedings of the fifth Australasian Conference on Data Mining, 61, 145-153 Roll, R., 1977, A critique of the asset pricing theorys tests part I: on past and potential testability of the theory, Journal of Financial Economics, 4:2, 129-176 Rubinstein, M., 1976, The valuation of uncertain income streams and the pricing of options, Bell Journal of Economics, 7, 407-425 Samuelson, L., 2001, Analogies, adaptation, and anomalies, Journal of Economic Theory, 97:2, 320-36 Sargent, T. H., 2007, Commentary, Review, Federal Reserve Bank of St. Louis (July/August), 301-303 Sharpe, S. A., 1999, Stock prices, expected returns, and inflation, Washington: Board of Governors of the Federal Reserve System Sharpe, S. A., 2001, Reexamining stock valuation and inflation: the implications of analysts earnings forecasts, Review of Economics and Statistics, 84:4, 632-648 Sharpe, W., 1964, Capital asset prices: a theory of market equilibrium under conditions of risk, Journal of Finance 19, 425-442 Shen, P. and J. Corning, 2001, Can TIPS help identify long-term inflation expectations? Economic Review, Fourth Quarter, Federal Reserve Bank of Kansas City Shiller, R. J., 1981, Do stock prices move too much to be justified by subsequent changes in dividends? American Economic Review, 71:3, 421-436 Shojai, S. and G. Feiger, 2009, Economists hubris: the case of asset pricing Journal of Financial Transformation, 27, 9-13 Sialm, C., 2009, Tax changes and asset pricing,American Economic Review, 99:4, 135683 Siegel, J. J., 2002, Stocks for the long run, New York: Third Edition, McGraw-Hill Thaler, R. H., 2001, Theoretical foundations I, Equity Risk Premium Forum, November 8, 2001, AIMR Whaley, R. E., 2008, Understanding VIX, Working Paper, Owen Graduate School of Management, Vanderbilt University Yardeni, E., 2002, Stock valuation models, Prudential Financial Research, Topical Study #56 (August 8th) Yardeni, E., 2007, Valuation chart book, Yardeni Research Inc., (March 20th).
PART 2
Our Understanding of Next Generations Target Operating Models

Andreas Andersen Principal Consultant, CAPCO Frankfurt Nicolas Faulbecker Associate, CAPCO Frankfurt
Abstract
The current situation in the world economics requires a rethinking of business strategies. The worldwide crisis in the financial segment increases the pressure on every company. All companies have to lean their operational processes and their high risk activities to ensure customer satisfaction. They really have to find new ways of operating the business. Current operating models have to be enhanced and should be more innovative. The future challenges to successfully manage a company necessitate a restructuring and rethinking. In our point of view, companies have to concentrate on their key qualities and services and need to slim down their organizational and operating structure. The paper starts with
describing common rationales of functional models which are essential for setting up the individual target operating model. After that, the target operating model will be enhanced with innovative aspects from the organizations corporate governance, human resources, and lean management point of view. This consolidated mix will build a framework, which fulfills upcoming challenges. At the end of the day, our approach illustrates of how the usage of a conventional target operating model could be stepwise transformed to the next generation in order to finally establish a self-optimizing organization system. It concludes that the prospects seem to be quite promising even though there are no significant experiences in practice yet. 149
The reference operating model based on functions and roles and responsibilities
Many financial institutions regularly examine their operating model and operating processes to optimize trade flows and to reduce costs. This fact offers opportunities to reconsider the whole functional and technical landscape in order to rebuild the existing operating model into a new target operating model (TOM). The first step to implementing a TOM is to set up a reference operating model, in order to create a benchmark for future activities. This reference model should cover best practices from the market on the one hand and individual experiences on the other. It will outline a unique functional and organizational structure by using model definitions equivalent to market standards and descriptions using universal terminology. This will contribute to the successful usage of the reference model by ensuring group-wide applicability. The starting point for creating this model is to describe and design the as-is state using functions and infrastructure. This creates transparency in the organizations infrastructure for every unit of its current functions. In detail, each functional area can be connected to its corresponding software solutions in order to derive an application landscape. This reference operating model will provide functional and technical requirements for the development of the new TOM. Next to these requirements, individual strategic principles and general business needs have to be defined in parallel (see steps two and three in Figure 2). Strategic principles help to provide a definition for the individual understanding of quality and efficiency. This will control the definition of general rules and guidelines for the use and deployment of resources and assets across the enterprise.
Strategic business needs reflect the definitions of the overall business strategy (i.e., volume, growth targets, and product launches). This will highlight the dependencies of the global target operating model (GTOM) of the company itself. These four components will build the base of the design of the GTOM. The GTOM connects the reference model, the as-is situation, and the future requested functions to one unit. The corresponding general reference model is transformed through the determined parameters into a stable solution. It conveys future course of actions to realize the TOM (i.e., centralizing, sourcing options) and develops an understanding on their degree of coverage. This outcome can be used as a decision template for the implementation of the target model and delivers the basis for specifications in the proposal towards possible services or system providers.
New generation of target operating models

Additional aspects incorporated into the classic TOM
The classic TOM concentrates on functions, including roles and responsibilities, and IT structures. In order to capture the overall evolution of the financial industry and the progression of requirements regarding the risks (market and organizational risks), costs, and resources, the classical way of generating and implementing TOM is not sufficient and needs to be enhanced. This enhancement should focus on how a target operating model may act as a leadership approach. This leadership approach concentrates on basic functions such as standardization or sourcing strategies, on the one hand, as well as on soft factors like human capital, on the other.
Focus on basic functions

The former part of the above-mentioned approach focuses on functions,
Analysis
Organizational unit
2
Strategic principles
3
Strategic business requirements Design
Product Function
Country
1
Capco reference model
5
Design global target operating model
6
Roll-out methodology & Roadmap
4
Legal entity
Capturing the as-is world
System
150
Figure 1 The six dimensions of TOM
Figure 2 The six steps of the TOM approach

roles, and responsibilities. Functions are located in specific areas and are related to asset classes or product classes. This step prevents a global or holistic view of general optimization capabilities. In complementation to that, the analysis of functions and processes should be enhanced by incorporating all areas of the financial institution (i.e., front and middle office, risk, finance, marketing, etc.).
can be used to identify topics, actions, and tasks which are similar in several different functional environments and which can be potentially harmonized and standardized. It is worth mentioning that this may cause some operational issues due to different transaction details or different technical requirements. In order to guarantee the correct and complete rebuilding of processes
First, processes and functions that are in different areas and that can be combined due to similar functional requirements and contents (i.e., cancellations and amendments) should be treated separately and should be classified within area independent processes. At this starting point a stepwise approach might be appropriate, starting with standardization and ending in sourcing strategies (Figure 3). The field of time-consuming amendments and cancellations of existing trades is a good example for the potential improvement opportunities of this kind of enhancement to the classical TOM approach. Trade amendments and cancellations are mainly focused on asset classes and IT systems. Each product has its own specific actions and issues, which are mainly caused by different IT systems. Processes are usually grown in the past and are orientated towards existing organizational structures and functional dependencies. Consequently, standardization processes
using standardization, the processes can be automated by using methods like workflow management or system migrations on the IT side. The automation of processes using workflow management procedures reduces the number of manual tasks within the process flow. Moreover, it supports employees with a stepwise execution of actions using new interfaces to affiliated IT systems. These automated features are constantly monitored in order to track all activities within the work cycle. As a second benefit, the workflow management automates the progress of processes and automatically indicates responsibilities and progress reports. These reports should be based on measurable figures and should result in key performance indicators in order to build standardized success metrics. This ensures transparency and enables financial institutions to monitor efficiency, performance, timelines, and resource capabilities. All these actions and optimizations are intended to create future opportunities to optimize the whole process landscape and cut operating costs. Sourcing is one of the most efficient ways to reduce operational costs and risks. The application of a well-proven sourcing approach which supports financial institutions with their best fit sourcing strategy might be the most promising policy (Figure 4).
Process standardization
Process automation
De nition of success metrics
Utilization of TOM as a leadership approach

In addition to focusing on basic functions, the role of TOM as a comprehensive leadership function need to be increased. The new generation of TOMs should combine functional, technical, and organizational aspects, including human capital and customers.
Sourcing strategies
Figure 3 Focus on basic functions
Human capital is one of the most important elements of any organization

Organization sourcing journey
Strategy
Market scan
with significant influences on performance, quality, and risks. Thus, the

Governance
Governance model Contract review, benchmarking Sourcing health check
Implementation
generation of a new central role for human capital regarding operational models might be reasonable. This new unit should directly report to the CIO and COO in order to demonstrate that the firm is giving credence to the contributions of human capital towards optimization and performance improvements. One of the key factors of human capital performance is willingness and motivation. Consequently, financial institutions are urged to create an environment where employees are given the opportunity to self-actualize
Sourcing activities
Sourcing feasibility analysis Sourcing strategy
Supplier evaluation (RFI/RFP)
Sourcing process pipeline strategy Target Operating Model (TOM) design, implementation & execution
Thought leadership
Capco research
Sourcing best practices
Industry experience/ knowledge
Supplier relationships
and, at the same time, are able to identify themselves with the companys corporate identity. As a consequence, companies should generate valid incentives in order to increase motivation. 151
Figure 4 Sourcing approach
First of all, changes within the process landscape should be pronounced as possibilities of change regarding performance and individual career development of each employee. Additionally, each employee is advised to provide suggestions for improvements with respect to performance enhancements or reformed processes. These suggestions should be rewarded or promoted to assign recognition to innovative employees. Next to the potential increase in the level of motivation among employees, the selection of products offers potential optimization opportunities. First, products for which inherent operating processes are either lean or have potential optimization opportunities should always be favored to complex products which involve additional overhead. Second, discussions with customers regarding their choice of transactions should be pushed in the direction of which trade flow requires the lowest manual work. This behavior should result in rewarding or promoting employees, too. New product initiatives should be handled by a very straightforward new product process which includes well informed employees in a very central function and high quality checks by the CIO and COO. The promotion of employees in existing career models may not satisfy each individuals expectations. For that reason, new innovative and forward-looking career models have to be generated. Employee skills are the most important factors of success. Consequently, the career plan of each area should involve part-time engagements in unfamiliar areas (i.e., traders or middle office employees should be engaged part-time in back office areas to understand the processes in detail and to be able to evaluate impacts of new product initiatives on back office processes). All these initiatives will encourage the transparency and efficiency of organizational changes and benefits and may have influences on the efficiency and leanness of front to back processes as it is described in the next section.
optimization stream and acts like Master Black Belts and Champions in the Six Sigma definitions. This new entity is used to serve as a connector between the operative side and the CIO/ COO. Applying the DMAIC (define measure analyze improve control) process management method [DeLurgio and Hays (2004)], the teams goal is to create a self-optimizing organization system which is flexible and transparent for all participants. The SWAT team performs several tasks. First, it directly reports to the CIO and COO, respectively, in terms of budgeting plans, cost minimizing, and timing issues. Second, it serves as a brain pool for continuous process optimization and innovation. Thus, high levels of authority and autonomy are crucial for the team. Third, the team is responsible for constantly motivating and educating staff. This can be done by remodeling incentives systems, for instance. As a consequence, processes in and among business units become more lean, fast, and reliable. It is also vital to establish another team that works alongside the SWAT team which is assigned to control, verify, and appraise the SWAT teams actions (solving the problem of who monitors the monitor [Sivaramakrishnan and Kumar (2007)]). This is essential to limit the SWAT teams excessive influence and to sustain a division of power within the organization. Consequently, the monitoring team contributes to providing the right incentives across the company, as it is told in principal agent related literature. Finally, it is likely that the SWAT team will be perceived more positively by the operative side, if there is a second authority. Thus, a more cooperative and productive way of collaboration is established. At the end of the day, the coordinated mix of Six Sigma methodology application combined with lean management and a continuous improvement
Creation of a self-optimizing organization system

Having integrated the above mentioned steps, it is possible to aim for six sigma standards. This approach is focused on two goals. It comprises high level quality targets via statistical techniques to measure process variation, on the one hand, and the implementation of a lean-thinking company mindset, on the other. The first aspect targets the organizations processes to improve quality. The latter focuses on efficiency, or process cycle times and process speed. As trading speeds increase and competition levels intensify, both quality management and architecture efficiency become essential to remain competitive in this complex environment.
De ne Measure
Control Analyze
Improve
In order to reach that aim, additional functions will be introduced that solely account for the organizations quality and efficiency management. 152 In detail, a SWAT-team will be implemented that leads the companys
Figure 5 DMAIC process cycle

task force provide the best chances of success for a self-optimizing environment within financial institutions.
Conclusion
This paper provides an overview of the methodology needed for the creation and implementation of a new generation target operating model within financial organizations. First, the typical features and methods of TOM are defined. The approach comprises different business units/ functions with its respective IT landscapes as well as several consecutive stages which have to be passed in order to set up the final model. These stages range from the creation of a reference model to the final methodology rollout with its strategic principles and business requirements within the company. Nevertheless, the conventional model approach leaves room for further optimization. Thus, a new perspective is introduced, which is rooted in our understanding of corporate governance and lean management and transforms the classical TOM into a new generation target operating model which also serves as a leadership approach. The new model is broadened with aspects like the education and motivation of staff, on the one hand, and the prioritization of manual effort-reducing products, on the other. Finally, it is feasible to strive for six sigma standards. A SWAT team is introduced that is responsible for the companys learning efforts and, at the end of the day, for the creation of a self-optimizing organization system within the institution. It is worth mentioning that this approach can be applied in a flexible way, occasionally, in order to meet financial institutions needs. With its comprehensive parameters our methodology appears to be more fruitful than existing model designs. Consequently, it would be quite interesting to track its performance in practice.
References
DeLurgio, S. A. and F. Hays, 2004, Using six sigma to improve credit and financial management competitiveness, The Credit and Financial Management Review, 4th Quarter 2004, Columbia Sivaramakrishnan, S. and P. Kumar, 2007, Who monitors the monitor? The effect of board independence on executive compensation and firm value, Available at SSRN: http://ssrn.com/ abstract=988543.
153
PART 2
The First Line of Defense in Operational Risk Management The Perspective of the Business Line
Udo Milkau Head of Strategy and Market Development, Operations / Services, Transaction
Banking Unit, DZ BANK AG, and part-time lecturer, Goethe University Frankfurt, House of Finance
Frank Neumann Head of Operational Excellence in Operations/Services, DZ BANK AG
Abstract
The Basel Committee wrote in its update of the Principles for the Sound Management of Operational Risk in June 2011 that the first line of defense is business line management. This underlines the continuous development from (ex-post) operational risk controlling to (ex-ante) operational risk management. From the point of view of a business line, this active operational risk management can be achieved by reuse of concepts from quantitative quality management together with a fundamental understanding of the underlying power law characteristics of operational risk loss distributions. In an integrated cycle, five steps are combined. 1. Definition of an operational risk strategy (to accept and manage, to reduce, to mitigate, to avoid) and awareness for specifics of operational risk. 2. Measurement of operational risk (loss distribution) with special regard to the asymptotic tail of the distribution and the limits of measurement. 3. Analysis of measured operational risk indicators with consequences for the business line. 4. Management process for the implementation of improvements along the whole hierarchy. 5.Controlling of results achieved by the business line management and, respectively, critical review of the fundamental assumptions of the entire approach. This active approach to operational risk management takes into account that a business line such as transaction banking with, for example, more than four billion high-volume payment transactions and more than seven million international and high-value transaction per year brings along operational risk generically, which has to be managed starting from the first line of defense. Consequently, the essential success factors are a broad understanding of the fundamental features of operational risk, open communication, and a permanent learning process for staff and management of the business line. 155
The Basel Committee published its update of the Principles for the Sound Management of Operational Risk [BIS (2011)] in June 2011. It reflects the extended knowledge and experience in operational risk management since the first version of the Sound Practices for the Management and Supervision of Operational Risk [BIS (2003)] in February 2003. In particular, this update points out that common industry practice for sound operational risk governance often relies on three lines of defense (i) business line management, (ii) an independent corporate operational risk management function and (iii) an independent review. From the perspective of a business line, i.e., the first line of defense, this viewpoint emphasizes an extension from (ex-post) operational risk controlling to include (ex-ante) management of operational risk. This paper describes the contribution of this the first line of defense to the whole framework of operational risk (OpRisk, for short) management of course, together and aligned with the other two lines. It is beyond the scope of this paper to discuss the various aspects and issues concerning controlling and quantitative modeling of operational risk, which have been discussed for more than a decade: from King (2001) and Cruz (2002) to recent overviews such as Gregoriu (2009) and Embrechts and Hofert (2011). Nevertheless, the whole framework of OpRisk including loss data collection exercises, quantitative impact studies, statistical methodologies, and best practices from other industries can be applied at the first
line of defense and (re)used for the day-to-day defense against losses resulting from problems in internal processes, people and systems, external events, and more and more obvious from complex correlation of internal issues and external events. One of those valuable assets in the framework is Basel Committees Results from the 2008 Loss Data Collection Exercise for Operational Risk (LDCE2008). As for the rest of this paper, only selected aspects, sometimes simplified and tailored for pragmatic implementation, will be exploited for the business line OpRisk management. But already an initial look at the aggregated OpRisk loss data provides insight into the essentials for the business line:
There is an obvious correlation between severity S of events (loss due to OpRisk events) and frequency F(S) of those events at least on the basis of averaged data.
This correlation can be fitted with a power law1 distribution F(S) S-. Power law distributions can be found in many empirical data including many destructive phenomena from earthquakes, fires, wars, terrorism, and avalanches to distribution of the mass of asteroids (created by destructive collisions) and fragments in heavy ion collisions [Clauset et al. (2009) and Milkau (1991)]. When severity S and frequency F(S) are plotted in a double logarithmic diagram (see Figure 1 / top) the power law distribution is a straight line with log10(F) = const log10(S).
107 106
The distribution of gross losses depending on severity (see Figure 1 / bottom) shows that every magnitude bin of severity contain approximately the same gross loss and the tail (with unknown possible maximum loss and a cut-off due to the limited observation time of only few years) contains even more gross loss.
Frequency (log)
105 104 103 102 101 <0.02 m 0.02-0.1 m 0.1-1.0 m
Fit F(S) = c * S-
In general, power law distributions have been described as fingerprints of

1.0-10.0 m 10-100 m >100 m
self-organized criticality (SCO), a concept introduced by Bak et al. [Bak (1988, 1993, 1995)]. SCO describes complex systems of many (nonlinear) interacting participants at the edge between order and disorder, where an outside impact creates spontaneous organization. One could argue that this is not given for OpRisk, as especially fraud is a single event without any interactions. But even the fraudulent activities of a rogue trader at UBS were not isolated events. In a memo reported by the
Gross losses in mln (lin)
25000 20000 15000 10000 5000 0 <0.02 m

1
0.02-0.1 m
0.1-1.0 m
1.0-10.0 m
10-100 m
>100 m
Severity (log)
For details, please refer to the LDCE2008 publication. The frequency-severity-distribution is plotted in double logarithmic scales, while the total amount-severity-distribution is a lin-log plot. Remark: as the lowest bin in the LDCE2008 data collection is defined < 20000 (and not < 10000), there is a slight inconsistency with fully logarithmic binning and the lowest severity bin contains a fraction of events and of gross loss, which would belong to a second bin from 0.01 mln to 0.1 mln.
Wall Street Journal on Oct. 5, 2011 [WSJ (2011)], UBSs Sergio Ermotti is quoted as saying: Risk and operational systems did detect unauthorized or unexplained activity but this was not sufficiently investigated nor was appropriate action taken to ensure existing controls were enforced. This
156
Figure 1 Distribution of number of losses (frequency) and gross loss amount by severity of loss reported by 2008 LDCE participants
At least, the asymptotic tail of the distribution can be fitted with a power law. For the lower end more sophisticated distribution like Generalized Pareto Distribution may be more accurate, but the asymptotic tail contains the most severe events with the majority of the damage potential.

supports the idea that even fraud has an underlying process with interacting parts over a lengthy period. Up to now, a general criterion for the onset of SCO and appearance of power law distributions is still missing and part of ongoing research work. But for the pragmatic discussion in this paper, the phenomenological result of a good fit with the power law can be used as a guideline even without a fundamental explanation of the underlying mechanism. The take away for the first line of defense is that no typical scale in the power law behavior of the OpRisk data exists (or put mathematically: the power law is scale-free). All magnitudes of severity contribute to the total gross loss in equal measure or, vice versa, there is no range of severity on which one can focus primarily. Nothing can be ignored: neither the noise of events at the lower thresholds of OpRisk monitoring (due to high-frequency), nor the mean events in daily business (with mediumseverity/medium-frequency), nor very rare events with frequencies of 1-in-100-years, 1-in-1000-years, etc. (due to highest-severity). This may sound simple but is challenging, as this fundamental feature of OpRisk is very different to the statistical behavior of other types of risk and has to be understood by the whole business line.
Considering these volumes at DZ BANK, quantitative quality management was always an integral part of the management of the transaction banking business line, as any bad quality with low STP ratios (nonstraight through processing with manual corrections) has a significant negative effect on the cost base. DZ BANKs quality management employed in transaction banking can be characterized by the following key elements:
Quantitative measurement of process performance indicators, which are critical to quality. Continuous improvement and learning process. Integration of the specialists on the shopfloor in the quality management process.
This approach is a Six Sigma2 type of quality management. As a toolset which compiles established methods3 Six Sigma is nothing new or groundbreaking.4 From the viewpoint of a practitioner in a business line, there are only minor principle differences between Six Sigma, TQM (Total quality management), Gemba Kaizen (in Japan), etc. [Hopp and Spearman (1995)]. However, Six Sigma can be used as a general lingua franca of quantitative quality management. This common language makes it much easier for subject matter experts and business line management to reach a mutual understanding and develop a mutual mindset of quality management. One part of this lingua franca is the well-known DMAIC cycle:

DZ BANKs business line transaction banking as an example for a the first line of defense
DZ BANK is the fourth largest bank in Germany and is acting as the central bank for approxiately 900 cooperative banks. DZ BANKs role in transaction banking can be simplified by from banks and back to banks: (i) bundling of transaction volume of the local banks for economies of scale, (ii) utilization of joint ventures to achieve a second step of economies of scale, (iii) continuous optimization and innovation, and (iv) delivery of added value back to the local cooperative banks (reduced cost, innovations, product management including all tax, legal, regulatory, and statutory issues, etc.). According to the OpRisk mapping of different business lines [BIS (2006)], DZ BANKs transaction banking consists of payment and settlement and agency services including retail payments processing (high-volume), international and high-value payments, cash handling services, securities transaction services, corporate actions, securities lending services, and depositary banking; plus payment and securities services related to the banks own activities (to be incorporated in the loss data of the respective business line). In this role as the cooperative central bank, DZ BANK bundles a large number and scope of transactions: from some four billion retail payment transaction to four million high-value payments per annum and from some six million securities transaction (i.e., retail brokerage transaction) to rather rare corporate actions once in a while (i.e., non-standard redemption offers). In parallel, there is tremendous spread of the nominal amounts in those transactions, between a few euros (retail payments) and 100+ million euros (i.e., in commercial payments for corporate customers).
Define the scope of the problem and the objectives. Measure the process parameters. Analyze the measured data as a basis for hard-fact improvement measures. Improve the process by implementing the selected solutions. Control the achieved results against the defined target.
As the DMAIC cycle is a well-established tool, it is appropriate to (re)use this approach for OpRisk management in the business line with the following adaptations for the five steps (shown in Figure 2): 1. Identification of possible risk(s) and definition of an operational risk strategy (accept and manage, reduce, mitigate, avoid). 2. Measurement of operational risk indicators with special regard to the end of the distribution, i.e., lowest-frequency/highest-severity events, and awareness of the limitation of those measurements. 3. Analysis of measured operational risk indicators on the level of the business line.
2 3 4
Six Sigma is a registered trademark and service mark of Motorola Inc. See, for example, Deming (1950); Feigenbaum, (1951), McGraw-Hill, Feigenbaum (1983), McGraw-Hill et al., and Juran and Gryna 1957). For the status of Six Sigma in the financial services industry see Heckl (2010)
157
4. Management process for the implementation of improvements, which is the crucial factor for sustainable improvements beyond some ad-hoc bug fixes to achieve a long-term learning curve. 5. Controlling of achieved results, adoption of the defined operational risk criteria, and critical review of the assumptions of the model including the assumption of the power law behavior itself. This DMAIC cycle is embedded in the overall OpRisk framework and central OpRisk controlling and aligned with the strategy of the business line including the defined appetite of business line for OpRisk. Likewise, all tools of OpRisk monitoring and controlling are available to be applied in the business line including the loss database (entry of events, use of external loss databases, analysis), OpRisk self-assessment, OpRisk indicators with corresponding alarm thresholds, and OpRisk manual. However, at business line level and especially at shopfloor level of the process experts OpRisk management gets an additional perspective: the mindset (or talent) of all those experts, their awareness, and their agility to manage OpRisk, where it materializes first. If one takes first line of defense rather literally, two quotations from quite unusual references elaborate the crucial importance of proper training and common mindset of those experts, who are doing the daily business, in an intriguing way: In the heat of battle, plans will go awry, instructions and information will be unclear and misinterpreted, communications will fail, and mistakes and unforeseen events will be commonplace [Fleet Marine Force Manual (1997)]. This can be seen as a modern version of a phrase written in 1832: It is, therefore, natural that in a business such as war, which in its plan built upon general circumstances is so often thwarted by unexpected and singular accidents, more must generally be
left to talent; and less use can be made of a theoretical guide than in any other [ Clausewitz (1832)]. If our experts in the business line do not possess the mindset (or talent) to understand what OpRisk means, sustainable improvements cannot be achieved. In the first line of defense a common understanding, a culture of open communication of risk events, and common learning processes are the crucial success factors [Levy et al. (2010)]. Even when staff members do not comprehend the difference between a power law distribution and a Gaussian distribution, they can be part of a OpRisk culture beyond sophisticated statistics. If they are not afraid to identify and report risk events even those events they are personally involved in then this is the starting point for learning and improvement. In the following sections the five steps of the OpRisk management in the business line are discussed along the DMAIC cycle, but always with a special focus on this mindset for OpRisk management.
Definition and awareness for specifics of OpRisk

According to the Basel Committee, OpRisk is defined as the risk of loss resulting from inadequate or failed internal processes, people and systems or from external events. This definition includes legal risk, but excludes strategic and reputational risk.5 For the business line this general definition and the defined event types have to be mapped on to the specific processes and conditions of the day-to-day banking business. Possible risk(s) have to be identified, for example, along process chains, in product portfolios, and according to the development of transaction volumes. When the risk(s) are identified, an operational risk strategy at the first line of defense can be defined, how to accept and manage, to reduce, to mitigate, or to avoid risk. This strategy has to address a wide range of possible loss events from the high-frequency/low-severity domain to the one-claim-causes-ruin [coined by Neslehova et al. (2006)].
OpRisk framework and central OpRisk controlling
Strategic planning process of business line including risk appetite for OpRisk
Accept and manage taking and managing risk is the fundamental business of a bank and this holds true for OpRisk. Especially in transaction banking with the huge number of transactions, there will be loss
Control Different horizons of expectations Adoption of definitions Review of assumptions Improve Management of improvements: Communication Culture Learning
Define Definition to accept and manage, reduce, mitigate, or avoid OpRisk Set-up of training of management and staff including awareness for extreme events Analyze Statistical analysis (ex-post) Analysis of future risk potential (ex-ante) Measure Measurement of loss events Limits of measurement
events in the high-frequency/low-severity region simply due the volume of business and the fact that a 100 percent correctness would be economically unaffordable. These loss events are the expected ones, and will be analyzed to see whether there is an economically justified opportunity to reduce them. From the perspective of an OpRisk culture in the business line, it is essential that expected losses are accounted for against a budget in the P/L of the business line. When expected have a budget, the question of who did something wrong becomes obsolete
158
Figure 2 DMAIC cycle for the OpRisk management in the business line
In this paper two types of risk are not covered, as they do not show up in the OpRisk data in transaction banking with statistically significance: model risk, which may be the most opaque type of risk [see in particular the recent OCC publication (2011)] and legal risk [see the analysis of Rosengren (2008)].

(except, of course, in the case of internal fraud); but the question what went wrong provides the opportunity for learning and improvement, if no one has to fear to be blamed for reporting loss events, because they are expected. Reduce in the high-frequency domain, technical improvements are the natural way to reduce OpRisk: i.e., through improved technical interfaces with rule-based monitoring, double-checking procedures, etc. Sometimes, rather simple measures like changes in graphical user interfaces, which can optimize the flow of clicks or the movement of the human eyes, can achieve significant enhancement. In the low-frequency domain, which is dominated by non-standard incidents, measures to reduce OpRisk are continuously improved process guidelines. However, the key factor to reduce low-frequency events in the first line of defense is a well-trained staff with good understanding of the whole process chain, people with the right mindset about OpRisk management, and the appropriate OpRisk culture. However, in times of general cost cutting, this well trained staff with the right mindset is a more and more limited resource. At some point, it has to be decided whether the number of well trained staff is still sufficient to reduce OpRisk. Otherwise, a decision to be made about whether (i) an increasing amount of losses will be accepted or (ii) the most risky types of transactions have to be avoided. Mitigate for transaction banking, mitigation is a powerful part of the OpRisk strategy, and it should be possible to achieve that by ensuring that a large fraction of the total loss amount are either covered by insurance or transferred to outsourcing partners. Outsourcing does not take away the responsibility of OpRisk management, But it6 provides clearly defined roles and measurable service level agreements, which per se help to reduce OpRisk due to improved process maturity [Milkau (2011)]. Avoid it is a generic principle of banking to avoid risk, if this risk is regarded as too high. This principle defines which types of transactions have to be avoided.7 Of course, transactions bearing high-frequency/ high-severity risk will be avoided per se. But the power law behavior has the implication that there is no a priori limitation of the total loss amount due to the tail of the distribution and even the rare events with highest severity contribute to the total loss in an equal measure. Consequently, attention has to be paid to the low-frequency/high-severity part. Devoid of any finger pointing, the payment of of 350 million by the German KfW bank8 to Lehman Brothers two hours before the American investment bank collapsed is an example of a (very) high-severity event. In principle,9 such an event can happen to any bank with high-value transactions in the world. As these evens with a probability of 1-in-1000-year are so rare, such external loss events provide valuable insight. A press release by KfW (2008) described the complexity of this event,10 a currency swap between KfW and Lehman Brothers, as: (1) swap conducted
on 10 July 2008 a with standard procedure; (2) conversion done properly on 14 July 2008 and reconversion scheduled for 15 September 2008; (3) situation at Lehman worsened on Friday, 12 September 2008; (4) open settlement risk was overlooked; (5) the situation was then not monitored over the weekend; (6) follow-up meeting was scheduled for Monday morning, 9:30 a.m. The payment, however, had left the bank at 8:37 in the morning. Two major lessons learned can be derived from the KfW event:
Such once-in-a-lifetime events with tremendous losses typically do not fit into a simple loss event type classification (fraud, process disruption, system failures, etc.), but result particularly from concatenation, for example, of an internal chain of errors (overlooked settlement risk, long lasting transactions, process gaps over the weekend) with the external events of insolvency (i.e., counterparty risk). The understanding of the underlying process interactions is a key to avoiding those rare events successfully.
In Figure 3, the KfW-event is compared to a Monte Carlo generated loss event time series for ten years. According to their personal experiences, people typically would bet that any losses would be limited to the known loss amount of those last ten years. But the power law distribution reminds us, to expect the unexpected as well. Any sentry of a first line of defense knows that it is a huge psychological challenge to be constantly at the ready for things that may never happen.
A careful training and repeated education of staff and management is required to keep the vigilance and to hope for the best, but be prepared for the worst. Especially the last issue leads over to the critical issue of understanding the limits (and inherent danger) of measurement in OpRisk management.
Nonetheless, potential hidden risks in an outsourcing relationship with a provider have to be monitored. When a provider takes a new client on board, there could be the risk of competition for resources (technical resources, but also human resources for non-standard processing with manual steps) and an increased risk for outages or delayed processing due to locked databases or similar technical bottlenecks. 7 It is worth stating that the strategy to avoid risk, if too high was already mentioned by Luis de Molina in 1597, a scholar of the school of Salamanca. Although Luis de Molina elaborated on banking risk in general and not specifically OpRisk, the idea of a risk to be avoided is obviously a 400 year old concept. 8 The KfW is a promotional bank under the ownership of the German Government and the federal states. 9 For example, due to the power law distribution with a non-zero probability even for tremendous losses. 10 For the purpose of this discussion, it would be splitting hairs to discuss whether this KfWevent belongs to transaction banking or to any other business line, as according to the Basel mapping of the business lines payment and settlement losses are only related to customers payments and a banks own activities have to be incorporated in the loss of the affected business line.
159
Loss amount magnitude (logarithmic scale)
Earthquakes in Japan
Losses of 18 Japanese banks

Threshold effects Tail with complex events
10
10
-1
3
10
-3
10
-4
10
-5
Data from Pisarenko et al. (2010); 39316 events 1923-2007 Fit F(M) = c
*
Tohoku 2011
Frequency (log)
10
-2
2 1
M6,5 7,0 7,5 8,0 8,5
Mmax, calc.
9,0 9,5 10,0
Original data and gure from Nagafuji et al. (2011) Fit F(S) = c
*
SSeverity (log)
10
-6
4,0
4,5
5,0
5,5
6,0
Magnitude*
Y1
Y2
Y3
Y4
Y5
Y6
Y7
Y8
Y9
Y10 Years
Sep 15, 2008
Figure 4 Comparison of the data from Pisarenko et al. (2010) for the distribution of earthquake magnitudes in Japan including the Tohokuevent of 2011 with the data of Nagafuji et al. (2011) for loss severity distributions among Japanese banks.
The Monte Carlo simulation is based on a power law with a parameterization log10(F/y) = constant log10(S/k) with = 1 and includes non-stationarity (i.e., time-dependence of the density of events). This simulated 10 years time series is compared to the KfW-event of Sep 15, 2008 (right box), which has a 1-in-1000 years probability. The three lines are illustrative indicators: (1) separation between the noise of high-frequency/low-severity events and the rest, (2) threshold to separate the peaks-over-threshold, and (3) a line to guide the eye to compare the event with highest severity within 10 years with the magnitude of the KfW-event. Please note that the ordinate is in logarithmic scale.
discuss the valuable work of Nagafuji et al. in detail, but three outcomes are evident:
The (rather granular) loss data can be fitted quite well with a simple power law distribution. There is a threshold effect at the low end potentially due to reporting bias (depending on different reporting thresholds). There is a long tail with very few lowest-frequency/highest-severity events, for which the end of the distribution cannot be extrapolated simply from the data but has to be estimated with sophisticated statistics.
Figure 3 Synthetically generated random loss event time series for ten years
Measurement and the limits of measurement

The development of qualitative approaches to measure and to model OpRisk was strongly pushed by regulatory guidelines known as Basel II with a focus to determine total losses and to calculate regulatory capital for OpRisk. There is growing scientific research, but also some controversial academic discussion, on the various aspects of modeling OpRisk, described in textbooks such as Gregoriu (2009) or articles such as Embrechts et al. (2011). But what can the business line do more than merely collect OpRisk data? Can risk management approaches in other fields be a source of additional insight for a business line like transaction banking? An answer to these questions can be found in the comparison of two recent research papers, entitled: Loss severity distributions observed among 18 Japanese banks, by Nagafuji et al. (2011) and Distribution of maximum earthquake magnitudes in future time intervals, of Pisarenko et al. (2010). Nagafuji et al. analyzed an aggregated dataset reported by 18 Japanese banks as a part of the LDCE2008 study with a total of 324623 submitted losses, 2502 losses greater than 20000, and a total loss amount of about 950 million. The results taken from the analysis of the aggregated data without providing absolute values in order to protect the anonymity of the data is shown in Figure 4 (right) in a double160 logarithmic representation. It would be beyond the scope of this paper to
More insight can be obtained by comparing this loss data distribution with a fundamentally different phenomenon: earthquakes. Both share the same feature of power law behavior, as presented by Pisarenko et al. (see Figure 4/left). Pisarenko et al. applied an analysis to the earthquake catalogue of Japan for the years 1923 to 2007. Once again, one can see a quite good fit with a power law, threshold effects at the lower end (potentially due to a detection bias in the early years) and a long tail with few, but heavy earthquakes until about magnitude 8 (but not higher). From the comparison the following two observations can be derived.
Magnitude M 5 6 7 8 9
Energy (in units of E0) E0 32 E0 1,000 E0 32,000 E0 1,000,000 E0
Frequency (illustrative) 1000 100 10 1 1/10
Total Energy (illustrative) 1,000 E0 3,200 E0 10,000 E0 32,000 E0 100,000 E0
Table 1 Magnitudes and total energy of earthquakes

First, mathematically spoken, the power law distribution has an infinite mean. As shown in Figure 1, the total loss in a magnitude bin does not decrease with increasing severity, although the frequency decreases. The same holds true for earthquakes as for every increase in magnitude M by one unit, the associated seismic energy increases by about 32 times. This leads to the schematic result of Table 1: the more (rare) higher magnitudes M are included, the higher is the total energy release.11 One of the criticisms often leveled at the use of power law distribution models to OpRisk loss data is this infinity of the mean total loss, if summed up to highest (low-frequency) severities [Neslehova et al. (2006)]. But the case of earthquakes with the same type of power law distribution tells us that nature simply does not care about an easy-to-handle mathematical treatment.
depending on their risk weightings and measures are taken before losses occur.
Analysis and what can be done ex-ante

As previously mentioned, collecting OpRisk event in a loss database and running analyses on those data, including the aforementioned EVT approach [Embrechts et al. ((2004); Chavez-Demoulin et al. ((2006)], is a usual approach to OpRisk with ex-post analysis. Once again, the question has to be asked, whether the business line can do more? Can transactions be classified ex-ante? Is there a chance to put a risk tag on each transaction, or at least on the non-standard ones? An alternative to the statistical approach in OpRisk modeling was already
Second, Pisarenko et al were able to calculate statistical estimations for a maximum Magnitude Mmax around 9.5 based on two methods of Extreme Value Theory (EVT) [Embrechts (1997)]. A magnitude Mmax of around 9.5 is more than one magnitude away from the historical data. Could this be realistic? The answer to this question was given by nature and is a very unfortunate and unhappy answer. The devastating earthquake which occurred offshore at the north-eastern Tohoku region in Japan on March 11, 2011, caused tremendous damage, with over 15,000 deaths and 10 trillion yen in damages directly attributed to the event. The earthquake was scaled as magnitude 9.0, which is the largest one ever recorded in Japan in its history. It was also the worlds first M9 earthquake in modern times. Such a tremendous earthquake was not expected, although the research work of Pisarenko et al. was published before this sorrowful disaster. To quote Hideo Aochi and Satoshi Ide (2011): Speaking seismologically, due to the frequent occurrence of M~7.5 events along the Japan Trench in northeastern Japan during the last century, it might have been generally accepted that such event series characterize regional fault behavior, without considering the possibility of M9 events. What is the lesson learned for OpRisk measurement in the business line? Obviously, careful monitoring and measurement is the starting point for any OpRisk management process. Loss events are collected, commented on, and documented in the internal database, expert assessments are taken into account and relevant external OpRisk data are included in the OpRisk management process. But the Tohoku earthquake teaches us to expect the unexpected and understand the limitations of measurements. It is very dangerous for the business line to concentrate too much on the measured loss events. The business line has to keep an open eye on the edges and beyond, has to expect the unexpected, and should stay away from too fast, too simple explanations of measured irregularities. However, there is one big difference between measuring earthquakes and OpRisk events, as the business line has the opportunity to reduce OpRisk actively, if, for example, transactions can be classified ex-ante
recommended by Carol Alexander (2000): to apply Bayesian Networks. A Bayesian Network is a representation of conditional probability [BenGal (2007)], in which various defined risk drivers are combined along the nodes of a network to an end-node, i.e., from asset classes, trading venues, notional value, settlement delays, etc., to the estimated number of fails and resulting losses of transactions with those initial parameters.12 To quote Alexander: a Bayesian Network describes the factors that are thought to influence operational risk, thus providing explicit incentives for behavioral modifications. Recent publications such as Neil et al. (2005, 2008), Aquaroa et al. (2010), and Sanford et al. (2011) elaborated on Bayesian Networks as tools for scenario simulations, visualization of OpRisk correlations, and by making things visible triggering those behavioral modifications. A simplified but pragmatic approach for the business line is an ex-ante classification of transactions based on empirical correlation between risk-drivers and measured loss events (equivalent to a network with only an input node layer and one result node). Risk-drivers can be, for instance, non-straight-through-processing (with manual intervention), ad-hoc changes in Standard Standing Instructions (SSI), non-standard process flows (i.e., settlement processes with chain of counterparty), situations after a system migration (due to a potential bias on technical problems), long-lasting and interrupted process chains, etc. Each risk-driver contributes a certain level of risk potential to a transaction with an amount which has to be derived from an analysis of historical loss data. The sum of all risk potentials gives an ex-ante indication of how risky a transaction could be. As the purpose of this approach is to raise awareness and to focus that awareness on those potentially risky
11 The maximal possible energy release is limited only by boundary of the systems itself, i.e., by the total tectonic energy in Earths surface. 12 A similar alternative was proposed by Grody (2008) with risk accounting for each transaction with an appropriate risk weighting as a measure for its potential for losses. This approach is comparable to a Bayesian Networks with (a tremendous number of) transactions as nodes at the top and accumulated risk weightings down the processing path to the end-node(s) of risk per product.
161
transactions, before real OpRisk events happen, it does not need to be too mathematical, but should be regarded as a risk tag on those transactions which require increased monitoring. Similar to many of the other ideas outlined in this paper, this approach can be found in other industries, such as in manufacturing as a well-known way to monitor a product through the production process, i.e., with bar codes or NFC tags. Such tools make it possible to control the production process by the individual product itself and to improve the quality, as each single production station can access all required information directly from the product at this station. Furthermore, it is a lesson learned that deviations from defined standard processes are the first step to increased risk potential. It is a widespread case that people tend to deviate from the standard process to help colleagues and to correct errors made in the previous process steps. Of course, this is well-intentioned, but generates uncertainty, misalignment, and uncontrollable variations. It requires a change in mindset not to help but to follow the defined process (i.e., to reject a transaction in the back office and send it back to either the front office for correct re-entry or to the middle office for enrichment). In industrialized processes such as transaction banking even the experts cannot overlook all consequences of a well-intentioned intervention in the following process chain.
Some rather illustrative examples for too fast, too biased, or too unstructured handling could be:
I spotted an OpRisk problem outside my personal responsibility, but I assume that somebody else is in charge and already dealing with this problem.
The unusual pattern which we detected in one type of transaction has to originate from the system migration we had last month. As there is a flu epidemic with many experts in my team off sick this week, let those events be forwarded to the technical migration project which will take care of it.
For a discussion about our operational risk, we take our in-sample loss data of the last 10 years and make an interpolation, as 10 years provide pretty good statistics also for predicting out-of-sample event and this approach guarantees that we will stay in the accepted limits.
Effective management of improvements starts with the mindset to pull the andon cord without hesitation in case of potential risk and ends with a visible and easy to understand documentation of the measures to establish a continuous learning process. To ensure sustainability of the improvement, all management levels in the business line have to be integrated in a defined procedure with a virtual andon cord:
Management of improvements from the andon cord to top management and back to the shop floor
For the first line of defense, OpRisk management is closely connected to communication (risk events and failures without bias and preconception), culture (open discussion about how to get better, but not of searching for the scapegoat), and learning (re-use and transfer of ideas in the business line and across business lines). Best practice from a different industry can illustrate how communication, culture, and learning can be integrated in a structured process. In the Toyota Production System [Ohno (1988)], each employee is a part of the quality culture and each team member is a quality inspector. Any team member who detects a problem can stop production by pulling the andon cord located next to the assembly line. A so-called andon board visualizes the location of the problem for the shop floor manager in charge. Either the problem can be directly solved at the production station, or the supervisor can stop the whole line in case of a larger problem. Depending on the time needed to solve the problem, an immediate communication and escalations process is started. Afterwards, the problem and the solution are documented at the shop floor quality panel as open visualization for staff and management. The concept of the andon cord as a technical synonym for a struc162 tured communication process can also help to manage OpRisk events.
General awareness that everyone is part of the overall risk culture, from OpRisk management as an executive function for the whole of the top management of the business line to the right OpRisk mindset of each staff member.
Ad-hoc reporting with defined escalation processes for events depending on severity or time without solution. Feedback about the solution or improvement back to the shop floor. Shop floor quality panel as a means of visualization for measured risk indicators and documented solutions. Business line cockpit with key performance indicators including OpRisk indicators. OpRisk management as a regular focus topic in middle management meetings. OpRisk reviews in management meetings of the business line. Dedicated annual OpRisk meeting of the top management team of the business line.
Especially the feedback to the shop floor and the visualization on the shop floor are quite new compared to the one directional reporting of loss events. This feedback requires a backward flow of information to enable a learning process at the first line of defense. It may sound simple but, nevertheless, is a challenge in the development from OpRisk controlling to active OpRisk management.

Controlling and the horizons of expectation

Unfortunately, standard statistical tracking of short-term improvements in the loss distributions is not feasible.13 The synthetically generated loss event time series shown in Figure 3 illustrates that (i) the variation from year to year by a factor of two or even more could mask smaller improvements, (ii) non-stationarity can hide improvements especially in the highfrequency/low-severity part, and (iii) at the low-frequency/high-severity end of the distribution the probabilities of 1-in-10 years and below are far away from any chance to see changes on a year-to-year basis. How should the business line validate the success of any improvement measure, if there is no statistical significant signal in the course of a year? In Figure 3, three different horizons of expectation can be distinguished. The two separation lines are indicative, as there is no fixed and a priori separation between the three different horizons of expectations, and they have to be estimated with some iterative approach for the best choice. Nonetheless, a segment with the frequent events with low severity (noise), a middle segment, and a top segment with only few events per year can be discriminated. In the top segment, all peaks over threshold have to be monitored case-by-case and checked for recurrences of events, detected (new) gaps in process chains, insufficient process design, events not included in OpRisk self-assessments yet, etc. Depending on the outcome of that monitoring measures can be realigned or new measures can be implemented. The middle segments can be monitored for the causes of loss event clusters in different processes such as, for example, processing of corporate actions (i.e., root-cause analysis). If either the same cause (i.e., incorrect or delayed settlement instruction) materializes to be significant or, respectively, a removed cause shows up once again, business line management has to take action. Causes could also be a significant high sickness absence rate among staff and other non-technical OpRisk indicators. In the segments of noise, the (exante) classification of transactions has to be monitored, and derivations can be indication of positive or negative changes in the risk profile of the business line if they are statistically significant;14 as the mentioned nonstationarity is hard to separate from real improvements (or deterioration). Last but not least, the controlling step has to include scrutinizing the power law distribution of loss events itself. The power law is only a model, and this model is based on some assumptions. These assumptions may look rather justified as the power law is a generic feature of many destructive processes but the assumptions can still be wrong. The business line has to take this model risk into account [Derman (1996); Sibbertsen et al. (2008)]. As our models and our assumptions can be wrong, they themselves could be a cause for OpRisk as well. If new models are available, they have to be evaluated, and if they are better, they have to be implemented.
Conclusion
The business line as a first line of defense in OpRisk management has its own role, independent from central OpRisk controlling and from the review and audit function. All three lines of defense in a bank, however, have to be fully integrated into the banks overall risk management governance structure. The first line of defense is responsible for managing the risks inherent in the products, activities, processes, and systems for which it is accountable, i.e., in transaction banking. This includes collecting loss events and reporting to the central OpRisk controlling, which has to aggregate those data, to challenge the data, and to analyze the data (controlling and quantitative modeling). For the active management of OpRisk in the business line, the DMAIC cycle provides a robust and pragmatic framework (with the right measure of statistics). The DMAIC cycle balances analytical tools and practical concepts for staff and management. All this has to happen along the power law distribution. The power law shows a generic difference to other statistical distributions typically used in risk management frameworks. Its characteristics govern OpRisk from the high-frequency/low-severity events to the lowest-frequency/highest-risk events and teach us to expect the unexpected. The approach described in this paper combines the DMAIC cycle with the specific characteristics of the power law, tries to circumvent too much of advanced statistics, and focuses on a framework which enables staff and management to understand and manage OpRisk in the business line. Consequently, the right risk culture, open communication of OpRisk events, and a continuous learning process are key success factors of the first line of defense. Phenomena described by the power laws exhibit so-called scale-free behavior, as there is no typical size for key parameters such a typical length, a typical magnitude, or a typical loss amount. While scientific research is developing, scale-free behavior is also found in complex systems, where it was not expected: for example, one latest discovery was in starling flocks [Cavagna et al. (2010)]. Further research is needed to achieve better understanding of the fundamentals of such complex systems; and one should also take such new insight from other fields of research into account in the management of OpRisk in the business line.
13 This is rather different to statistical quality management with concepts such as Six Sigma, which provide enough data to control the success of improvement measures, such as in non-STP payment transactions as demonstrated in Milkau (2011). 14 Statistical significance is not easy to achieve. For a meaningful filter, false positives and false negatives results have to have quite low probability to be statistically significant. A good example of the difficulties in providing usable filters is the discussion about the application of Benfords law in forensic accounting and auditing as an indicator of manipulation and fraud [Diekmann and Jann (2010)], as often proofs for the usefulness of Benfords law are given ex-post with single examples, which are known as fraudulent, but not ex-ante in blind tests with control groups.
163
References

Alexander, C., 2000, Bayesian methods for measuring operational risk, in Alexander, C. (ed.), Mastering risk, volume II, Pearson/Financial Times Press Alexander, C., 2003; Managing operational risks with Bayesian networks, in Alexander, C. (ed.), Operational risk: regulation, analysis and management, Pearson/Financial Times Press Aochi, H. and S. Ide, 2011, Conceptual multi-scale dynamic rupture model for the 2011 Tohoku earthquake, in Earth Planets Space special issue: First results of the 2011 off the Pacific Coast of Tohoku Earthquake Aquaroa, V., M. Bardoscia, R. Bellotti, A. Consiglio, F. De Carlo, and G. Ferri, 2010, A Bayesian Networks approach to operational risk, Physica, A 389, 1721-1728 Bak, P. and M. Paczuski, 1995, Complexity, contingency and criticality, Proceedings of National Academy of Science of the United States of America, 92:15, 66896696 Bak, P. and K. Sneppen, 1993, Punctuated equilibrium and criticality in a simple model of evolution, Physical Review Letters, 71:24, 40834086 Bak, P., C. Tang, and K. Wiesenfeld, 1988, Self-organized criticality, Physical Review, A38:1, 364374 Basel Committee on Banking Supervision, 2003, Sound practices for the management and supervision of operational risk, BIS, February Basel Committee on Banking Supervision, 2006 International convergence of capital measurement and capital standards a revised framework comprehensive version, BIS, June Basel Committee on Banking Supervision, 2009, Results from the 2008 loss data collection exercise for operational risk, BIS (LDCE2008) Basel Committee on Banking Supervision, 2011, Principles for the sound management of operational risk, BIS, June Ben-Gal, I., 2007, Bayesian Networks, in: Ruggeri F., Faltin F. & Kenett R. (eds.), Encyclopedia of statistics in quality and reliability, Wiley & Sons, Hoboken, New Jersey Cavagna, A., A. Cimarelli, I. Giardina, G. Parisi, R. Santagati, F. Stefanini, and M. Viale, 2010, Scale-free correlations in starling flocks, Proceedings of the National Academy of Science of the United States of America, 107:26, 11865-11870 Chavez-Demoulin, V., P. Embrechts, and J. Neslehova, 2006, Quantitative models for operational risk: extremes, dependence and aggregation, Journal of Banking and Finance, 30:10, 2635-2658 Clauset, A., C. R. Shalizi, and M. E. J. Newman, 2009 Power-law distributions in empirical data, SIAM Review, 51, 661-703 Clausewitz, C., 1832, On War, translation by Graham, J. J., published by N. Trbner, London, 1873, Book 2, Chap. 2, 23. (www.clausewitz.com/readings/OnWar1873/BK2ch02.html; download Dec. 25, 2011), in German: Vom Kriege (http://gutenberg.spiegel.de/buch/4072/12): Es ist also natrlich, da ein Handeln wie das kriegerische, was so hufig in seinem auf allgemeine Umstnde gebauten Plan durch unerwartete individuelle Erscheinungen gestrt ist, berhaupt mehr dem Talent berlassen bleiben mu und von einer theoretischen Anweisung weniger Gebrauch machen kann wie jedes andere. Cruz, M. G., 2002, Modeling, measuring and hedging operational risk, Wiley & Sons, Hoboken, New Jersey Deming, W. E., 1950, Some theory of sampling, Wiley, New York Derman, E., 1996, Model risk, Risk, 9, 3437 Diekmann, A. and B. Jann, 2010, Benfords Law and fraud detection: facts and legends, German Economic Review, 11:3, 397-401 Embrechts, P. and M. Hofert, 2011, Practices and issues in operational risk modeling under BaselII, Lithuanian Mathematical Journal, 50:2, 180-193 Embrechts, P., R. Kaufmann, and G. Samorodnitsky, 2004, Ruin theory revisited: stochastic models for operational risk, in Bernadell, C. P. Cardon, J. Coche, F. X. Diebold, and S. Manganelli (eds.), Risk management for Central Bank foreign reserves, European Central Bank, Frankfurt, 243-261 Embrechts, P., K. Klppelberg, and T. Miklosh, 1997, Modelling extremal events for insurance and finance, Springer, Berlin Feigenbaum, A. V., 1951, Quality control: principles, practice and administration, McGraw-Hill, New York Fleet Marine Force Manual, 1997, Warfighting, U.S. Marine Corps Gregoriu, G. N., 2009, Operational risk toward Basel III best practices and issues in modeling, management, and regulation, John Wiley & Sons, Hoboken, New Jersey Grody, A. D., P. J. Hughes, and S. Toms, 2010, Risk accounting a next generation risk management system for financial institutions, Journal of Financial Transformation, 29, 43-56 Heckl, D., J. Moormann, and M. Rosemann, 2010, Uptake and success factors of Six Sigma in the financial services industry, Business Process Management Journal, 16:3, 436-472 Hopp, W. J. and M. L. Spearman, 1995, Factory Physics: foundations of manufacturing management, Irwin Professional Publishing, Chicago, 3rd edition, McGraw-Hill/Irwin, New York King, J. L., 2001, Operational risk: measurement and modelling, Wiley & Sons, Hoboken, New Jersey
Juran, J. M. and F. M. Gryna, 1957, Quality control handbook, McGraw-Hill, New York KfW, 2008, Press Release, Comments by KfW Bankengruppe on article KfW transfer was apparently intentional in Frankfurter Allgemeine Zeitung of 22 September 2008 Levy, C., E. Lamarre, and J. Twining, 2010, Taking control of organizational risk culture, McKinsey Working Papers on Risk, No. 16, June, McKinsey & Company Milkau, U., 1991, Emission mittelschwerer Fragmente in asymmetrischen Schwerionenstssen bei mittleren und relativistischen Einschuenergien, GSI Report GSI-91-34 (ISIN 0171-4556) Milkau, U., 2011, A Six Sigma type of quality management in international and high-value payments, Journal of Payment Strategy and Systems, 5.4 De Molina, L., 1597, Tratado sobre los cambios, Cuenca; reprint: Instituto de Estudios Fiscales, Madrid, ICI, 1990; translation: Treatise on money Nagafuji, T., T. Nakata, and Y. Kanzak, 2011, A simple formula for operational risk capital: a proposal based on the similarity of loss severity distributions observed among 18 Japanese banks, FSA Staff Perspectives, Financial Research Center/FSA Institute, Japan, May Neil, M., N. Fenton, and M. Tailor, 2005, Using Bayesian Networks to model expected and unexpected operational losses, Risk Analysis, 25:4, 963-972 Neil, M., D. Marquez, and N. Fenton, 2008, Using Bayesian networks to model the operational risk to information technology infrastructure in financial institutions, Journal of Financial Transformation, 22, 131-138 Neslehova, J., P. Embrechts, and V. Chavez-Demoulin, 2006, Infinite mean models and the LDA for operational risk, Journal of Operational Risk, 1:1, 3-25 Office of the Comptroller of the Currency (OCC), 2011, Supervisory guidance on model risk management, Bulletin OCC 2011-12, April 4 Ohno, T., 1988, Toyota production system: beyond large-scale production, Productivity Press, Portland, Oregon Pisarenko, V. F., D. Sornette, and M. V. Rodkin, 2010, Distribution of maximum earthquake magnitudes in future time intervals, application to the seismicity of Japan (1923-2007), Earth Planets Space, 62, 567-578 Rosengren, E. S., 2008, Risk-management lessons from recent financial turmoil, Conference on New Challenges for Operational Risk Measurement and Management, Federal Reserve Bank of Boston, May 14 Sanford, A. D. and I. A. Moosa, 2011, A Bayesian network structure for operational risk modeling in structured finance operations, Journal of the Operational Research Society, advance online publication, 11 May Sibbertsen, P., G. Stahl, and C. Luedtke, 2008, Measuring model risk, Journal of Risk Model Validation, 2:4
164
PART 2
Optimal Bank Planning Under Basel III Regulations

Sebastian Pokutta Friedrich-Alexander University of Erlangen-Nrenberg Christian Schmaltz Aarhus School of Business and the True North Institute
Abstract
We provide a modeling framework for banks business planning under Basel III. For this purpose, we write banks planning as a formal optimization problem where Basel III minimum requirements/ratios enter as constraints. The linear program provides dual variables that are interpreted as compliance cost for each Basel III ratio. We analyze the effects of Basel III on banks product mix for a simplified, deterministic two-product case. In what follows, we generalize the model by incorporating parameter uncertainty, adjustment cost, multiple time steps, and products. 165
As any other company, banks have to determine their optimal business strategy. Regulation plays a crucial role in determining feasible business strategies for a given sector. More precisely, the more regulated a sector is the more restricted are the set of feasible strategies. Banks were already highly regulated before the financial crisis and the number of constraints entering the business planning process was high. As a consequence of the financial crisis, bank regulation has been further extended. These new regulatory requirements, Basel III, enforce an even stricter regulation. Under its predecessor Basel II, banks had to fulfill a single (minimum capital) ratio. The new regulations set forth by Basel III require banks to fulfill four ratios. This complicates banks strategic planning. The fact that each product usually affects several ratios at the same time adds further complexity. In order to manage the increased complexity, we propose a formal Basel III optimization model. It seeks to assist bankers in their planning, but it can also help regulators to better predict how banks will adjust their business strategy to the new Basel III requirements. We assume that banks seek to maximize the expected margin income. As products expose banks to risks (loans expose the bank to default risk, deposits expose banks to liquidity risks), any business strategy comes with inherent risks. The risk profile of the optimal business strategy must be within defined risk limits reflecting risk aversion. This risk aversion in turn might be set internally by management or might be given as an external constraint from the regulator. Similar to the classical portfolio optimization, banks maximize their expected returns while limiting inherent risks via constraints. The recent financial crisis and resulting bailouts with public funds revealed that the chosen business strategies had not been in line with such risk limits. The regulatory response was the publication of a new regulatory reform package (Basel III). In its essence, Basel III narrows down the set of potential business strategies by tightening (regulatory) acceptable risks. Acceptance is defined as minimum buffer-to-risk ratios. In particular, the four Basel III ratios are (1) capital ratio (CR), (2) leverage ratio (LR), (3) liquidity coverage ratio (LCR), and (4) net stable funding ratio (NSFR). The Basel III framework affects banks corporate planning via two rationales. On the one hand, products affect ratios; hence, product changes imply changes in the ratios and might risk the non-compliance with a given ratio. Consequently, the integrated treatment of products and risk profiles in a comprehensive optimization model is necessary. On the other hand, products usually do not only affect a single ratio, but they affect several ratios simultaneously. For example, retail deposits enter the LCR for short-term liquidity risk as a source of risk whereas in the NSFR for illiquid funding risk they appear as a risk buffer. Thus, while a product negatively impacts one ratio it can at the same time positively affect another one. In order to study, model, and optimize business planning 166 under these constraints, an integrated model is necessary.
Model description
We start our analyses with a simple bank having only two products. These two products are loans and deposits. Apart from deposits, the bank holds capital as buffer against losses from loan defaults. Apart from loans, the bank also holds a liquidity reserve to protect against deposit withdrawals. We use this simplified setup to analytically show fundamental effects. We gradually extend the basic model for product uncertainty (stochastic scenarios), sticky costs, multiple time steps, and multiple products. We study the interdependencies of the ratios and their impact on the optimal product mix.
Related work
Our paper is naturally related to previous work on bank optimization and recent work on Basel III. Bank optimization is a complex task. Consequently, the literature on bank optimization has split the topic into selected aspects: (1) the determination of the optimal liquidity reserve and (2) the determination of the optimal capital structure. A survey on the first aspect is provided in Santomero (1984). However, the surveyed models are usually one-period models and model banks with a very restricted product mix. Furthermore, they do not consider the capital side. Hence, they cannot assist real-world bank planners. A stochastic programming approach for the optimal management of the corporate liquidity reserve has been proposed by Schmid (2000). The literature on capital structure takes an external investor perspective and addresses the question of the optimal ratio debt versus equity. Examples for this literature are Black and Cox (1976), Leland (1994), and Leland and Toft (1996). However, the literature seeks to derive internal liquidity and capital buffer measures. By contrast, our model bank uses exogenous regulatory limits prescribed in the reform package Basel III. As Basel III is a recent regulatory package, the available literature so far is limited to the regulatory text Basel Committee on Banking Supervision (2011). It introduces and details the capital, leverage, liquidity coverage, and net stable funding ratios. Perotti and Suarez (2011) do examine the joint impact of the LCR and NSFR on risk taking and social wealth, but they do not explicitly model the capital dimensions in terms of the leverage and minimum capital ratio. The closest work to our multi-product model is De Nicol et al. (2011). They model a shareholder value-maximizing bank with non-bank loans and non-bank deposits as well as interbank lending and borrowing. Loans and deposits are on demand (non-maturing). Interbank funding is a net-position, i.e., a bank cannot simultaneously have both interbank lending and borrowing. In a similar fashion, we consider loans, deposits, and interbank transactions. However, our loans are long term (two periods). Furthermore, we have a third funding source: bond issuances. Like loans, our bond issuances are long term. We introduce long-term products to cover the effect of changing ratio parameters: in the NSFR the

Optimal Bank Planning Under Basel Iii Regulations
parameters for products with remaining maturities beyond a year are different to those below a year. Products on demand have constant parameters. In order to capture the dynamic effect of greater and less than one year, we introduce long-term loans and bond issuances. The credit shock modeled in De Nicol et al. (2011) only leads to a reduction in loan income, but not to a reduction of outstanding loan volume. In our setup, the credit shock affects the loan volume, but not the loan income. We believe that defaults are modeled more realistically by reducing the loan volume than reducing the loan income. Finally, De Nicol et al. (2011) do not use the regulatory parameters, but claim that the actual incoming cash flows have to be covered by the actual outgoing cash flows. However, the Basel III-LCR does not allow the use of internal models, but prescribes the use of the same regulatory parameters for all banks. We, therefore, believe that our model is closer to the Basel III regulation.
(A) Balance sheet
(B) Basel III: feasible solution set

k(PD) Possible initial constellation 100% lev
Reserve (R) Deposits (D) Assets
LCR
F NS R
CR
LR
Loans (L)
Basel III
Capital (C) asf/ rsf Liabilities 100%
Figure 1 Overview of the two-product model
Contribution
We provide an optimization model that bank planners can use for their corporate planning under Basel III. It shows the interaction between the four obligatory ratios and how they impact banks business strategy. In that respect, our model can also assist regulators to understand how banks might change their business mix in response to Basel III. Moreover, when considering solely regulatory aspects our optimization model and the characterization of its optimal solutions imply that there are only a few optimal configurations. We start our analyses with the two product case that can be analyzed in an analytical fashion using linear programming theory. This is followed by different variations of our model. Finally, we present a multi-period and multi-product model. Our work seeks to model the planning process as an optimization problem covering the range from a simple two-product to a realistic multi-product setup. As the information disclosed by banks is not granular enough to actually calculate Basel III ratios, our contribution is the formalization of an optimization which is often based on heuristics due to its complexity and which has gained even more complexity with Basel III.
withdrawals it holds a liquidity reserve (R). The resulting balance sheet is displayed in part (A) of Figure 1 Overview of the two-product model. Banks operating under Basel III regulation have to fulfill four key ratios: a minimum capital, a leverage, a liquidity coverage, and a net stable funding ratio (we use abbreviations CR, LR, LCR, and NSFR respectively) [BCBS (2011)]. For the considered bank model with two products, the ratios can be written as follows:
Minimum capital ratio (CR)

The minimum capital ratio can be expressed as: C k(PD) L (1)
This constraint requires a minimum capital that equals the unexpected loss k(PD) inherent in a loan exposure with a given PD. The expected loss is usually covered by the loan rate and used to build up provisions. Losses beyond the expected loss are to be buffered by capital. The function k is the 99.9 percent-quantile of the Basel II Vasicek model. It is volume-independent (i.e., it does not reflect risk concentrations).
The two-product model

In this section, we consider the simple case where we only have two products. In this case, the optimal solutions can be explicitly characterized, which provide structural insights into the nature of the regulations put forth by Basel III. We start with a single scenario and then we include more stochastic information. This condition requires the bank to hold a minimum amount of capital for a given asset size, independent on the riskiness of the assets with lev [0, 1]. This volume-based measure contrasts with the risk-based measure CR. It serves as backstop for situations where the risk-based measure is not reliable (i.e., in extreme market circumstances).
Leverage ratio (LR)

The limit on leverage can be expressed as: C/(L+R) lev (2)
Single scenario static case

As a starting point for our analyses we reduce the bank to its core activity transforming short-term deposits into long-term loans. The corresponding balance sheet positions are deposits (denoted by D) and loans (denoted by L). Loans are exposed to default at a given probability of default (PD) and deposits are subject to early withdrawals. In order to buffer for default losses, the bank holds capital (C), and to buffer against early
Liquidity coverage ratio (LCR)

The liquidity coverage ratio can be expressed as: R c D (3) 167
This ratio ensures that banks hold a minimum liquidity reserve that covers
a potential withdrawal of a fraction c [0, 1] of deposits. The scalar c is volume-independent.
D min {1 k(PD) L, 1-lev, [(1-L)/c], [(1- rs L)/(1- as)]}
(6)
This implies that there exists a maximum deposit (or equivalent a mini-
Net stable funding ratio (NSFR)

The net stable funding ratio can be written as: as D + C > rs L (4)
mum capital). The candidates for optimal solutions (extremal points) are displayed in
This requirement ensures a minimum of stable (long-term available) funding to cover a fraction rs [0,1] of the loans. For example, capital provides 100 percent stable funding whereas only a fraction of as [0, 1] is assumed to be available in the long-term in the case of deposits. The NSFR imposes the golden rule of banking that illiquid assets should be adequately, i.e., long-term, funded. The parameter as measures the available stable funding, i.e., the fraction of deposits that are available long term. The parameter rs measures the required stable funding, i.e., the fraction of loans that are expected to be long-term outstanding. As the setup does not have size-dependent parameters, we normalize total assets to one. This implies the following relations: R = 1- L; C = 1- D, and allows us to remove two of the four variables L,R, D, C. For example we could consider the decision variables L, D only and in this case we can plot the solution set in the two-dimensional space as suggested by part (b) of Figure 1, with assets being the loans and liabilities being the deposits. Any bank setup is characterized as a point in the D, L-space. The figure clearly shows how the Basel III ratios constrain the feasible region. A bank maximizes the expected margin income. Loans and deposits are products. They result from banks role as intermediaries and they charge a margin for this role. Consequently, products have a positive margin (denoted by mL and mD, respectively). In contrast to that, the liquidity reserve and capital are not products, but are composed of (capital market) instruments. Instruments are usually sold and bought at fair market prices. Consequently, we set their margins to zero. As loans can default, their margin contribution is stochastic. By consequence, their margin income enters as expected margin income into the objective function. Expressed in a formal way, the bank solves the following optimization program:
max D,L
Figure 2. The upper left corner corresponds to an entity that is 100 percent equity-funded and that invests all funds in liquid securities. Clearly, this is not a bank but an investment fund. Analogously, the lower left corner also corresponds to an investment fund due to its 100 percent equityfunding. The intersection of LCR and NSFR constraints corresponds to the business model of a commercial bank with a substantial portion of loan and deposit businesses. The intersections of CR and NSFR constraints, LR and CR as well as LR with the x-axis, correspond to the profile of investment banks as the proportion of illiquid loans is low in relation to liquid securities (reserve). We will now consider an example showing how the different configurations might depend on the rates. Example 1 We consider the following parameters: lev = 0.03; k(PD) = 0.05; c = 0.10; rsf = 0.85; asf = 0.80.
100%
1 2
LCR
NS FR
LR
CR
Assets
Basel III
E[ML] L + mD D1
Liabilities
100%
(5)
Assets Liabilities D 0% 50% 80% 85% 97% 0% C 100% 50% 20% 15% 3% 100% L
1
The optimization model (5) can be solved by linear programming as both the objective function and constraints are linear. It is known from linear programming that the optimal solution is always situated at an extremal point (or vertex) of the feasible set. As we have only two decision variables L, D here such a vertex is given by the intersection of two tight constraints. The optimal vertex (which corresponds to optimal levels L, D) solely depends on the profitability (margin) ratio between loans and deposits. The Basel III constraints (1), (2), (3), and (4), can be summarized to the following single constraint: 168
2 3 4 5 6
R 0% 10% 60% 70% 100% 100%
Business model Mortgage fund Commercial bank Investment bank Investment bank Investment bank Investment fund
100% 90% 40% 30% 0% 0%
Figure 2 Potential business models as solutions

The feasible region for these parameters is depicted in Figure 3. In this case, the net stable funding constraint is outside of the feasible region, i.e., it does not constrain the choice of D and L at all. We obtain the following optimal vertices (depending on the chosen rates): L = 0.000 L = 1.000 L = 0.905 L = 0.600 L = 0.000 D = 0.000
Assets
Feasible region, numeric example

100%
LCR
NSF
LR R
k(PD) c lev asf rsf
5% 10% 3% 80% 85%
D = 0.000 D = 0.954 D = 0.970 D = 0.970
Basel III
CR
Economic information from linear programming

The capital ratio, the liquidity coverage ratio, and the net stable funding ratio compare the outcome of a risk model to a buffer. The risk models are reflected in the scalars. As the scalars are the same for all banks, the feasible set is the same for all banks if they all had loans and deposits as their dominating business activity. If furthermore, loans and deposits are of similar profitability across banks and banks maximize their expected margin income, the optimization model predicts that one should observe banks converging to the same (very similar) loan and deposit fractions. In fact we can obtain much more information from Model 2. An optimal solution to this linear program provides several additional economic indicators that are worthwhile to consider. For this let us consider the model with the proposed substitutions:
max D,L E[ML] L + mD D
Liabilities 100%
Figure 3 Feasible region for example D is on the x-axis and L is on the y-axis
Moreover it follows from linear programming theory that at any optimal point is defined by exactly two tight constraints in the non-degenerate case which we can assume here by -perturbations. Consequently, a bank with two products considered here in the simplified model will only encounter two of the ratios to be binding at a time. This simple but crucial insight is, once again, highlighted in Figure 2 and Example 1. In fact, one can interpret the various choices of exactly two binding constraints as determining the (regulatory) type of the institution.
(7) The dual linear program associated with Model (7) is given by:
s.t. D + k(PD) L 1 D 1 lev D + 1/c L 1/c D [(1 rs)/(1 as)] L 0 D, L 0 (CR) (LR) (LCR) (NSFR)
min CR,LR,LCR,NSFR CR + (1 lev) LR + 1/c LCR CR + LR + LCR + NSFR mD
(8) (Deposit rate) (loan rate)
k(PD) CR + 1/c LCR [(1 rs)/(1 as)] NSFR E[ML] CR + LR + LCR + NSFR 0
Now, given an optimal solution x* = D*, L* to Model (7), there exists an optimal dual solution * = (*CR, *LR, *LCR, *NSFR) to the dual problem of Model (7) (see Model (8) below) so that s** = 0 where s* = (1 D* k(PD) L, 1 lev D*, (1-L*)/c D*, [(1 rs)/(1 as)] L D*) is the slack vector of the optimal solution x*, i.e., complementary slackness holds. Suppose now that for a given optimal solution x*, a certain constraint i is tight. Then the associated dual variable denotes the marginal cost of this constraint, i.e., being able to relax the constraint by increasing its right-hand side by would result in additional profit by i* for small . This dual program can be viewed as an optimal decomposition of the rates E[ML] and mD into acceptable risk levels under Basel III. More precisely, while the primal linear program (7) is the program that the banks solve, i.e., maximizing profit under risk constraints, the dual linear program (8) is the one that is solved by the regulator, i.e., it minimizes the risk given certain required minimum rates for deposits and loans. Consequently, this linear programming model highlights in a nice way the interaction between the banks and the regulator. In particular, for an optimal primal-dual solution pair the objective values of the primal (7) and the dual (8) coincide. As the dual has only two constraints, at most two of the four variables of dual can be non-zero (in the non-degenerate case), thus showing in a dual fashion 169
that it is two of the four constraints constraining the risk. We also would like to stress that standard sensitivity analysis from linear programming can be readily applied to Model (7). Moreover, the reduced costs of non-basic variables can be used to estimate the costs of changing the business model from one optimal solution to another.
Assets 100%
-4% reserve
Ad-hoc scenarios (stress tests) a robust optimization approach

Regulators require banks to meet their regulatory ratios under any circumstances, i.e., in all potential states of the economy and at any given point in time. Being always compliant with the Basel-III regulations implies in particular that even after an adverse scenario, a so-called stress scenario, banks still have to meet their ratios. In our two-product example, a stress scenario could be a considerable loan default and/or a considerable deposit withdrawal. Any change of the balance sheet is reflected as a simultaneous shift of one asset and one liability position as Figure 4 suggests a 10 percent loan default translates into a 10 percent capital reduction. Hence, the bottom and right-hand side borders move upwards and to the left respectively. A 4 percent deposit withdrawal translates into a simultaneous reduction of reserve shifting the left-hand and the upper borders to the right and below respectively. The shrunk square represents the theoretically maximum area. In fact, for exactly this reduced square we have to meet the ratio requirements. As a consequence, the feasible set is considerably reduced when we want to be compliant in the adverse scenarios. Applying our optimization model to this reduced area leads to a solution that holds additional buffers for adverse scenarios. Such a buffer is mandatory for the capital (capital conservation buffer) in Basel III. The size of the additional buffer is chosen such that the bank fulfills the ratio requirements even after a stress test. We address the impact of stress scenarios using the standard techniques from robust optimization. For a detailed introduction the reader is referred to Ben-Tal et al. (2009) and Bertsimas and Sim (2004). In order to robustify Model (7) we define an uncertainty set U = {(L, D) : (L, L) is stress scenario} which contains the stress scenarios in term of changes in L and D.
-4% deposits
-10% capital
-10% loans
Liabilities 100%
t1: Stress test: 10% losses on loans, 4% deposit withdrawals Basel III constraints t0: t1: Basel III feasible set t0: t1:
(C) Basel III: feasible solution set in a two-step dynamic setting
Figure 4 Feasible set in static and dynamic setup
Using linear programming duality, this in turn can be expressed again as a linear program with few auxiliary variables [Bertsimas and Sim (2004)].
Probabilistic scenarios
While in the previous a robust optimization approach for stress test-like scenario was presented we will now focus on a probabilistic version of our model. Whereas in the case of stress scenarios the probabilities of the considered events are unknown and therefore robust optimization is the method of choice, in the case of probabilistic scenarios we assume that the probabilities of the scenarios are known. Then we can use methods from stochastic optimization to solve our model under stochasticity assumptions. Until now, only constant variables, or stochastic variables reduced to
Then the robustified version of Model (7) is given by

max D,L ML L + mD D
their expected values, have entered the optimization problem. The assumption of a stress test was a first attempt to account for the inherent (9) stochasticity (risk) in loans and deposits. However, the introduction of a scenario distribution provides more flexibility to introduce managements risk aversion. Instead of always fulfilling the constraints (compliance confidence of 100 percent), it would be more reasonable under certain as-
s.t. (D + D) + k(PD) (L + L) 1 (D + D) > 1 lev (D + D) + 1/c(L + L) 1/c (D + D) [(1 rs)/(1 as)] (L + L) 0 170 (D, L) U (D, L) U (D, L) U (D, L) U (CR) (LR) (LCR) (NSFR)
sumptions to satisfy the ratio requirements with high probability, i.e., at a (lower) confidence level only. We can formulate these probabilistic guarantees with chance constraints:
max D,L
E[ML] L + mD D
(10)

s.t.: P[Loss] LOSS P[C < k(PD) L] CR P[(C/L+R) < lev] lev P[R c D] LCR P[as D + C rs L] NSFR (CR) (LR) (LCR) (NSFR)
1-D(mD) lev 1-L(mL) c D(mD) as D(mD) + (1 D(mD)) rs L(mL) X/m 0
(LR) (LCR) (NSFR) (MVF)
The last constraint characterizes the mvfs to be decreasing in the marginvolume space. Note that mvfs are only necessary for products, not for instruments. In contrast to products, banks can determine the volumes of instruments mainly via their capital management (capital buffer), liquidity management (liquidity buffer), risk management (risk transfer), and debt management (structural funding).
Introduction of adjustment cost

The current optimization problem is not sensitive to the initial product mix. However, a bank will find it difficult to completely close/open a business line due to the fixed costs associated with long-term contracts (like staff, IT infrastructure, etc.). We model these costs as adjustment costs that increase in relation to the changes (the greater/smaller the changes the greater/smaller the costs). This introduces a local stickiness (optimization) to the current product mix and reduces on average the distance between initial and optimal mix.
max D,L E[ML] L + mD D + aD |D D0| + aL |L L0|
Multi-period and multi-product extension

Multi-period extension
So far, the current model has only addressed the choice of an initial product mix. However, later adjustments to new market conditions were not possible, it was assumed that management was passive. While this is a perfectly fine assumption for understanding the interplay of the various
(11)
factors, it is counterintuitive in practice. In fact, management learns and usually adjusts the business mix based on new information that are available at a later point in time. Our model can be easily extended into this direction by including time indices for the decision variables and considering a time horizon. In this case in particular, adjustment costs are important as otherwise the bank would perfectly adjust business from one time step to another.
s.t. as before where D0 and L0 denote the initial deposit and loan volume respectively.
Margins as decision variables

Although banks plan volumes, volumes are not quantities that banks can directly decide. What banks do set are rates and margins. The volume decision remains with customers by accepting or refusing the offered contract. The modeling approach that translates margins into volumes is a margin-volume function (mvf). Mvfs are decreasing in the marginvolume space. Steep mvfs characterize very monopolistic products where customer volumes hardly react to any margin adjustments. Flat mvfs characterize very competitive products where customers hardly tolerate margin changes. Mvfs are an approach to more directly incorporate customer preferences and behavior. Setting rates and margins gain in importance due to Basel III as these reforms increase the liquidity and capital cost for many products. Banks have to decide whether they want to offload the additional costs on to client rates to recover pre-Basel III margins or whether they partly absorb the increased cost preserving preBasel III rates (not margins!). Mvfs usually introduce nonlinearities into the optimization problem. Formally, the optimization problem with margin setting can be written as follows: max mD, mL E[ML(mL)] L(mL) + mD D(mD) s.t.: 1-D(mD) k(PD) L(mL) (CR) (12)
Extension to an arbitrary number of products

Apart from the extension to include various time steps, the decision vector can also be extended to include an arbitrary number of products. In the basic version, the bank has only two positions on each balance sheet side, implying that one position is endogenously determined. However, real-world banks have many more products and instruments. In particular, the two products should be interpreted like a representative loan or representative deposit. Substituting stylized products by several concrete products introduces more granularity. On the instrument side,
Assets A t Interbank lending IB t+ Loans L t-1,t
Liabilities L t Interbank borrowing IBtDeposits Dt Securitized Debt B t-1,t
Loans L t,t
Securitized Debt B t,t Capital C t
Figure 5 Balance sheet for multiple product setup
171
banks directly set their margins. While this extension can be easily incorporated into our model by sheer size, a closed-form analytical characterization is not possible anymore. Nonetheless, the characterization of the models based on linear programming theory still holds.
B. Laws of motion: Lt,t = lt Lt,t+1 = Lt,t (1 - DF) Bt,t = bt Bt,t+1 = Bt,t IBt+= ibt+
Product selection
Figure 5 shows the resulting balance sheet. In particular, the bank is engaged in the traditional loan-deposit business, lends and borrows at the interbank market, and issues debt. Our product motivation was twofold: economical and technical. The economic motivation to include loans, deposits, interbank lending/borrowing, and debt issues is to cover the broad majority of products of a commercial bank. The technical motivation is to cover various aspects of the regulatory setup. The regulatory model prescribes dynamic weights for the net stable funding ratio: for products with a remaining maturity of less than one year the percentages are different than those above one year. Every long-term product migrates from the above 1Y bucket to the below 1Y bucket on its way to maturity and changes the weightings. In order to capture this effect, products with a maturity structure are required. Our positions loans and securitized debt have agreed initial maturities of more than one period. Interbank borrowing and lending as well as deposits are modeled as on demand (non-maturing). This contrasts the work of De Nicol et al. (2011) where only on demand-products are used. This approach cannot capture the dynamic aspect of changing regulatory parameters. Note that the number of products (decisions variables) is related to the number of regulatory constraints (ratios) that might be binding at an optimal solution, i.e., the one that maximizes the objective function. In a twoproduct setup, only two of the four Basel III ratios can be binding at any optimal solution. With four or more products, all ratios might be binding.
IBt = ibt Dt = Dt-1 (1 - WD) + dt C. Constraints: Collateral constraint: b (Lt,t + Lt-1,t) (ibt + Bt-1,t + Bt,t), b [0,1] Economic solvency: Ct 0 B3: Minimum capital ratio: Ct kVasicek (PDL)(Lt-1,t + Lt,t),ci [0,1] B3: Leverage ratio: Ct lev (Lt-1,t + Lt,t + IBt+),lev [0,1] B3: Liquidity coverage ratio: IBt+ c1 Dt + c2 IBt asf, rsf i [0,1] D. Domains: bt,t ,ibt+,ibt, dt, lt,t [0,1] E. Stochasticity: DF = Ber[0,1] WD = Ber[0,1] Our linear program models bank management that seeks to maximize terminal shareholder value (cumulated and compounded earnings after tax). We denote earnings after tax as P&Lt. The P&L consists of earnings before tax (+ EBTt ) less the tax itself ((EBTt )) and less adjustment cost to in-/ decrease the production volume. Adjustment cost can be thought of as costs that occur when laying off/engaging employees or when reducing/ expanding the IT infrastructure. The P&L is an integral element of the internal available cash flow. As impairments are value, but not cash flow relevant, the cash flow definition corrects for impairments. Beyond the cash flow relevant portion of the P&L, the cash flow is driven by new and maturing assets and liabilities. We assume that all profits are retained such that no dividends are paid out. The laws of motion describe the evolution (transition) of products from t to t+1: the evolution of loans and deposits is uncertain: a fraction DF of loans default, and a fraction WD of deposits are withdrawn. Both fractions are the stochastic drivers of the model. In fact, the Basel III minimum capital and liquidity coverage ratios require banks to hold capital and liquidity as buffers against the unexpected portion of defaults and withdrawals. If the two quantities were deterministic, no unNo funds disappear: CFt = 0
B3: Net stable funding ratio: asf Dt + Bt,t + Ct rsf1 Lt-1,t + rsf2 Lt,t,
Model description
Let () be a concave function defined on R+, (.) a convex, asymmetric function defined on [-,+], and m(.) a quadratic, convex, asymmetric function defined on [-,+]. max
lt,bt,dt ibt+,ibt
( 1 )(T-1-t) Et [P&Lt] t=0 b
A. Definitions: Net interest margin: NIIt = L (Lt-1,t) + L (Lt,t) + rIB (IBt+ - IBt t ) D (Dt) rB (Bt-1,t + Bt,t) Impairments: Impt = DF (Lt-1,t + Lt,t) Earnings before Tax: EBTt = NIIt Impt Profit and loss: P&Lt = EBTt - (EBTt) - m(Lt,t) - m(dt) Final cash flow: CFt = P&Lt + Impt + Lt-2,t - lt,t - WDDt-1 + dt Bt,t Bt-2,t 172
+ Start balance sheet: C0 = L0 + IB0 - IB0 - D0 - B0
ibt+
+ ibt +
expected portion would exist and no buffers/ratios were needed. Note that our credit shock leads to default on loans which contrasts with the work of De Nicol et al. (2011) where the credit shock only affects the return from loans (i.e., defaulted loan rates), but not their volumes.
Updating capital: Ct = Ct-1 + P&Lt

Also note that this paper only looks at single banks but not the full network of banks (systemic risk). In fact, banks constitute a network via their interbank lending and borrowing (designated IB+t and IB-t for lending and borrowing respectively). In our model, banks interact with the interbank market, hence the anonymous aggregate of all other banks. We do not model a banking network tracking the bilateral lending/borrowing links. Consequently, the model cannot capture the feedback effects that the bank suffers/enjoys from the reaction of other banks to our banks interbank decisions. Models that capture the feedback effects from other banks are usually based on banking network models [Pokutta and Schmaltz (2011)]. The downside of the interbank market is twofold. Firstly, loans to other banks might default or deteriorate in credit quality. Secondly, the interbank market might dry up and the optimal volume of interbank funding ib-t might simply not be realizable in the money market. Although not catering for the stochasticity of the fundable interbank volume, we mitigate this effect by requiring that interbank funding is fully collateralized (secured funding). Collateralized funding has proven to be quite stable during the recent financial crisis. Hence, any desired collateralized interbank volume should be realizable in reality as well, even in adverse market circumstances. Hence, there is no need to capture stochastic interbank funding limits if the model only allows for secured funding. The bonds market could also dry up as well, such that the optimal bond volume bt,t is simply not fundable. Similar to the interbank funding, we require bonds to be collateralized (covered bonds) to mitigate the effect that we do not explicitly model stochastic bond issuance limits. We adopt the constraint proposed by De Nicol et al. (2011). The no funds disappear constraint ensures that incoming cash flows equal outgoing cash flows in each period. If the outgoing cash flow exceeds the incoming (including the funds generated from bond issuance and interbank borrowing) the bank is illiquid. The economic solvency constraint postulates that the capital position should be positive. Otherwise, the bank is insolvent. The Basel III constraints are analogous to those from the previous models: they are linear in the product volumes as the regulatory parameters are constant and do not depend on the volume.
Price of compliance
The linear program ensures via the Basel III constraints that all ratios are fulfilled in every scenario and every time step. The reduction in the eligible product set comes at the cost of P&L reduction compared to the unconstraint banks (the same linear program but with relaxed Basel IIIconstraints) as (13) suggests. Price of compliance := LPBasel III [lt, bt, dt, ibt+, ibt ] LPUnconstraint [lt, bt, dt, ibt+, ibt ] (13)
Note that the P&L reduction is an aggregate measure across all four ratios. Information about the (marginal) P&L reduction of a single constraint is obtained from the shadow prices as detailed above. The shadow price of a given constraint measures how much one would gain in terms of P&L by relaxing the constraint by 1 percent.
Model implementation
The formal description of a model is mathematically sound and facilitates the comparison to other models, but it is not necessarily a ready-to-use guide for implementation. Consequently, we want to provide a complementary implementation guide to financial executives for the multi-period, multi-product model (that nests the simpler models above). Our model provides assistance in planning for profitability and Basel III compliance for banks. It addresses both CFOs (for profitability) and CROs (for risk compliance) of banks as it naturally encompasses risk (in the form of regulatory constraints labeled as B3-constraints) as well as profitability (in form of the P&L as an objective function). Beyond its support for quantitative decision making, our model promotes the integrative thinking between risk and finance. The model can be interpreted as an (algorithmic) CFO-CRO-team that determines the most profitable product mix that is still Basel III compliant. In particular, the multi-product version determines the optimal volumes of long-term loans (i.e., consumer loans, labeled lt), long-term funding (i.e., corporate bonds, labeled bt), deposits (i.e., demand deposits, labeled dt), tended to multiple periods it does not only provide the optimal business
and short-term interbank lending and funding (labeled ibt+ and ibt ). Exmix, but the most profitable sequence of product volumes. With respect to data, our model requires a set of standard inputs that
Solving the model

As the program is convex and under suitable assumptions even linear, it can be quickly and efficiently solved with standard optimization software. However, note that the complexity grows exponentially in the number of time steps given that we consider at least two scenarios in each time step. Although the program might not necessarily be linear, in the convex case we can still extract dual information from the Lagrange multipliers of the associated KKT system. Again, we obtain shadow prices for each constraint reflecting how expensive (in terms of marginal value for the target function) each constraint is in the optimum.
should be available in every bank: for all products, product interest rates, LCR and NSFR-parameters are necessary. Additionally, assets need the standard risk parameters probability of default (PD) and loss-given defaults (LGD). These risk parameters are converted into required minimum capital via the Vasicek/Gordy-model of the IRBA approach in Basel II/ III. The required capital amount is the coefficient for the B3 capital constraint. For loans and deposits, two additional parameters, default and 173
withdrawal rates are needed to parameterize the stochastic of the model. Estimations of these rates are usually available from the credit and liquidity risk management departments. Once the product parameters are known, the model can be solved by any linear programming solver, such as CPLEX 11.1.0 [IBM (2008)], Gurobi 1.1.1 [Gurobi (2009)], or Scip 1.1.0 [Achterberg (2006)]. Our model uses both a standard objective function and standard (linear) constraints such that an implementation can be realized with the manuals that come with the software packages. Global banks are organized across many departments. Hence, the decision variables of our model are located in different departments. Loans (lt) and deposits (dt) are planned in the business units. Debt issues are labeled funding plan in a bank and located in the treasury. Interbank lending and depositing is decided in the liquidity management, which is also part of treasury. The main planning challenge in banks is to provide an optimal outcome across many decentralized planned inputs. Our model is a centralized planner that only uses product (risk and profitability) information as inputs, bundling them into an optimal product mix. In this regard, our model overcomes sometimes poorly coordinated decentralized planning. Although accommodating liquidity and funding management, an active capital management with dividend and stock issuances as decision variables could make our model more realistic. However, the current version accommodates a passive capital management, i.e., retained earnings as the only source for increasing the capital basis. An active capital management is left for future research.
probability of future defaults (PD). The potential deterioration of credit risk parameters oblige banks to hold a capital buffer. A potential deterioration is only to be considered for a systemic credit shock. As our credit shock affects the whole loan book and as the Basel III credit risk model (Vasicek model) is driven by a systemic credit risk shock, we can reuse the driver for the actual defaults DF to also drive PD deterioration. We presented a linear programming-based framework to analyze the impact of the new Basel III regulation on banks. While we considered the case of banks with two products in order to maintain analytical tractability, we presented a more general version with several products. The strength of using linear programming for characterizing the effects of the Basel III regulation is the strong duality theory available for the linear program which provides us with crucial economic information such as marginal prices, reduced costs, and which leads to a natural formulation of the dual problem. This dual problem naturally resembles the economic role of regulators: while banks try to maximize profits given the ratio constraints, regulators try to minimize the risk that the ratios are not respected.
References

The products of our multi-product model have been chosen to cover several product features (see product selection above). Consequently, new products can be easily integrated. Loans with different ratings/PDs are treated as a new loan category with a new coefficient in the capital constraint. Similarly, deposits with a different deposit rate are treated as an additional deposit type. Term deposits can be treated as debt issues, since it does not matter for the model whether a funding instrument is traded on a secondary market (securitized debt) or not (term deposits). Currently, the maturity spectrum only covers on demand and longterm. This spectrum can be extended by introducing additional time and decision points in the model. Similar to the representative products, new products must be specified by their product rates, LCR and NSFR coefficients and additional parameters for assets and deposits. A natural extension of our framework is the impact of the credit shock on PDs: banks capital ratios are not only affected by actual defaults (reduction in available capital), but also by credit quality deterioration which translates into increased probability of defaults. Within the Basel III framework, the latter translates into higher (minimum) capital charges. A 174 credit shock should, therefore, not only push actual defaults, but also the

Achterberg, T., 2006, Scip a framework to integrate constraint and mixed integer programming, Technical Report 4-19, Zuse Institute, Berlin Basel Committee on Banking Supervision, 2011, Basel III: a global regulatory framework for more resilient banks and banking systems, June, download at: http://www.bis.org/publ/ bcbs189.html Ben-Tal, A., L. El Ghaoui, and A. S. Nemirovski, 2009, Robust optimization, Princeton University Press Bertsimas, D. and M. Sim, 2004, The price of robustness, Operations Research, 52:1, 3553 Black, F. and J. C. Cox, 1976, Valuing corporate securities: some effects of bond indenture provisions, Journal of Finance, 31, 351367 De Nicol, G., A. Gamba, and M. Lucchetta, 2011, Capital regulation, liquidity requirements and taxation in a dynamic model of banking, http://ssrn.com/abstract=1762067 or doi:10.2139/ssrn.1762067 Gurobi, 2009, Gurobi 1.1.0 Mixed Integer Linear Programming Solver, Technical Report, Houston IBM, 2008, CPLEX 11.0 Mixed Integer Solver, Armonk Leland, H. E., 1994, Corporate debt value, bond covenants, and optimal capital structure, Journal of Finance, 49:4, 12131252 Leland, H. E. and K. B. Toft, 1996, Optimal capital structure, endogenous bankruptcy, and the term structure of credit spreads, Journal of Finance, 51:3, 9871019 Perotti, E. and J. Suarez, 2011, A pigovian approach to liquidity regulation, International Journal of Central Banking, 7, 3-41 Pokutta, S. and C. Schmaltz, 2011, A network model for bank lending capacity, http://ssrn. com/abstract=1773964 or doi:10.2139/ssrn.1773964 Santomero, A. M., 1984, Modeling the banking firm, Journal of Money, Credit & Banking, 16:4, 576602 Schmid, O., 2000. Management der Unternehmensliquiditt. Paul Haupt, Stuttgart
PART 2
A Risk Measure for S-Shaped Assets and Prediction of Investment Performance

Qi Tang Reader in Mathematics, Department of Mathematics, University of Sussex1 Haidar Haidar Department of Mathematics, University of Sussex Bernard Minsky Head of Portfolio Analysis and Risk Management, International Asset
Management Ltd
Rishi Thapar Senior Quantitative Research Analyst, International Asset Management Ltd
Abstract
In this paper, we study the option valuation of S-shaped assets. S-shaped assets are frequently encountered in technological developments, grant funding of research projects, and to a degree, hedge funds and stop-loss controlled trendfollowing investment vehicles. We conclude that the quantity 2/ (variance of return/expected return) replaced the traditional variance risk measure 2 in the Black-Scholes option
valuation formula. We further study the interesting property of 2/ in predicting the turning point of performance of a portfolio of hedge funds in the early months of 2008 (and indeed, for earlier historical turning points).
The authors thank the support of London Technology Network for support in technology related training which brings S-shaped assets into our view.
175
In technology development, it is well known that the S-curve descriptor is widely used in assessing the maturity of technology projects [Nieto et al. (1998)]. Our study is about the valuation of assets whose price changes follow the pattern of an S-shaped asset. That is, the asset prices either remain unchanged or go up at each time interval of observation. In the equity investment world today, there are mainly two kinds of funds run by professionals:
deviate from the traditional higher risk implies higher return behavior. The stress in the pre-crisis financial market implies that higher risks lead to lower returns. This observation is not obvious when using a particular fund or index as observation reference, it can only be observed through a systematic approach as we explain below in detail. In this paper, we follow the standard binomial formulation to establish the value of call/put options for S-shaped assets. Under appropriate assumptions, we further derive the approximate Black-Scholes (BS) European option pricing formula. It is interesting to note that in our situation, the involvement of the variance of the asset returns in the BS formulae is replaced by 2/ (variance of asset returns)/(expected return). And from this, we further reveal how hedge funds (as a group) anticipate the arrival of financial crises. In our mathematical derivation, we use formal asymptotic analysis methods. Compared to standard BS formula derivation, the additional assumption for the variant Black-Scholess approximation formula to hold is that 2/ is small compared to 1. Theoretically, this can be justified in S-shaped assets when the increase in the asset value is in one direction and is reasonably uniform. In this case, it is easy to demonstrate that 2 2. Practically, this is justified by the hedge funds return data provided by International Asset Management (IAM) in non-financial-crisis periods (see graphs in the Appendix). There have been many discussions in recent years as to why standard deviation is not an appropriate measure of risk. Newer risk measures such as Sharpes Ratio, VaR [Linsmeier and Pearson (1996)], CVaR [Acerbi and Tasche (2002)], Omega [Shadwick and Keating (2002)], maximum drawdown [Chekhlov et al. (2003)] and many others have been introduced over the years. In particular, Tang and Yan (2010) observed that even using the standard binomial option value derivation formula, it is possible to derive corrective terms in addition to standard variance that indicate that the trend of asset price movement could get tired and turn in the opposite direction. These observations are backed up by real financial data testing. In this article, we argue that 2/ is a new kind of risk measure or a new starting point to search for a new class of risk functions. Some of the interesting behaviors it exhibits give it a new dimension in assessing absolute return assets. We have discovered that for IAM selected quality hedge funds,2 if we use Monte-Carlo methods to generate portfolio weights between 0 and 1 and plot the corresponding (2/, ) points, when the performance deteriorate, the shape of the cluster of points of (2/, ) begins to change shape
Mutual funds these funds seek relative return [Harper (2003)], and their performance is measured against certain benchmarks, such as S&P 500, MSCI world index, or other indices. So if the index returned -5 percent and the fund returned -2 percent for the period, the fund is regarded as an excellent fund as it beats the market by 3 percent.
Hedge funds these funds seek geared, absolute return, and their performance is usually measured against 0 (or in some cases, against the risk-free bank deposit interest rate, which is much harder to achieve). Hence, at every report time window, the fund is expected to have a positive return (or in some cases, better than depositing the capital in the bank) compared to the end of the previous period. Consequently, in the case of hedge funds, negative returns are regarded as highly undesirable.
In real life, when we read the hedge fund monthly returns report, occasional monthly negative returns do appear. In the normal times, the negative returns are very rare, but during financial crises, we notice that the number of negative returns is more pronounced. Hence, we anticipate that if we model hedge funds using the concept of S-shaped assets during normal times, we may have interesting findings when we approach financial crises. It has been widely noticed that the bond market feels financial crises before they happen. We are out to demonstrate, in this paper, using the three most recent financial crises as examples, that hedge funds feels financial crises before they take place with some clear measurable indicators. In particular to this topic, we notice that in his report to the U.S. House of Representatives, Lo systematically analyzed the existing methods of studying systematic risks in the financial market, he postulated that the hedge funds should be regarded as a group, and suggested that researchers should use some systematic approaches to look at the risks that the financial markets are facing [Lo (2008)]. In particular, he quoted some of the network approaches in studying systematic risks in the hedge fund world. Taking the idea of systematic risks, we look at the hedge fund world from a different angle, but follow a similar approach: we use differently weighted portfolios of a large set of hedge funds and calculate the risk measure derived by us, to conclude that when ap176 proaching a financial crisis the hedge fund world as a whole begins to
Available selected funds are 16, 27, 38, 62, and 229 in periods Jan/96-Dec/98, Nov/99Dec/01, Feb/01-Apr/04, Jun/03-Nov/05, and Jan/05-Jun/11 respectively.

dramatically with ample warning in advance note that withdrawals from hedge funds usually face a three-month notice period or even temporary suspension in bad times, hence warnings in advance are particularly important. We point out that our prediction method is for hedge funds only and cannot be readily applied to other investment vehicles where absolute return is not the primary investment objective. We also applied the same technique to draw similar conclusions for the 1998 and 2001 financial crises. Due to the fact that there were not as many hedge funds during those historical periods, we used all the hedge funds available at the time and we need to be aware that these are not the same data pool in each case.
t = p(u-1) 2t = p(1-p)(u-1)2 2/ = (1-p)(u-1) (5)
These relationships imply that there is at least one variable which can be set freely. We have to make the following assumptions to further our discussions: Assumption 1 in our scenario where asset prices can only go up or stay stationary, we assume that the variance 2 based on daily returns of the asset is a small quantity compared to . Justification is the expected unit time period return, it is usually a constant of a few percentage points. In our case where asset prices can only go up or stay stationary, the daily increase rate of the asset cannot exceed . Since 2 is the average of (daily return average return)2, hence we can justify that, under usual circumstances where there is no high volatility in the underlying asset prices, we should expect: 2 c2 << , hence, when t is regarded as a small quantity (u-1)2 = (2/ + t)2 << u-1 = 2/ + t This relationship cannot be directly implied from (5).
The assumptions concerning expected return and risk

We adopt the standard binomial formulation in our discussion. Assuming that: 1. is the expected unit time period return 2. The volatility of the security is 3. S is the price of the security at t= 4. St is assumed to be a random variable which either takes the value S0u (u>1 which is the up movement) with probability p, or stay at the value S0 with probability 1-p. Hence we have E(St) = pS0u + (1-p)S0 Var(St) = S2 p(1-p)(u-1)2 0 (1) (2)
Binomial formulae
Put option
In addition to the price movement patterns assumed above, we add the following assumptions: short selling is permitted; fractions of the security are permitted to be traded; there are no trading transaction costs and no dividend; and there are no arbitrage opportunities. Denote by P0 the value of the put option on this security at t0, and by P+ (P-) the corresponding option values at t = t0 +t if the underlying prices goes up (stays the same): P0 = P(S0, t0), P+ = P(S0u, t0+t), P- = P(S0, t0+t) It is clear that for put option, PT = Max(E-S(T), 0). Now consider a portfolio consisting of one put option, and a short posi-
Let be the expected unit time period return, be the one time period risk of the asset. Using standard formulae that E[(St S0)/ S0)] t => E(St) S0(1 + t) Var[(St S0)/ S0)] 2t => Var[(St) S2 2t 0 We get {pS0u + (1-p)S0 S0(1 + t), S2 p(1-p)(u-1)2 0 S2 2t 0 (3)
tion of quantity to be specified later. We establish the value at t0+t: if the price has stayed, the portfolio has value P- S0; and if the price has moved up, the portfolio has value P+ S0u.
Asymptotically (when t is small), we can regard the above as equalities. Solving u, we obtain u = 1 + 2/ + t We can also obtain the following relationships: (4) We choose so that the portfolio has the same value in both cases: P- S0 = P+ S0u, => = (P- P+)/S0(1-u) We now have, using standard non-arbitrage theory, (6) 177
Proposition 1 by the principle of non-arbitrage, we have P0 S0 = Present Value Of (P- S0) Substituting (6) into (7) and rearranging, we get {P0 S0 + (P- S0) exp(-rt) = (P- P+)/(1-u) + [(P+ P-u)/(1-u)]e-rt; PT = Max[E-S(T),0]} where r is the risk-free interest rate. The proof is to use standard arbitrage arguments and we omit the details [interested readers can refer to Higham (2004) and Willmot et al. (1995) for details]. (8) (7)
= [(P- P+ )/(1-u)](1 e-rt) + P- e-rt
+ [P(S0, t0) + P/t (S0, t0) t](1-rt) + O(t2)+ O((u-1)2t) Here we have used (5) in setting O(t2/(u-1))=O[(p/) t] O(t). Notice that from (5) and Assumption 1, the terms O(t2) and O((u-1)2t) are higher order terms compared to O(t). Further simplifying the expression: 0=[P/S(S0, t0)S0 + 2P/S2(S0, t0) (1-u)]rt + P/t(S0, t0) t P(S0, t0)rt + O(t2)+ O[(u-1)2t]. Dividing t on both sides, we get, using (4), 0=[P/S(S0, t0)S0 + 2P/S2(S0, t0) S2 (2/ + t)]r + P/t(S0, t0) 0 P(S0, t0)r + O(t)+ O[(u-1)2]. Hence
= [P/S (S0, t0)S0 + 2P/S2 (S0, t0) S2 (1-u)]rt 0
Call option
Proposition 2 similarly, let C be the value of the call option, using same notations, we have {C0 S0 + (C- S0) exp(-rt) = (C- C+)/(1-u) + [(C+ C-u)/(1-u)]e-rt; CT = Max[S(T) E,0]} (9)
0=[P/S(S0, t0)S0 + 2P/S2(S0, t0) S2 (2/]r + P/t (S0, t0) P(S0, 0 t0)r + O(t)+ O[(u-1)2]. Following Assumption 1, under normal circumstances, we can regard O[(u-1)2] as small quantities, hence we can drop the terms of order O(t) and O((u-1)2) to get P/S (S0, t0)S0 r + 2P/S2(S0, t0) rS2 2/ P(S0, t0) r + P/t(S0, 0 t0) = 0 Using S0, t0 as standard independent variables, we have shown Proposition 3 under our assumptions, the put option value for our uponly asset is determined by P/t + rS P/S + r2S2/ 2P/S2 Pr = 0 with P(T) =
where r is the risk-free interest rate.
Asymptotic approximations for values of options

This section is rather mathematical. The main conclusion is that if we compare the differential equations derived for the standard Black-Scholes equation, we have 2/ in our equation replacing 2 in the standard BS equation. This implies that 2/ is a more appropriate quantity in describing risks for S-shaped assets.
Put option
Take put option as an example. Following the ideas developed in Friedman and Littman (1994), using (4) and (5), we have P+ = P(S0u, t0+t) = P(S0, t0)+ P/S (S0, t0) S0(u-1) + 2P/S2 (S0, t0) S2 (u-1)2 0 = P(S0+ S0(u-1), t0+t)
Max[E-S(T),0]
(10)
This formula is also true when r, the risk-free interest rate, is dependent on t (possibly also on S).
+ P/t (S0, t0)t + O(t2) + O[(u-1)3] + O[(u-1)t] P- = P(S0, t0+t) = P(S0, t0)+ P/t(S0, t0) t + O(t2)
Call option
Same assumptions and arguments lead to Proposal 4 under our assumptions, the call option value for our up-only asset is determined by C/t + rS C/S + r2S2/ 2C/S2 Cr = 0 with C(T) = Max[S(T)-E,0] (11)
2 P- P+ = P/S (S0, t0)S0(u-1) 2P/S2 (S0, t0) S0 (u-1)2
O(t2)
+O((u-1)3)+O(t(u-1))
P(S0, t0) = (P- P+ )/(1-u) + [(P+ P-u)/(1-u)]e-rt 178 = (P- P+ )/(1-u) + [(P+ P-)/(1-u)]e-rt + P- e-rt

This formula is also true when r, the risk-free interest rate, is dependent on t (possibly also on S).
0.018 0.017 0.016
0.018 0.017 0.016 0.015 0.014 0.013
The solutions to the PDE problems

Assume that r, , and are constants. We note that the expressions in (10) and (11) are similar to the Black-Scholes equation (replacing 2/ by 2, we will have Black-Scholes equations), and satisfy similar boundary conditions that: S 0 => PE, C0, and S => P0, CS. Following the same argument as for Black-Scholes equations, let d(S,t) = [2 log(S/E) + (2 + 2/)r(T-t)] 2 r/(Tt)
0.018 0.017 0.016
0.012 0.011 2.5
0.012 0.011
6.5
x 10
x 10
Figure 1 Return against our risk function (the portfolios generated have performed reasonably. Risk-reward seems to be in proportion)
0.015 0.014 0.013 0.012 0.011
and F(z) = 1/2 Z exp (-y2/2)dy - for put option, we have P(S,t) = E exp[-r(T-t)] F[ r/(Tt) -d(S,t)] S F[-d(S,t)]. For call option, we have C(S,t) = S F[d(S,t)] E exp(-r(T-t)) F[d(S,t) r/(Tt)].
0.01 0.009 0.008 0.01
0.011 0.01 0.006 0.008 0.012 0.014 0.016 10/0609/07 (RiskF,mu) 0.018 0.02
0.012
0.014
0.016 0.018 0.02 0.022 12/0611/07 (RiskF,mu)
0.024
0.026
0.028
Figure 2 Return against our risk function (The portfolios generated have performed reasonably. However, the second graph begins to show that larger risks may not bring higher returns)
How the S-risk function tells the trend change in a portfolio of hedge funds
In the previous sections, we have established an important fact: as far as option value is concerned, under not too volatile market conditions, our S-risk function 2/ is a more suitable measure of risk. Now we demonstrate how this new risk measure can be used to predict the performance trend change of a portfolio of hedge funds. It is well known that one of the selling points of hedge funds is the absolute return. It is, therefore, possible to view hedge funds as a kind of S-shaped assets (in real life, this is not true but not too far from truth in a rational market especially when market is calm and 2/ << 1). International Asset Management (IAM) is a fund of hedge funds based in London. IAM researches the hedge fund market and builds portfolios of hedge funds for its clients. IAM provided us with anonymized hedge fund returns data dated from November 1999 to June 2011, which covers the summer of 2008 when volatility of the financial market shot up. We use the Monte-Carlo method to generate random nonnegative portfolio weights (w1, w2, , w60) twenty thousand times such that w1 + + w60 = 1. Using one year historic data to calculate and plot the graph of (2/, ) (Figures 1 to 6). These tell us that using our S-risk function 2/, the warning signs would have been clear by January to February 2008. Taking into account the time it takes to arrive at a decision and an average of two months notice
10
x 10
7 6 5
x 10
4 3 2 1
0 1 120 100 80 60 04/200703/2008 (RiskF,mu) 20 40
4 0.02
0.03
0.09
0.1
0.11
Figure 3 Return against our risk function (The first graph shows the performance deterioration clearly: high risks seem to bring lower returns bad sign for hedge funds. The second graph shows that the value of the risk measure becomes large and is no longer suitable to be used to measure risk)
period required for withdrawal, investors still would have been able to avoid the disaster for hedge funds performances of the second half of 2008 if they had taken action by March 2008. Similar simulation has been applied to the 1998 and 2001 financial storms. During these periods, we have fewer hedge funds available, hence we have chosen all funds data that existed during the period (16 funds for 1997-1998, 27 funds available for 2000-2001) and did not make any selections using any criteria. The results show that (see Figures 4 to 6 at the end of the article): 1. For 1998, the warning signs became clear in June 1998 (using nine months back data, the warning signs appeared in April, but we need to investigate further to determine whether shorter historical data gives many more false warnings). The aftermath has seen some strong neg179
ative returns from the portfolios of the hedge funds. This again clearly indicated the Russian currency crisis and the beginning of the end of Long Term Capital Management. The recovery took place in May 1999 when risk and return became positively correlated again. 2. For 2001, the warning signs became clear in November 2000. The impact of this downturn is much milder. In fact, the returns of the simulated portfolios remained in the positive territory, but the turning point shape of the (2/, ) graph has been persistent until May 2003. That means, for this considerable period of time, higher risk resulted in lower returns for the hedge fund portfolio. It is interesting that from November 2000 to May 2003, although higher risks resulted in lower returns, the actual returns of our Monte-Carlo portfolios never turned negative. Around June 2003 (shortly before and afterwards), our riskreturn graph finally turned the trend to the normal pattern where higher risk implies statistical higher returns. The hedge fund industry really flourished around June 2003, many new funds were created and the good times had arrived.
1,6 1,4 1,2 1 0,8 0,6 0,4 0,2 0
0.022 0.02 0.018 0.016 0.014 0.012 0.01 0.008
0.018 0.016 0.014 0.012
Fund universe index 3,5 3 2,5 2 1,5 1 0,5

Sep/00 Dec/00 Mar/01 Jun/01 Sep/01 Dec/01 Mar/02 Jun/02 Sep/02 Dec/02 Mar/03 Jun/03 Sep/03 Dec/03 Mar/04 Jun/04 Sep/04 Dec/04 Mar/05 Jun/05 Sep/05 Dec/05 Mar/06 Jun/06 Sep/06 Dec/06 Mar/07 Jun/07 Sep/07 Dec/07 Mar/08 Jun/08 Sep/08 Dec/08 Mar/09 Jun/09 Sep/09 Dec/09 Mar/10 Jun/10 Sep/10 Dec/10 Mar/11 Jun/11
Fund index
S-risk "beta=-1"
S-risk "beta=-4"
Figure 7 The effect of S-risk function on hedge funds investment
S&P 500
Figure 8 The effect of S-risk function on S&P investment
0.008 0.006 0.004 0 0.02 0.04 0.06 0.08 0.1 0.12 (RiskF,mu) 07/9706/98 0.14 0.16 0.18
Visual observations confirm the arrival of bad investment periods including the financial crisis in advance. Visual observations do not confirm patterns of distribution of (2/, ), so we regressed 2/ on of the samples generated by Monte Carlo. The coefficient b of generated by the regression 2/ = + b highlights the sensitivity of the S-risk function to the return. A negative b represents higher risk associated with lower return. A b value of -1 represents a unit return loss for one more unit of risk. We suggest that investors should shorten their investing periods and avoid some predicted bad periods when they receive the warning sign (i.e., b < -1) and wait until the sign recovers. The value of b to depends on the investors risk tolerance. We plot the effect of the S-risk function
0 0.02 0.04 0.06 0.08 (RiskF,mu) 12/9911/00 0.1 0.12
0.01 0.02 0.03 0.04 0.05 0.06 (RiskF,mu) 06/9705/98 One month before turning point
0.07
Figure 4 The run up to the 1998 downturn
0.028
0.022
0.018
0.016
0.014
0.012
0.02 0.04 0.06 0.08 0.1 (RiskF,mu) 11/9910/00 One month before turning point
0.12
on investing one unit of money in the hedge funds universe index, which represents an equal weight ETF of the existing funds, and in S&P 500 for the period from September 2000 to June 2011. Note that for the S&P 500 investors can sell the stocks immediately while for hedge funds we allow two months notice period to withdraw the money from the funds. There are no restrictions on buying. From what we have discussed, it can be deduced that during the financial crisis of 2000 and 2008, hedge funds performance as a whole falls far short of that of the stock market indices; however, it recovered very fast from the bottom. So the choice of -1 as a threshold leads to missing many good investing opportunities. We allow fewer restrictions on the sign and for comparison we choose a less strict threshold indicator of -4 to extend the investing time periods. We plot the effect of the S-risk
Figure 5 The run up to the 2001 downturn
10 8 6 4 2 0 2 4
x 10
0.018 0.016 0.014 0.012
0.008 0.006 0.004 0.002
6 600
400 200 400 (RiskF,mu) 11/9710/98 Four months after turning point
600
0.3 0.4 0.5 0.6 0.7 0.8 (RiskF,mu) 04/0003/01 Four months after the turning point
0.9
180
Figure 6 The aftermaths (the 1998 (first graph) downturn turning point implied serious negative performance risk. The 2001 (second graph) downturn turning point implied more risks, but no significant negative performance risk as a portfolio)
Sep/00 Dec/00 Mar/01 Jun/01 Sep/01 Dec/01 Mar/02 Jun/02 Sep/02 Dec/02 Mar/03 Jun/03 Sep/03 Dec/03 Mar/04 Jun/04 Sep/04 Dec/04 Mar/05 Jun/05 Sep/05 Dec/05 Mar/06 Jun/06 Sep/06 Dec/06 Mar/07 Jun/07 Sep/07 Dec/07 Mar/08 Jun/08 Sep/08 Dec/08 Mar/09 Jun/09 Sep/09 Dec/09 Mar/10 Jun/10 Sep/10 Dec/10 Mar/11 Jun/11
S&P
S-risk "beta=-1"
S-risk "beta=-4"

function using b of -1 and -4 for the hedge fund universe in Figure 7 and for S&P 500 in Figure 8. We can further rearrange the regression equation to get 2 = + b2, which represents a reverse efficient frontier when b is negative as more risk leads to less return.
References

Conclusion
The case where asset price cannot go down (or cannot go up) can often be used to approximate the cases of technology projects (the widely known S-curve theory) funding assessment and operational loss assessment. In these circumstances, the amount of investment that is required to support the project or to cure the cause of loss needs to be evaluated. We propose this model as a first step in an effort to value these kinds of assets. The PDE model, with exact solution, is a good approximation of the binomial formulation. Due to the explicit solution formula, the PDE solution can be used in a much wider context much more efficiently. The most meaningful part of the discussion here is that we found that the risk measure 2/ is a suitable assessment for performances of assets with expected absolute returns. In this particular context, the quantity 2/ replaced the traditional 2 in the Black-Scholes option value formula as an indicator of risk of the asset. Our assumptions are simpler than that for the standard Black-Scholes equations. To make the simplification, we do have to add an empirical type condition that 2/ is small compared to 1. These conditions are usually satisfied when the market conditions are good. We give a theoretical justification/clarification (Assumption 1), which is further justified by the hedge funds data before the market turmoil during the periods 06/06-05/07, 08/06-07/07, 10/06-09/07, and 12/06-11/07. It is clear that as market conditions deteriorated in 02/07-01/08 and 04/2007-03/2008, this assumption no longer holds. But correspondingly, the visual trend of funds performances also changes in time to give good warning about the market storm. Finally we conclude that by using risk control, investments in hedge funds can be improved by 10 percent during 2007-2008 financial crisis. By contrast, investments in the S&P 500 can be improved by over 40percent. This means that hedge funds as a whole, provide strong risk mitigating abilities when facing financial storms.
Acerbi, C. and D. Tasche, 2002, Expected shortfall: a natural coherent alternative to Value at Risk, Economic Notes, 31, 379-388. Chekhlov, A., S. P. Uryasev, and M. Zabarankin, 2003, Drawdown measure in portfolio optimization, http://ssrn.com/abstract=544742 Friedman, A. and W. Littman, 1994, Industrial mathematics, SIAM Haidar, H., 2008, Pricing options for special prices movement, MSc dissertation, University of Sussex, http://www.maths.sussex.ac.uk/~haidar/files/haidar_dissertation.pdf Harper, D., 2003, Introduction to hedge funds part one, http://www.investopedia.com/ articles/03/112603.asp Higham, D. J., 2004, An introduction to financial option valuation, Cambridge University Press Linsmeier, T. and N. Pearson, 1996, Risk measurement: an introduction to Value at Risk, working paper, University of Illinois Lo, A. W., 2008, Hedge funds, systematic risk, and the financial crisis of 2007-2008, http:// ssrn.com/abstract=1301217 Nieto, M., F. Lopz, and F. Cruz, 1998, Performance analysis of technology using the S curve model: the case of digital signal processing (DSP) technologies, Technovation, 18: 6-7, 439457 Quinn, J., 2007, Investors withdraw $55bn from hedge funds, The Telegraph [online], 5 September
Shadwick, W. and C. Keating, 2002, A universal performance measure, Journal of Performance Measurement, Spring, 59-84 Tang, Q. and D. Yan, 2010, Autoregressive trending risk function (AR-risk function) and exhaustion in random asset price movement, Journal of Time Series Analysis, 31:6, 465-470 Willmott, P., S. Howison, and J. DeWynne, 1995, The mathematics of financial derivatives, Cambridge University Press
181
182
PART 2
ILLIX A New Index for Quantifying Illiquidity

Tim Friederich Chair of Mathematical Finance, Technische Universitt Mnchen Carolin Kraus Chair of Mathematical Finance, Technische Universitt Mnchen Rudi Zagst Chair of Mathematical Finance, Technische Universitt Mnchen
Abstract
Illiquidity is a major issue in todays risk management, yet there exists no straight-forward quantification of liquidity or illiquidity. We present eight possible measures of liquidity which are partially based on micro-structural market data and examine their evolution over time in the context of the development of financial markets. These eight measures are used for creating the new illiquidity index ILLIX. We outline the calculation of this index and show that it can describe well the liquidity situation in the North American market over the period between 1998 and 2009. 183
The illiquidity of financial markets has always been a widely acknowledged source of risk among practitioners, and its importance and the attention to it has increased during the recent financial crisis. Liquidity describes the ease and speed with which a financial asset can be converted into cash or used to settle a liability, as defined by the ECB. Being cognizant of liquidity has become an indispensable aspect of risk management. However, when we want to monitor liquidity, suddenly the question arises as to how this factor can be measured. In contrast to stock prices and most financial data, liquidity itself is a hidden factor. Numerous papers in the literature have proposed using observable market factors as measures of liquidity or illiquidity. Chacko and Stafford (2004) define illiquidity as the spread between the fundamental value of a security and the price at which the security is actually traded. If this gap is large, illiquidity is high and vice versa. Besides this rather general definition, a variety of possibilities for measuring liquidity more directly have been discussed in the literature: Amihud and Mendelson (1986) propose the bid-ask spread as the most intuitive measure of liquidity. Kyle (1985) shows that trading volume and market capitalization are positively related to the liquidity of the market. Amihud (2000) suggests the ILLIQ, describing the price response to the turnover, as a measure of illiquidity. The open interest also comprises liquidity information via examining the volume of derivatives traded [DraKoln (2008)]. Other measures are not defined for single securities but for entire markets or asset classes, such as the credit spread which quantifies the risk premium of corporate bonds [Huang and Huang (2003)]. The assumption here is that the higher the spread, the lower the liquidity. Adrian and Shin (2010) propose the LIBOR-OIS spread as an indicator of illiquidity observable through the interbank market. Finally, Whaley (2008) claims that the VIX, which describes the implied volatility on the U.S. stock market, is a good measure of how investors evaluate the liquidity of a market. All of these measures express liquidity from a different point of view. Some studies have already examined the commonalities among several liquidity measures and concluded that the dependency between these measures is high, which means that they are driven by the same unobservable factors, i.e., factors which cannot easily be publicly quoted as a single measure and thus are not readily observable. Chordia et al. (2000) extensively analyze the cohesion of several measures of liquidity in regression analyses on single stocks and at an aggregated market level. They regress market- and industry-wide liquidity (as an average over a set of stocks) on the liquidity factors of single stocks and find high dependencies. Korajczyk and Sadka (2008) include price impact components in their analysis (together with spreads, turnover, and the Amihud measure) and find in a principal component analysis that the first three components can explain up to 55 percent of the variation of one liquidity factor across assets. However, their principal components measure vari184 ability of one liquidity factor over different stocks. They further analyze
the canonical correlations of the different liquidity measures and find high correlation among them. Both of these studies, as well as Hasbrouck and Seppi (2001), focus on analyzing the price impact of liquidity, stating that systematic liquidity is a factor comprised in asset prices. The aforementioned studies concentrate on the ex-post analysis of the similarities and extensively scrutinize their impacts on, for example, price movements. However, they, as well as most other studies, focus on the commonalities of one liquidity measure among different stocks rather than on the commonalities between the different liquidity measures themselves. In contrast to these outlined studies, in this paper we construct one single index of (il)liquidity which may serve as a trend indicator of the liquidity of a market. We propose an algorithmic methodology for a continuous calculation method (as opposed to ex-post) which takes into account the findings of the previously mentioned research and identifies the common hidden factors. Among widely used measures based on asset price spreads and trading information we also include aggregated market information (in the form of credit spreads, the LIBOR-OIS spread, and the VIX). We use micro-structural market data of the 500 companies of the S&P 500 for our analysis. We combine the information comprised in these measures to achieve one comprehensive indicator, the illiquidity index ILLIX, which nicely explains the evolution of illiquidity in the North American market over the past 12 years. To the best of our knowledge, no research so far has included data gathered from the time of the financial crisis and beyond.
Measures of liquidity
Studying liquidity bears one major difficulty: while we can easily obtain prices for stocks, interest rates, all sorts of derivatives, etc., liquidity is unobservable. Thus, we first have to ask the question: how can we measure liquidity? As no definition in terms of a formula for liquidity exists, we do not have a one and only measure of liquidity. In this article, we present eight measures which have been proposed to explain the latent factor of liquidity and which are partially based on micro-structural market data. The 500 companies of the S&P 500 index contained in the index on September 15, 2009 form the data basis for the calculation of the first five measures which are based on single stock data. We calculate the five measures for every company in order to obtain 500 time series per measure. To achieve one time series for each measure representing the North American market in total, we average all 500 time series of this measure for the single stocks, i.e., all stocks are assigned equal weight. The last three measures are indices already related to the entire market, for which we get data directly from providers such as Bloomberg.
Bid-ask spread
As described above, liquidity refers to the ability of an investor to buy or sell his security as close as possible to its actual value. If the market is illiquid, the bid and ask spread widens, resulting in costs for the

investor. The bid-ask spread is calculated in relation to the mid-price. The mid-price is the average of bid and ask prices and is used to normalize the spread in order to quote the spread in relative terms. The ask price reflects a premium for immediate buying, and the bid price similarly includes a concession required for immediate sale. Hence, the spread of bid and ask prices is a natural measure of illiquidity [Amihud and Mendelson (1986)]. The resulting bid-ask spread for the entire examined market, namely the average over the 500 bid-ask spreads of the single S&P 500 stocks, shows a declining trend over the last 15 years (1995 2009). There are several reasons for this, such as the globalization of financial markets, increased competition among investors resulting in greater trading volume, and improved price transparency worldwide [Vandelanoite (2002)]. In the following analysis, we are not interested in the longterm trend caused by exogenous factors such as globalization, but in the development of liquidity. Hence, we eliminate this decrease by an exponential trend line. We choose an exponential function representing an annual decrease which is constant in relative terms and prevents the trend from achieving negative values. This technique is described in the appendix. As a result, we achieve the time series shown in Figure 1.
40
30
20 Crisis 10 9/11 0 Crisis
-10
-20 1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 1 Bid-ask spread (trend-adjusted)
3 Dot com bubble 2
Examining Figure 1, the most striking peak, of course, is in late 2008; the trend-adjusted bid-ask spread exhibits a peculiar jump after the Lehman bankruptcy in September 2008. But already at the beginning of the financial crisis in the second half of 2007, it had increased significantly. The Asian crisis in 1997, the Russian crisis in 1998, and the terror attacks in September 2001 can also be clearly made out in the graph (see the labeled humps in Figure 1). In contrast to that, the burst of the dot.com bubble is not reflected as we would expect. The bid-ask spread shows a decline between November 2000 and June 2001. Yet, stock prices were generally falling after the dot.com bubble had burst. Thus, in that time period, the bid-ask spread shows a development which we would not have expected in general.
-3 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 -2 0
-1 Financial crisis
Figure 2 Trading volume (trend-adjusted)
15 Dot com bubble
10
Trading volume and market capitalization

The next two measures, trading volume and market capitalization, have a lot in common; both quantify liquidity in the sense that their values rise when liquidity increases. They represent trading intensity. Kyle (1985) has shown that the intensity of trade by informed investors is positively related to the liquidity of the market, and that the latter in turn depends on the volume of trades performed by noise traders. Generally, over the last 15 years, both the number of stocks and the trading volume rose. However, while trading volume and market capitalization appear to show an increasing exponential trend across the board, they actually only reveal the enormous growth due to globalization of the trading business rather than any accompanying increase in liquidity. Thus, we again adjust these time series for an exponential trend (Figures 2 and 3 present these two measures after eliminating the exponential trends). Figure 2 and 3 demonstrate an increase in liquidity during the dot.com bubble up until 2000 and a decrease
-5 Financial crisis -10
-15 1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 3 Market capitalization (trend-adjusted)
185
after the bubble burst, in contrast to the bid-ask spread. The second peak of the trading volume in October 2000 results from the bailouts after the crash and did not represent an improvement in the liquidity situation itself at this time. Both measures decrease due to the bursting of the dot.com bubble and slowly recover from roughly 2003 until 2007. They both fall rapidly during the financial crisis in 2008. Unlike the market capitalization and the bid-ask spread, the trading volume does not yet indicate a recovery from the stressed situation in late 2009. Furthermore, both trading volume and market capitalization do not yet reflect either the Asian or the Russian crisis, as the bid-ask spread clearly does.
the Lehman bankruptcy and exhibits a final peak in March 2009 when stock prices collapsed worldwide. During this period, stocks displayed extremely negative returns, which resulted in a high ILLIQ.
Open interest
The last measure based on the data of the single S&P 500 stocks is the open interest, which we calculated as the average number of outstanding put and call options for each of the 500 companies. Similar to the manner in which the trading volume provides information about the liquidity situation of a stock, the open interest describes the liquidity of an option, which again is related to the liquidity of a stock [DraKoln (2008)]. In contrast to the previous measures (which we directly calculated on a monthly basis), we use the daily average over the 500 companies from which we first had eliminated the exponential drift. The result is displayed in Figure 5. At first sight, it becomes clear that the shape is different from all the
ILLIQ
Amihud (2000) constructed a more elaborate measure for quantifying illiquidity in his paper, which he called ILLIQ. In contrast to the measures presented so far, the ILLIQ is not that immediately observable in the market. The following formula describes the ILLIQ for stock i in month t.
ILLIQ i ,t =
1 Dt
V
d =1
Dt
d i ,t d i ,t
other measures. The reason lies in the nature of the open interest: The (1)
12
This measure contains the absolute daily returns r divided by the daily turnover V, which is defined as the ratio between trading volume and market capitalization. The ILLIQ calculates the average for one month t with number of days Dt. In other words, the ILLIQ can be interpreted as the price response associated with the daily turnover, thus serving as a measure of price impact. Hence, the ILLIQ yields high (absolute) values when returns are either very negative, or very positive, as long as the turnover is small; high turnover divides the ILLIQ into small values. After computing the ILLIQ time series for all 500 companies, we average over all the companies, as already described, for the bid-ask spread, trading volume, and market capitalization to get the ILLIQ for the entire market. The resulting time series is shown in Figure 4. As the ILLIQ, and the factors used for calculating this measure, do not show any trend which can be explained economically, we do not adjust the ILLIQ for an exponential trend, as opposed to the measures introduced earlier. In Figure 4, we can make out the Asian and the Russian crises in 1997 and 1998, respectively. Also the 9/11 terror attacks are represented by a peak at that time. In general, we can observe a slight increase until January 2000 and a decrease in the trend from this time until 2004. This evolution of the measure is in contradiction to the situation in the market during the dot.com bubble and after it burst, as we would assume an increase in illiquidity after the bubble burst and not before. Yet, the high values of the ILLIQ during the dot.com boom result from the high (positive) returns. During the financial crisis, the time series is more explicable by the market situation. The ILLIQ rises with the acquisition of Bear Stearns 186 by JP Morgan Chase. Finally, the ILLIQ assumes its maximum value after
-150 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 -100 -50 0 100 open interest (trend-adjusted) season pattern
2 4 10 Lehman
8 Russian crisis Asian crisis 9/11 Recession
0 1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 4 ILLIQ
50
Figure 5 Daily open interest (exponentially trend-adjusted and season pattern)

40
the Lehman bankruptcy. Yet, the open interest does not, for example, account for the turmoil within the markets after 9/11.
20
Credit spread, LIBOR-OIS spread, and VIX

0
The last three of the eight measures we present are measures which themselves are indices: credit spread, LIBOR-OIS spread, and VIX. The
Burst of dot com bubble Bear Stearns
-20 Russian crisis
credit spread is computed as the difference between the five-year corporate bond yields (U.S. Industrial BBB) and five-year treasury bond yields (U.S. Treasury Strips). It illustrates the risk premium between a highly secure investment (in the form of treasuries) and a less secure investment (in the form of BBB-rated corporate bonds). The reason why a spread exists is that a company must offer a higher return on their bonds because their credit is worse than that of the government. The higher the return the higher the spread, and therefore the higher the illiquidity of an investment in this company [Huang and Huang (2003)]. Figure 7 exhibits a similar pattern to the ILLIQ, also showing the impact of the Russian
-40
Lehman -60
-80 1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 6 Open interest (exponentially trend-adjusted and de-seasonalized)
5 Lehman
crisis and the loss in liquidity during the financial crisis at its peak after the Lehman bankruptcy. Similar to the ILLIQ, the credit spread does not indicate the dot.com bubble and its collapse as we would expect from an illiquidity measure.
3 Bear Stearns 2
The LIBOR-OIS spread is the difference between the three months LIBOR and the three months Overnight Indexed Swap, and therefore describes a risk premium from the banks point of view [Hui et al. (2009)]. According to the theory of Adrian and Shin (2010), banks avoid investing in the interbank market if asset prices fall during a crisis, reducing market liquidity. Thus, a higher risk premium is claimed for longer-term loans. As banks avoid these loans, the spread between the LIBOR and the OIS, which usually ranges just below the LIBOR, rises. Data for the LIBOR-OIS spread was not available before December 2001, which does not allow us to compare the effects of the recent financial crisis with any other crisis. In the first few years, the index remains very stable at about 10 basis points. In the middle of 2007, the index jumps to a level of 85 after Northern Rock announced rescue financing agreements, shows another hump when Bear Stearns was finally taken over, and rose to its highest point when Lehman became insolvent. Naturally, the LIBOR-OIS spread is highly sensitive to events directly related to the banking sector. As we eventually want to calculate the illiquidity index ILLIX from 1998 onward, we extrapolate the LIBOR-OIS spread by regressing the other measures (except for the open interest) for the years 2002 and 2003 to the LIBOR-OIS over the same time span, and apply these weights to the time series of the six measures to gain a generic LIBOR-OIS spread between 1998 and 2001. The extrapolated time series is indicated by the dotted line in Figure 8 which is in line with the interpretation given by Adrian and Shin (2010), although the OIS had not yet been reported at that time. 187
Russian crisis
0 1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 7 Credit spread
conspicuous zig-zag-pattern shown in Figure 5 occurs every month, and once a year on a greater scale. This is due to the maturity of the options, which mostly end on every third Friday of a month, and the fact that most options mature at the end of a year. If we want to draw conclusions on the liquidity of the markets from the open interest we have to eliminate this intrinsic behavior of the open interest. We apply an exponentially stretched seasonal pattern in order to de-seasonalize the time series. This season pattern is displayed in Figure 5 by the dark line. The technique is described in detail in the appendix. By computing the average over all days of one month, we get the final (monthly) open interest time series which is displayed in Figure 6. Due to the fact that open interest is a liquidity measure, like the trading volume and the market capitalization, the decline during the Russian crisis in 1998 demonstrates the loss of liquidity in the market. Furthermore, the dot.com bubble and its bursting are indicated by this measure in the middle of 2000. The open interest clearly shows the takeover of Bear Stearns and the collapse of Lehman Brothers with a sharp decline after
350
These eight measures give us a pool of liquidity information, where each measure expresses different aspects of liquidity. Yet, none of them is the one and only quantification of liquidity. As we observed from the figures, some measures explain certain developments in the financial markets better than others, and sometimes a single measure would not give us any indication of a certain incident. However, by simply looking at the figures of the measures, even at this point we can observe similar
Northern Bear Rock Stearns
300 Lehman 250
200
150
patterns in some market phases among several measures. This indicates already that some of these measures are driven by the same underlying intrinsic factors, as previously stated by Chordia et al. (2000) and Korajczyk and Sadka (2008). The question arises, how much information on liquidity is comprised in each of these measures, and how many measures do we need to capture all this information sufficiently?
100
50
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 8 LIBOR-OIS spread
The ILLIX a new illiquidity index

70
Based on the information comprised in these eight measures our aim is to

Lehman
60
generate an index that reflects the illiquidity situation in the North American market over time. Thus, we try to eliminate measures with the same information to keep the index structure as least complex as possible.
50 Russian crisis 9/11 30
40
Recession
That is why we test the measures for commonalities and arrange them according to their relevance. Hence, we implement an iterative algorithm to select the most important measures every month and combine them into one indicator which we call ILLIX, the ILLIquidity indeX. As six of the eight measures are available from January 1998 onwards, we start the calculation with the time series of the eight measures between 19981 and 2003 as the initialization. However, we do not want to use data which
20
10
0 1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
was not available at the point (date) of calculation. The aim is to create an index whose composition is determined ex-ante, such that it allows for continuous calculation. This is why we then, after the initialization, expand the time series month by month in an iterative process (guaranteeing measurability in the context of financial mathematics) and continue
Figure 9 VIX
The last measure is the Volatility Index (VIX) which represents the expected future market volatility over the next 30 calendar days, implied by the current prices of the S&P 500 index options. It has been dubbed the investor fear gauge [Whaley (2008)] and is therefore also used as a measure of illiquidity. At first glance, its shape seems similar to the ILLIQ or the credit spread. Especially between 1997 and 2003 and also during the financial crisis, the ILLIQ and the VIX exhibit similar patterns: if the volatility increases then the risk of price jumps rises too. Hence, the absolute returns and therefore the ILLIQ go up. But the VIX shows some more peculiarities other than the ILLIQ between 1997 and 2003, for example during the Russian crisis. Moreover, we can see the effects of the terrorist attack in September 2001. As the VIX quantifies the implied volatility, it mainly contains the investors expectations, which react very sensitively to the situation in the markets. Subsequent to 9/11, investor fear increased and that is why the implied volatility rises. Subsequently, there were turbulences which demonstrate the instable situation after the 188 dot.com crash.
until the last available month, which is August 2009 for this study.
Generating the ILLIX

For the generation of the index, we apply the technique of the principal component analysis (PCA). With the help of this tool we obtain the main parts of all the information available to us via the first principal component. The PCA allows us to transfer a number of correlated variables into an orthogonal space of uncorrelated variables. These uncorrelated variables are called principal components (PC), which are ordered by the explanatory power, i.e., the first PC accounts for as much of the variation
For this reason, the LIBOR-OIS spread has been extrapolated back tp 1998, as described above. For the same reason, the open interest is extrapolated for the first nine months of 1998 with the same technique as described for the LIBOR-OIS spread based on the first two available years (October 1998 until September 2000). However, due to seasonal patterns we do this with the generated monthly open interest time series, i.e., after the adjustments.

in the data as possible in one dimension [For more details about the PCA see, for example, Alexander (2008a, b), and Jolliffe (2002)]. We use this first PC as an approximation of the intrinsic factor of illiquidity which drives all our measures, yet is not directly observable. For every month, i.e., for every point in time we update the index, we proceeded as follows: we perform a PCA on the eight trend-adjusted (and de-seasonalized) measures at the specified point of calculation. As an assumption of the principal component analysis, the measures are standardized from now on. This means that the mean of every measure is 0 and the variance is 1. Applying the PCA, the first PC gives us the main part of all the information available to us. As already discussed, some of the measures may be driven by the same information, so that some measures may be redundant in explaining the first PC. In order to select only the relevant measures with additional information, we use the Variance Inflation Factor (VIF) [Longnecker and Ott (2004)]. By avoiding redundant measures we get a less complex structure of the resulting ILLIX, making it more stable. The VIF is a measure of multicollinearity and is defined as the inverse of 1 minus the R-squared which we get from a regression of one measure against all the others. This is why the higher the VIF, the higher the multicollinearity of one measure to the rest. In our algorithm, we used a VIF of five as a cut-off value which is a commonly used barrier [Menard (1995)], i.e., measures with a higher VIF were eliminated for that month, as their contribution is not significant. A VIF higher than five means that the R-squared is 0.80 (that is 80 percent of the variation is explained by the linear model i.e., the other measures, and 20 percent is noise or in this case, additional information which has not been accounted for yet) and shows us that the measure can be well explained by other measures and gives us consequently no significant new information. In this case, we can eliminate the measure more or less without losing any information which is not already comprised in the other factors. At the same time, we set an additional minimum of four and a maximum of six measures which have to be included. The minimum and maximum constraints are of course heuristic. Yet, they have been proven to provide the most stable results during our study. Furthermore, those barrier conditions are already satisfied by the VIF condition in most cases (the lower boundary is not broken at all in the proposed setting). The actual selection of the measures works as follows: we calculate the VIF for each of the remaining measures (i.e., all eight measures in the first step). As long as the number of measures is higher than six we remove the measure with the highest VIF (even if the highest VIF is lower than five). In the next step, we calculate the VIF values for the remaining measures again (because they are now regressed to a different set which contains one measure less). We further eliminate the measure with the highest VIF, if this VIF is higher than five (or we have more than six measures) until all measures have a VIF lower than five (or we only have four measures left).
2
14 Lehman 12 10 8 6 4 2 0 -2 -4 1998 9/11 Recession Russian crisis Dot com bubble Northern Bear Rock Stearns
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Figure 10 The illiquidity index ILLIX over time
Having selected all those measures containing significant information, we regress them against the first principal component to obtain the weights of the ILLIX.2 The ILLIX arises from the selected measures multiplied by these weights. As we are describing illiquidity, naturally, those measures quantifying liquidity usually have a weight with a negative sign and the measures quantifying illiquidity in general have a positive weight. In the previous paragraphs, we explained the procedure of the algorithm for calculating the value of the ILLIX for one month. For the next iterative calculation step to determine the ILLIX value for the next month, the algorithm adds the values of the eight time series for the following month and continues with the procedure as described above. By means of the stepwise calculation we provide an index which uses all the market information up to the point of calculation, yet no data which would not have been available at that point (as would have been the case for an ex-post index calculation, as the weights for the first step would also depend on the entire input time series). The final ILLIX is displayed in Figure 10.
Describing the liquidity situation with the ILLIX

Analyzing Figure 10 and the weights in Table A1 (in the Appendix), we first have to remember that the index was calculated until the end of 2003 ex-post (in order to have a minimum sample size of six years of data for the calculation for reasons of stability), and then updated with the most current data every month from January 2004 on. Table A1 summarizes the composition of the index in each time step from December 2003 until August 2009. It gives the weights of the measures which are selected by the algorithm. If a measure is not selected, this is indicated by a weight of 0 in the table. Due to the fact that the index was
The new regression is necessary due to the reduced number of factors. Of course, if all the factors were used, the weights would already be given by the first PC.
189
calculated ex-post for the six years until December 2003, there is only one set of weights for this time. From then on, the index is calculated each month with the then most current information being reflected in the adjustment of weights. There are some interesting conclusions which can be drawn from the table: for example, the ILLIQ is never selected for the index calculation. Yet the VIX is always contained in the index. This fact again emphasizes the similar characteristics and behavior of these two measures and that the algorithm avoids redundancies via the VIF.3 Furthermore, the trading volume and the market capitalization are similar in many ways, and consequently only one of the two is integrated in the calculation of the ILLIX at the same time. Moreover, the table shows a detail which we would not expect at first sight, namely that although we have identified the trading volume and the market capitalization as measures of liquidity, they have a positive weight in the illiquidity index until October 2008. When looking at Figures 2 and 3, the trading volume and market capitalization show a slight increase between 2003 until 2008, for example, and strong turbulences after this period. This pattern is similar to that for the illiquidity measures bid-ask spread, VIX, credit spread, and LIBOR-OIS spread which are contained in the index over the same time. Consequently, the trading volume shows the same characteristics as the illiquidity measures and is positively weighted. The positive weight can also be regarded as a means of balancing the high weight of the illiquidity measures. But the positive weight decreases towards October 2008 and finally becomes negative again in November 2008. Analyzing the performance of the index in Figure 10, we get a nice description of the evolution of illiquidity. In 1998, the index jumps from about -2 to 3 points during the Russian crisis, marking the first little hump in the shape of the index, when illiquidity on the market increased. After the Russian crisis, in general, it shows a declining trend when markets were booming, above all, in the technology sector. With the dot.com bubble the index rises in early 2000 and depicts the falling stock prices with a second hump in that year. Here, the index demonstrates this turbulent time in contrast to the bid-ask spread which has hit a low. Furthermore, we can see the aftermaths of 9/11 by a peak in September 2001. Finally, it achieves a high of 0.3 points in September 2002 due to the ongoing recession, similar to the bid-ask spread, trading volume, and VIX at this time. In 2003, the index returns to a lower level, reflecting the recovering and again booming economy. From July 2005 until the beginning of the current ongoing financial crisis, the index remains very stable. After the beginning of the financial crisis, with the turmoil surrounding Northern Rock in 2007, the index increases very fast. It also indicates when Bear Stearns was taken over, and then hits its maximum after the Lehman bankruptcy in September 2008. The Emergency Economic Stabilization Act of the U.S. government resulted in a fast decline in contrast to the 190 trading volume which can be observed from October 2008 onwards. A
comparison of the current financial crisis to the recession of the beginning of the century clearly depicts the different nature of those two recessions and that liquidity has been a great issue during the financial crisis. All in all, the ILLIX reflects the illiquidity situation in the North American market very well.
Conclusion
This paper introduces a methodology for making liquidity a computable factor. By generating the index ILLIX, we introduce a method which turns the latent factor illiquidity into a measurable indicator. The ILLIX is based on eight measures, whose trends and seasonal patterns were eliminated in a first step, if necessary. These eight measures include micro-structural data of single stocks such as price spreads or trading information as well as data on a market level such as the volatility index. In the second step we applied the principal component analysis in order to find the first PC representing the main driver of all measures. This accounts for the findings of previous research which identified commonality among all liquidity factors. The most important measures were selected via the VIF to explain the first PC. The regression coefficients gave us the weights for the resulting ILLIX. Whereas previous studies are all based on expost analyses, this method allows us to construct an index which can be calculated in each time step with only the information on the market available at that time. The aim of the ILLIX is to provide a trend indicator for illiquidity. We saw that none of the single factors alone can sufficiently reflect the liquidity in the market. The ILLIQ, for example, rose during the dot.com bubble, hinting at high illiquidity, and decreased during its collapse in contrast to what we would expect for an indicator of illiquidity. The trading volume, for example, nicely exhibits the dot.com bubble and its collapse, but has shown no indication of recovery from the financial crisis yet. The ILLIX combines and filters all the information and provides an indicator whose developments can explain liquid and illiquid market situations.
Note that this does not mean that the ILLIQ would not be of importance. The information is already comprised in the other measures to a large extent and accounted for in the actual weights of the other measures. Excluding the ILLIQ from the calculation would lead to a different ILLIX.

References

Method of exponentially stretched seasonal pattern

Before dealing with the seasonal pattern, we eliminate the general exponential trend as described above and gain the trend-adjusted time series x. For the open interest, this (daily) time series is visualized in Figure 5 as the exponentially trend-adjusted open interest. We then account for the seasonal pattern. In the first step, we apply the flat seasonal pattern [for more details see Schlittgen and Streitberg (2001)]. For this we define dummy variables sm(t) for each month m = 1, 2, , 12 and day t:
1 sm(t) = 0 if day t belongs to month m else
Adrian, T. and H. S. Shin, 2010, Liquidity and leverage, Journal of Financial Intermediation, 19, 418-437 Alexander, C., 2008a, Market risk analysis 1: quantitative methods In finance, John Wiley & Sons, New York Alexander, C., 2008b, Market risk analysis 2: practical financial econometrics, John Wiley & Sons, New York Amihud, Y., 2000, Illiquidity and stock returns: cross-section and time-series effects, Journal of Financial Markets, 5, 31-56 Amihud, Y. and H. Mendelson, 1986, Asset pricing and the bid-ask spread, Journal of Financial Economics, 17, 223-249 Brockwell, P. J. and R. A. Davis, 1991, Time series: theory and methods, Springer, New York Chacko, G. and E. Stafford, 2004, On the cost of capital market transactions, working paper, Harvard Business School Chordia, T., R. Roll., and A. Subrahmanyam, 2000, Commonality in liquidity, Journal of Financial Economics, 56, 3-28 DraKoln, N., 2008, Winning the trading game: why 95% of traders lose and what you must do to win, Wiley & Sons, Hoboken, NJ Hasbrouck, J. and D. J. Seppi, 2001, Common factors in prices, order flows, and liqudity, Journal of Financial Economics, 59, 383-411 Huang, J. and M. Huang, 2003, How much of the corporate-treasury yield spread is due to credit risk? Working Paper, NYU Hui, C., H. Genberg, and T. Chung, 2009, Liquidity, risk appetite and exchange rate movements during the financial crisis Of 2007-2009, Hong Kong Monetary Authority Working Paper Jolliffe, I. T., 2002, Principal component analysis, Springer, New York Korajczyk, R. and R. Sadka, 2008, Pricing the commonality across alternative measures of liquidity, Journal of Financial Economics, 87, 45-72 Kyle, A., 1985, Continuous auctions and insider trading, Econometrica, 53, 1315-1335 Longnecker, M. T. and R. L. Ott, 2004, A first course in statistical methods, Thomson Brooks/ Cole, Belmont, CA Menard, S., 1995, Applied logistic regression analysis, Sage University Series on Quantitative Applications in the Social Sciences, Sage Thousand Oaks Schlittgen, R. and B. H. J. Streitberg, 2001, Zeitreihenanalyse, Oldenbourg, Mnchen Vandelanoite, S., 2002, Takeover announcements and the components of the bid-ask spread, Working Paper, University of Paris Whaley, R.W., 2008, Understanding VIX, Working Paper, Vanderbilt University
Based on these dummy variables, in the first step, we define the flat season pattern as follows:
s(t)= msm(t)
m =1
12
For estimating the 12 coefficients bm we use the least square method:
min (x( t) s( t))

b t
The flat seasonal pattern indirectly assumes the same jumps in every January, every February, and so on. However, for the open interest, these jumps also seem to increase exponentially over the years. Thus, we then calculate the exponentially stretched seasonal pattern. This means that because the number of outstanding options increases over time, the number of options expiring at maturity also increases exponentially.
Appendix
Exponential trend elimination
For eliminating the exponential trend of the bid-ask spread, trading volume, market capitalization, and open interest, we calculate exponential trend lines and subtract them from the original time series. Hence, we estimate the coefficients a and b of the exponential function, which are real numbers, by the least square method:
The exponentially stretched seasonal pattern is gained by multiplying an exponential term with the flat seasonal pattern, which resulted from a least square regression:
min (x ( t) s( t) a exp (b t))

a,b t
This exponentially stretched seasonal pattern in also displayed in Figure5. Deducting the seasoaln pattern from the original time series and taking the average over all days in one month yields the monthly time series which is shown in Figure 6.
min (y( t) a exp(b t))

a ,b t
where y(t) is the value of the measure at time t. We then subtract the exponential trend with the estimated parameters a and b from the original data y in every point in time t.
191
bid-ask spread until Dec 03 Jan 04 Feb 04 Mar 04 Apr 04 May 04 Jun 04 Jul 04 Aug 04 Sep 04 Oct 04 Nov 04 Dec 04 Jan 05 Feb 05 Mar 05 Apr 05 May 05 Jun 05 Jul 05 Aug 05 Sep 05 Oct 05 Nov 05 Dec 05 Jan 06 Feb 06 Mar 06 Apr 06 May 06 Jun 06 Jul 06 Aug 06 Sep 06 Oct 06 0.4583 0.4405 0.4332 0.4235 0.4111 0.4025 0.3970 0.3876 0.3890 0.3829 0.3738 0.3653 0.3569 0.3478 0.3442 0.3379 0.3321 0.3269 0.3225 0.3201 0.3182 0.3141 0.3091 0.3051 0.3011 0.2964 0.4583 0.4573 0.4552 0.4572 0.4569 0.4569 0.4541 0.4517 0.4491
trading volume 1.1515 1.1522 1.1501 1.1482 1.1475 1.1428 1.1391 1.1357 1.1283 1.1246 1.1252 1.1238 1.1248 1.1199 1.1131 1.1060 1.0998 1.0953 1.0890 1.0858 1.0798 1.0748 1.0710 1.0688 1.0664 1.0578 0 0 0 0 0 0 0 0 0
market capitalization 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1.0051 1.0036 1.0021 1.0012 1.0020 1.0004 0.9992 0.9970 0.9952
ILLIQ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
open interest -0.1904 -0.1833 -0.1710 -0.1667 -0.1508 -0.1511 -0.1456 -0.1459 -0.1419 -0.1369 -0.1255 -0.1146 -0.0981 -0.1011 -0.0879 -0.0830 -0.0751 -0.0766 -0.0724 -0.0781 -0.0738 -0.0666 -0.0575 -0.0488 -0.0356 -0.0401 -0.0414 -0.0500 -0.0513 -0.0567 -0.0598 -0.0622 -0.0642 -0.0581 -0.0481
credit spread -0.1176 -0.0902 -0.0695 -0.0542 -0.0330 -0.0229 -0.0149 -0.0082 -0.0038 0.0001 0.0037 0.0082 0.0123 0.0183 0.0285 0.0384 0.0449 0.0473 0.0511 0.0549 0.0571 0.0581 0.0600 0.0600 0.0615 0.0629 0.3534 0.3573 0.3609 0.3622 0.3635 0.3637 0.3641 0.3626 0.3635
LIBOR-OIS spread 0.5306 0.5366 0.5379 0.5427 0.5467 0.5517 0.5528 0.5560 0.5561 0.5581 0.5615 0.5642 0.5652 0.5694 0.5718 0.5732 0.5794 0.5832 0.5877 0.5863 0.5901 0.5957 0.6021 0.6049 0.6073 0.6167 0.3421 0.3359 0.3327 0.3241 0.3194 0.3186 0.3184 0.3194 0.3201
VIX 0.3108 0.3371 0.3546 0.3666 0.3849 0.3943 0.4036 0.4161 0.4217 0.4309 0.4420 0.4543 0.4670 0.4769 0.4840 0.4940 0.4987 0.5034 0.5086 0.5113 0.5152 0.5204 0.5235 0.5278 0.5325 0.5376 0.5918 0.5930 0.5953 0.5992 0.6020 0.6036 0.6062 0.6106 0.6136
192

bid-ask spread Nov 06 Dec 06 Jan 07 Feb 07 Mar 07 Apr 07 May 07 Jun 07 Jul 07 Aug 07 Sep 07 Oct 07 Nov 07 Dec 07 Jan 08 Feb 08 Mar 08 Apr 08 May 08 Jun 08 Jul 08 Aug 08 Sep 08 Oct 08 Nov 08 Dec 08 Jan 09 Feb 09 Mar 09 Apr 09 May 09 Jun 09 Jul 09 Aug 09 0.4461 0.4418 0.4400 0.4365 0.4360 0.4333 0.4301 0.4270 0.4270 0.4384 0.4581 0.4630 0.4678 0.4731 0.4839 0.4896 0.5012 0.5079 0.5148 0.5233 0.5337 0.5395 0.5620 0.5175 0.6853 0.6942 0.6843 0.6648 0.6518 0.6439 0.6399 0.6384 0.4536 0.4561
trading volume 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -0.3323 -0.3798 -0.3945 -0.4063 -0.4140 -0.4178 -0.4174 -0.3080 -0.3100
market capitalization 0.9930 0.9899 0.9868 0.9846 0.9830 0.9807 0.9783 0.9780 0.9775 0.9581 0.9865 0.9918 0.9866 0.9858 0.9713 0.9612 0.9435 0.9313 0.9194 0.8963 0.8582 0.8093 0.6316 0.2172 -0.2003 0 0 0 0 0 0 0 0 0
ILLIQ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
open interest -0.0373 -0.0268 -0.0291 -0.0392 -0.0449 -0.0467 -0.0491 -0.0506 -0.0495 -0.0389 -0.0394 -0.0384 -0.0380 -0.0333 -0.0623 -0.1081 -0.1394 -0.1702 -0.1926 -0.2172 -0.2409 -0.2710 -0.3233 -0.3906 -0.3963 -0.3546 -0.3452 -0.3679 -0.3785 -0.3829 -0.3836 -0.3797 -0.4561 -0.4553
credit spread 0.3642 0.3658 0.3665 0.3707 0.3714 0.3699 0.3689 0.3687 0.3691 0.3732 0.3895 0.3933 0.3969 0.4000 0.4076 0.4102 0.4114 0.4124 0.4097 0.4183 0.4295 0.4432 0.4631 0.4564 0.6722 0.7327 0.7352 0.7250 0.7236 0.7209 0.7170 0.7117 0.5342 0.5326
LIBOR-OIS spread 0.3205 0.3206 0.3211 0.3202 0.3206 0.3230 0.3259 0.3277 0.3276 0.3332 0.2086 0.1769 0.1675 0.1365 0.1411 0.1405 0.1441 0.1407 0.1389 0.1460 0.1646 0.1912 0.3025 0.4707 0 0 0 0 0 0 0 0 0.3689 0.3676
VIX 0.6176 0.6225 0.6254 0.6267 0.6271 0.6297 0.6319 0.6331 0.6346 0.6356 0.6593 0.6619 0.6674 0.6712 0.6754 0.6786 0.6849 0.6848 0.6850 0.6853 0.6881 0.6827 0.6758 0.6467 0.6054 0.6198 0.6203 0.6212 0.6221 0.6247 0.6287 0.6357 0.6177 0.6194
Table A1 Weights of the different measures in the ILLIX over time
193
194
PART 2
How Homogeneous Diversification in Balanced Investment Funds Affects Portfolio and Systemic Risk
1
Rocco Ciciretti SEFeMEQ Department, University of Roma Tor Vergata, and EPRU, University of Leicester School of Management Raffaele Corvino Collegio Carlo Alberto, University of Torino
Abstract
The recent financial crisis highlighted the dangers of systemic risk. In this regard no common view appears to exist on the definition, measurement, and real impact of systemic risk on the financial system. This paper aims to analyze the relationship between systemic risk and portfolio diversification, highlighting the differences between heterogeneous and homogeneous diversification. Diversification is generally accepted to be the main tool for reducing idiosyncratic or portfolio-specific financial risk, however, homogeneous diversification also has implications for systemic risk. Using balanced investment funds data, the empirical analysis first investigates how diversification affects the two components of individual portfolio risk, namely systematic, and
idiosyncratic risk, and then uses an estimation procedure to examine the change in asset allocation and its impact on global systemic risk. The results suggest that funds portfolio diversification reduces at the same time the portfoliospecific risk and increasing the likelihood of a simultaneous collapse of financial institutions given that a systemic event occurs.
Earlier version of this paper is SSRN-CEIS Research Paper 204. The authors thank all the participants of the 11th International C.R.E.D.I.T. Conference (Venice), the XX International Tor Vergata Conference on Money, Banking, and Finance (Rome), the 2012 ASSONEBB Conference on Sovereign Risk (Rome), and the XIII Workshop on Quantitative Finance (LAquila) for useful comments. All remaining errors belong to the authors. Usual disclaimer applies.
195
Introduction
Every happy family is the same. Every unhappy family is miserable in its own way [Leo Nikolayevich Tolstoy (1877), in Summers (2000)]. The recent financial crisis highlighted the dangers of systemic risk, and led to academicians, as well as financial executives, to consider its implications on the functioning of the financial market. The debate on the definition of systemic risk as well as on the sources of the last turmoil is still open. For example, Schwarz (2008), while discussing systemic risk, states that if a problem cannot be defined it cannot be solved, and Tirole (2002) argues that two crises are never identical and each one shows own distinctive elements. Given that the last crisis may be considered as an example of systemic crisis, our research investigates a potential root of systemic risk, namely the degree of homogeneity among market agents as consequence of their portfolio diversification strategies. The common thread among definitions of systemic risk in the literature is that it has an adverse effect on the stability of the financial system [Brownlees and Engle (2010); De Bandt and Hartmann (2010); Lehar (2004); De Nicolo and Kwast (2002)]. Hence, if the agents are homogeneous the likelihood that a systemic event will affects them all in the same way increases. Thus, portfolio diversification, usually considered as one of the most important tools for mitigating risk and implemented by financial investors to reduce portfolio risk, may increase the likelihood of a systemic crisis.2 The aim of this paper is to examine these two sides of the diversification process by analyzing the impact of diversification on different types of financial risk. More precisely, we investigate over the past 10 years how diversification has impacted portfolio and systemic risk. The former may be divided into two components: i) systematic risk, which stems from the sensitivity of portfolio returns to market returns, and is usually measured through the portfolio b factor the correlation between the portfolio and market returns; and ii) idiosyncratic risk, which depends on the specific portfolio factors and is the portion of portfolio risk not explained by market factors. In the financial literature, systematic risk is considered non-diversifiable while the latter may be reduced through an adequate portfolio diversification strategy which neutralizes the risk-components related to portfolio-specific factors [Goetzmann and Kumar (2008); Fama and MacBeth (1973)]. Consequently, if the portfolio idiosyncratic component is reduced, the level of mutual homogeneity among market agents increases, making them vulnerable to a simultaneous collapse when a negative systemic event occurs. Thus, starting from different conditions and expectations, market agents become homogeneous because of their portfolio diversification strategies, increasing the level of systemic risk in the financial system. 196
Our investigation proceeds along three consecutive steps. Firstly, portfolio systematic risk, or the beta factor, is estimated and analyzed. Secondly, the relationship between the idiosyncratic portfolio risk and portfolio diversification is investigated. Finally, the impact of portfolio diversification and homogeneity level of the financial system on the likelihood of a simultaneous downturn is assessed.
Related literature
The debate about systemic risk is recent and the related literature is still limited. Moreover, there is not a common view on the definition of systemic risk. Many strands of research into the threats caused by systemic risk on the economic system have been developed. Nevertheless, let us briefly review some of the available ones below. Four approaches are: (1) risk that an event affects a large number of financial institutions and markets at the same moment, (2) a domino-effect that occurs through common exposures of financial institution to a certain asset, (3) a banking default or a broader market participants default as key factors, and (4) a negative externality involving real effects. In the first approach, systemic risk may be thought of as the likelihood that a trigger event, such as an economic or financial shock, may have significant adverse implications on a large portion of financial institutions or markets. This strand of literature [Brownlees and Engle (2010); Kupiec and Nickerson (2004); and Dow (2000)), defines systemic risk as the risk of a simultaneous collapse of market agents acting in the financial system. Dow (2000) suggests that systemic risk produces its effects in four different ways: disruption of a payment system due to one or more banks defaults, depression of banking asset values, general fear of losing savings (simultaneous withdrawals from banks), and reduction of national income linked to macroeconomic changes. Kupiec and Nickerson (2004) describe other potential ways of systemic risk impacting on the financial system, such as price volatility, corporate liquidity, and efficiency losses. In the second group, we have Kaufman (1996), De Bandt and Hartmann (2000), Sheldon and Maurer (2008), Schwarcz (2008), who suggest that systemic risk acts as a domino-effect due to linkages between the financial institutions. Kaufman (1996) refers to the cumulative losses caused by an event that ignites successive losses along a chain of financial institutions or markets. De Bandt and Hartmann (2000) relate systemic risk to experiencing of a systemic event. This involves institution Y being severly impacted because of an initial shock that has impacted institution X even if Y was fully solvent at the beginning. This is also supported by Bartram et al. (2007) who show that institutions with good economic fundamentals can also be indirectly affected by systemic risk in a crisis.
It can be defined as a systemic event that affects a considerable number of financial institutions or markets, in a strong sense and severely impairing the general well-functioning of the financial system [De Bandt and Hartmann (2000)].

The domino effect is explicitly defined as the likelihood that a failure of one bank triggers a chain reaction causing other banks distress through interbank loans [Sheldon and Maurer (2008)] and as a trigger event that causes a chain of bad economic consequences [Schwarcz (2008)]. In the third approach, the banking default is the key element for defining systemic risk. Eisenberg and Noe (2001) refer to the number of waves of default needed to cause a firms default in a closed financial system. Lehar (2004) assesses systemic risk as the probability that a certain number of banks within a time period become insolvent due to a fall in the value of their assets below that of their liabilities. This view stems from Mertons (1974) structural models where banks become insolvent, and default occurs, when the value of their assets falls below a given threshold. Considering not only a bankruptcy condition but all market participants default, the Bank for International Settlements [BIS (1994)] defines systemic risk as the risk that a failure of a market participant to meet its contractual obligations may cause other participants to default. Such definition is shared by the U.S. Commodity Futures Trading Commission (2008), which describes systemic risk as the risk that a market participants default impacts other participants due to the interlocking nature of financial markets.
to share their projects with other banks to achieve a lower default probability and lower repayment to creditors. The same process makes up a clustered network in which each bank holds the same portfolio, so that each banks signal is of interest for investors. Wagner (2006) considers an economy with two banks which have to set the optimal level of diversification. Full diversification is undesirable because it reduces the risk at each individual institution but increases the risk of a systemic crisis. The bank has incentives to fully diversify because it externalizes the costs, thus increasing the likelihood of failure in other banks. The level of diversification has to be arbitrarily small, depending on the difference between costs of individual failure and a systemic crisis. Diversification may increase the likelihood of a contagion too, exposing banks to the consequences of the failure of other banks in which the first one diversified its investments. Allenspach and Monnin (2007) test for the hypothesis that there is an empirical link between common exposures to shocks and systemic risk for the period 1993-2006. If all banks choose to diversify, they are all exposed to the same risk factors. Considering a broader notion of systemic risk that includes the contagion
In the last approach, De Nicolo and Kwast (2002) and Kambhu et al. (2007) describe systemic risk as a negative externality, either through the direct linkages given by intermediaries exposures and through a broader disruption directly affecting the financial markets. Such market failure has an impact on cost of capital, producing a reduction in credit provision as well as in real activity. The authors underline the fact that real effects of systemic risk constitute the main threat. However, they distinguish systemic risk from financial crises. In De Nicolo and Kwast (2002), the magnitudes of financial failure have to be so high as to induce real consequences such as reductions in output and employment. In Kambhu et al. (2007), the effect is a reduction of productive investment due to the decreasing credit provision. But in the authors opinion the optimal level of systemic risk is not zero. Few similar studies may be found in the literature about the relationship between diversification, portfolio, and systemic risk. De Vries (2005) argues that diversification reduces the frequency of individual bank failure when a shock is smaller and easily borne by the system, while it increases the likelihood of a systemic failure when a stronger shock occurs. Allen et al. (2010) analyze systemic risk focusing on the banking sector and the interconnections among the banks looking especially to the signals perceived by investors who have to roll over their investments in the same banks. The banks are involved in a network and each banks condition is a signal for the entire banking system. The network is the result of the diversification process of the banks who desire
of financial turmoil across different countries or regions, Schinasi and Smith (2000) relate the diversification between risky and riskless assets, especially looking to the rebalancing of portfolios among these two different classes of securities, with the contagion effect from one region, where the shock occurs, transmitted to the other region. Focusing on the Russian default of 1998, this paper shows that one shock leads the leveraged portfolio to reducte its other risky positions (in other regions, markets, and industries), in according with management rules, thus discovering the implicit and potential danger within portfolio diversification.
The estimation model

The framework proposed below aims to analytically describe the relation between diversification, portfolio risk, and systemic risk through a multi-step analysis that begins from the portfolio return decomposition and explanation. The goal of this model is to show how diversification activities impact different forms of financial risks to capture the net effect of diversification on portfolio and systemic risk. Many papers [Fama (1972); Becker and Hoffmann (2008); Goetzmann and Kumar (2008)] have focused on the consequences of diversification on individual risk-taking without looking at its impact on the entire system. Another strand of literature [Allenspach and Monnin (2007); Allen et al. (2010), Wagner (2006)] attempts to assess how banking diversification affects the risk that the whole banking system will collapse. By contrast, the model below aims (i) to evaluate the impact of diversification on different components of risk for a representative agent, and (ii) to assess its consequences for the entire financial system. 197
Portfolio return and the b factor

Consider an economy with i agents, with i going from 1 to n. Each agent holds a portfolio composed of different asset classes (from here, we identify each agent with his own portfolio. In other words, i identifies at the same time the agent as well as the portfolio). Each portfolio consists of k asset classes (with k going from 1 to m), and each asset class has a weighting of wk within the agents portfolio. Thus, the portfolio is a basket of k-asset classes, and the relative portfolio size adds up to 1 and is described as follows: 1 = PSit = m witk, where PSit is the relative size of k=1 portfolio i at time t (with t going from 0 to s), and witk is the relative weight of each k asset class in portfolio i at time t.
We can evaluate how the portfolio diversification of the portfolio of agent i and the other j i agents portfolios influences the portfolio risk of the i-th agent, focusing on the idiosyncratic components.3 For this purpose we build a measure of idiosyncratic portfolio risk, based on the standard deviation of the residuals (hereby, RSD) in (1) defined as it(it) [Fama and MacBeth (1973)]. The analysis may be implemented using the idiosyncratic risk of portfolio i as the dependent variable, and where the independent variables are the diversification index of the i-th portfolio, and the average measure of other portfolios diversification followed by a set of variables that describe the asset allocation choices of agent i: it(it) = 0 + 1DIVit + 2DIVt + yy Viyt + it (3)
For the aim of this paper it is useful to refer to the strand of literature related to the traditional financial theory of market models where portfolio returns are explained by different components: i) a constant term, ii) the portion of portfolio return explained by the co-movements between the portfolio and the market returns (systematic factors of portfolio return), and iii) the portion of portfolio returns not explained by either the constant term or the systematic factors, identified as idiosyncratic or specific component [Black et al. (1972)]: Rit = i + biRmt + it (1) where 0 is the constant term, DIVit is the diversification degree of portfolio i at time t, DIVt is the average degree of diversification of the financial system at time t, and Viyt represents the asset allocation variables (with y, going from 1 to q, being the number of the variables), and it being the error term.
The relationship between diversification, asset allocation of economic agents, and systemic risk
The third and final stage of the estimation model focuses on systemic risk and its relationship with diversification. More precisely, it aims to determine how diversification influences the degree of heterogeneity of asset allocation among market agents. In fact, if systemic risk is defined as the risk that a given event produces a simultaneous collapse of all market agents and the entire system, then this condition occurs with a larger probability when the agents are similar and vulnerable to similar threats. In this case, the given event affects all agents in the same way. To measure the level of agents heterogeneity we construct a dispersion index of portfolio asset allocation:
where, i is the constant term, Rit is the return of portfolio i at time t, and Rmt is the market return at time t. bi is the factor that explains the sensitivity of the i-th portfolio return with respect to the return on the market, and it is the portion of portfolio return neither explained by market return nor by the constant term. This portion is defined as the idiosyncratic component of portfolio return, the portion of portfolio return explained by portfolio specific factors.
The impact of diversification on portfolio risk

It is now interesting to investigate the relationship between the model described in equation (1) and the portfolio diversification process. In order to do this, we need to construct a diversification measure of the agents portfolios. For this purpose, we use Herfindahls measure of concentration and compute its complement as proposed by Woerheide and Persson (1993), Lang and Stulz (1994), Byrne and Lee (2003), and Goetzmann and Kumar (2008): HIit = m k=1 w2 itk where DISPit is the dispersion index of portfolio i at time t, witk is the relative weight of k-th asset class at time t in portfolio i, wtk is the average weight of k-th asset class at time t. DISPit measures the extent to which the weights of k asset classes in portfolio i are different from the average weights of the k asset classes in all n agents portfolios for each time t. From this index DISPit it is possible to define an average value for each time t among all portfolios to measure the level of heterogeneity in terms of asset allocation of the financial system, also taking into account the where HIit is the Herfindahl concentration measure of portfolio i at time t; that is the sum of the squared relative weights of k asset classes (witk). Our diversification index DIVit is the complement of (Hit): DIVit = 1 Hit 198 (2)
3 The latter diversification term may be computed as follows: DIVt = 1/n i=1DIVit. This indicator measures the average degree of diversification of the financial system at each time t.
n
DISPit = 1/m m (witk wtk )2 k=1
(4)
agents portfolio sizes as follows:

HETt = 1/n * n DISPit i=1
(5)
Descriptive findings
Data
The dataset consists of 233 balanced investments funds out of a universe of 1,500 large balanced investment funds from November 2001 to December 2010.4 In the dataset we have monthly variables that may be grouped in two categories: i) the main characteristics of funds; ii) variables that capture the composition and the allocation strategy of funds. The first group of variables include i) return, the performance of fund t in a particular month t; and ii) fund size, which is a measure of month-end net assets of fund i in a given month t, recorded in millions of euros. The second category of variables is related to the asset allocation strategy of funds. The main variables are: asset allocation bonds, asset allocation equity, asset allocation cash, asset allocation other (AAb, AAe, AAc, and AAo respectively). They measure the monthly percentages of fund investments allocated to each one of these asset classes for different sub-asset allocations. On the equity portion of the asset allocation, the first sub-category of asset class deals with the geographic allocation: North America (Ena), United Kingdom (Euk), Eurozone (Eeuro), Emerging markets (Eem), Asian developed countries (Easia), and Japan (Ejapan). The other sub-category for asset allocation equity relates to super-sectors, which include many
where HETt is the weighted average heterogeneity index of the financial system at time t. Considering the definition of systemic risk, it is worth investigating when and in what condition a simultaneous collapse occurs and what is the relationship between portfolio diversification, heterogeneity, and a market agents simultaneous downturn indicator. As stated by Brownlees and Engle (2010), and Acharya et al. (2009), a systemic event may be defined as a market loss that surpasses a given threshold (TS) and systemic risk is the expected shortfall suffered by market agents when the systemic event occurs. Hence, Brownlees and Engle (2010) build an expected return estimation model that takes into account different factors in addition to market return. In particular, they measure the expected loss suffered by a portfolio when market losses surpasses TS. The sum of these expected shortfalls is considered as a proxy for systemic risk. Following this approach, it is possible to compute the simultaneous downturn (hereby, SD) rate as the portion of portfolios that record a certain shortfall when the market loss surpasses a given threshold TS: SDrate = [Number of portfolios (Riz < TS)] [Number of portfolios] (6)
similar and homogeneous industries. The equity super-sectors are: manufacturing, information, and services (Eman, Einf, and Eserv, respectively). Bonds constitute the second asset category for which sub-asset allocation observations are available. It is possible to separate bonds allocation into five super-sectors: United States government, United States corporate, Non-U.S. government, Mortgage, and Cash (Busgov, Buscorp, Bnonus, Bmortg, and Bcash respectively). These represent the portions of bond assets that are allocated to bonds of the U.S. government; U.S. private company bonds; bonds issued by public authorities outside the U.S.; bonds related to the many different kinds of mortgages that have been securitized and transformed in market bonds; and bonds with maturities of less than
where z is a specific period in t where the condition Rm<TS occurs. Consequently, following Brownlees and Engle s approach and relating it with the traditional market model estimation described above, we may build a return estimation model which takes into account market return, portfolio diversification, and heterogeneity, to evaluate how diversification and heterogeneity affect this simultaneous downturn rate through the aggregate funds returns. From the simple market model in equation (1) we move to the following return estimation model: Rit = i + b 1i Rmt + b 2iHETt + b 3iDIVit + it
^ ^ ^ ^
(7)
12 months, respectively. The last sub-asset allocation refers to the credit quality rating. For each month, the percentage of total asset allocated to bonds rated aaa, aa, a, bbb, bb, b, below b are available.
where it is the error term. In this way, given a market return, it is possible to assess the impact of diversification and heterogeneity on the return of portfolios and on the average return of the whole financial system; and on the simultaneous downturn rate. The latter allows us to evaluate how diversification and heterogeneity impact the number of portfolios that fall at the same time. Once the coefficients have been estimated, the arbitrary fixed level of market return (market crash) is applied to calculate the average of the portfolio returns, and the SD rate. Finally, to isolate the effects of diversification, the heterogeneity index HETt is fixed so that the trend of average return and the SD rate depends only on the diversification index (DIVit). The same procedure is applied to isolate the effects of heterogeneity fixing the portfolio diversification index.
Summary statistics
Looking at the fund performance measures, returns suggest some further considerations. For the whole sample, returns have a mean of 0.5% [Table 1, column (3)] and a skewness of -.86 [(column (5)]. The presence of fat tails is confirmed by a value of kurtosis equal to 6.84 (column 6). Extreme values in monthly returns are described by 19.02% and 20.71%, respectively [columns (1), and (2)]. Looking at asset allocation, it is quite
Data are provided by Morningstar Italia. The selected funds are those funds that have over 70% of non-missing observations on the asset allocation variables from 2001 to 2010. We decided to start our analysis from the first month after the economic recession of 2001 (November 2001), according to the Federal Reserve Bank of St. Louis estimation.
199
Var/Stat return fs AAe AAb AAc AAo Easia Eem Eeuro Ejapan Euk Ena Einf Eserv Eman Bnonus Bcash Bmortg Buscorp Busgov Ba Baa Baaa Bb Bbb Bbbb Bub
(1) min -19.02 0.00 -2.96 -32.15 -543.79 -488.76 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -2.83 0.00 0.00 0.00 0.00
(2) max 20.72 37300* 458.23 598.85 493.80 100.10 100.00 61.42 100.00 69.09 100.00 100.00 72.67 100.00 100.00 100.00 100.00 99.97 100.00 99.26 100.00 100.00 100.00 73.00 59.78 60.79 60.80
(3) mean 0.50 1010* 51.72 33.70 9.09 5.53 1.55 1.70 10.22 3.21 9.55 50.42 15.62 45.58 38.80 49.11 24.51 7.85 17.16 7.43 16.42 11.25 45.16 2.98 3.44 8.65 0.88
(4) sd 2.69 1850* 19.49 18.92 12.34 15.11 2.52 2.98 17.01 5.03 19.93 32.90 6.72 8.81 9.87 34.04 24.00 13.95 19.05 13.00 12.63 10.62 24.77 6.90 6.55 9.09 2.80
(5) sk -0.86 5.69 -0.10 2.79 -0.30 -0.17 7.77 4.21 3.32 5.27 3.40 -0.08 0.97 0.62 0.52 -0.30 1.38 1.77 1.80 2.83 1.22 1.61 0.14 3.60 3.04 1.98 7.52
(6) kurt 6.84 54.29 11.31 52.79 433.22 110.85 176.34 37.45 14.88 49.10 13.88 1.67 6.59 8.45 5.99 1.56 4.48 5.01 6.45 13.46 5.58 7.15 2.60 19.02 14.34 7.93 99.28
(7) p1 -8.17 0.00 0.99 0.00 -2.19 -2.95 0.00 0.00 0.00 0.00 0.00 0.00 0.12 21.28 15.87 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
(8) p10 -2.58 30.5* 23.50 11.39 1.65 0.00 0.00 0.00 0.00 0.00 0.00 0.30 8.06 36.62 27.57 1.20 0.00 0.00 0.61 0.00 2.17 0.00 8.83 0.00 0.00 0.00 0.00
(9) p25 -0.77 155* 39.42 22.62 3.68 0.13 0.00 0.00 0.78 0.00 0.83 20.96 11.59 40.99 33.32 6.91 6.86 0.00 3.82 0.00 6.85 3.03 28.42 0.00 0.00 2.35 0.00
(10) p50 0.70 485* 56.19 32.46 6.87 0.79 0.76 0.56 5.99 2.14 4.07 50.74 15.17 45.28 38.46 59.56 17.61 0.24 10.14 1.10 14.47 8.70 44.61 0.00 0.54 6.11 0.00
(11) p75 2.04 103* 63.79 42.09 11.63 4.67 2.19 2.21 9.73 4.69 6.65 78.85 18.97 49.79 43.68 77.84 33.98 8.18 24.87 10.00 23.18 16.40 61.02 2.19 3.77 11.68 0.20
(12) p90 3.42 2320* 73.97 58.04 19.06 14.79 4.13 4.71 21.50 6.93 16.27 95.35 23.22 55.18 49.78 88.93 60.28 32.20 41.99 23.81 32.20 24.23 77.50 9.50 11.04 19.43 3.02
(13) p95 4.45 4040* 78.94 65.66 27.28 29.74 5.91 7.50 50.33 10.30 62.78 100.00 59.98 27.17 54.77 94.96 75.80 40.57 56.02 32.99 40.41 32.10 91.23 17.74 18.19 26.88 5.18
(14) p99 6.51 8660* 89.25 80.14 47.46 79.12 9.84 13.35 97.44 23.94 100.00 100.00 37.10 69.26 69.53 100.00 100.00 51.38 88.65 62.50 55.75 49.91 100.00 33.72 31.10 44.38 12.96
(15) Obs. 33977 25676 23194 23191 23190 23194 22980 22980 22980 22980 22980 8178 22966 22966 22966 21428 8005 21428 21428 21428 7125 7125 7125 7125 7125 7125 7125
Legend: Variables (raws): return = fund return; fs = fund size; AA = asset allocation (equity, bond, cash, other); E = sub-asset allocation equity (Asia, Emerging Markets, Eurozone, U.K., Japan, North America, information, services, manufacturing); B = sub-asset allocation bond (U.S. government, U.S. corporate, non-U.S. government, cash, mortgage, aaa, aa, a, bbb, bb, b, under b) Statistics (columns): min = minimum value; max = maximum value; mean = average value; sd = standard deviation; sk = skewness; kurt = kurtosis; p(n) = percentile; md = median value; Obs = number of observations in the sample. The values of percentiles, mean, sd, and range (except for fund size) are expressed in percentage. * = the values are expressed in millions Source: Elaboration on Morningstar data
Table 1 Summary statistics overall sample
clear that funds prefer to invest mostly in equity markets [see AAe, column (3)]. This tendency is stronger for North American funds where over half of the funds investments are allocated to equity markets. The preferred equity market for all funds is surely the North American [see Ena, column (3)]. Asian and emerging equity markets represent a very small portion of investments [see Easia and Eem, column (3)]. Asset allocation classes Cash and Other show very high negative returns (see AAc and AAo, column (1)). This may be a signal of the aggressive short strategies of funds. However, funds seem to be really risk-averse about their bonds allocations. They allocate the largest portion of their bond investments to low risk assets [see table 1, column (3), variables Ba, Baa, and Baaa]. The average percentage of aaa bonds is 45 percent 200 for the whole sample. By contrast, the portion allocated to the riskiest
bonds (below B) is less than 1 percent. Within the sub-asset classes of bonds, funds mostly allocate to non-U.S. government and short-maturity bonds (see column (3), variables Bnonus and Bcash). There are no strong differences between North American and European funds, except for a particular bond sector, mortgages (Bmortg). The North American funds allocate one bond out of 10 to this market, the European funds only allocate one out of 200.5 Counterintuitively, the European funds are on average larger than their North American counterparts. It is useful to focus on the differences between two different time periods: the period preceeding the recent crisis, between January 2005 and June
We define the country of origin for each fund according to the inception domicile provided by Morningstar.

Panel A: pre-crisis period (January 2005 - June 2007) Var/Stat Return Fs AAe AAb AAc AAo Easia Eem Eeuro Ejapan Euk Ena Einf Eserv Eman Bnonus Bcash Bmortg Buscorp Busgov Ba Baa Baaa Bb Bbb Bbbb Bub (1) min -18.25 0 -0.79 -2.25 -119.21 -488.76 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.47 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (2) Max 20.72 20300* 184.26 170.79 493.80 100.10 100.00 61.42 100.00 69.09 100.00 100.00 72.67 100.00 97.53 100.00 100.00 99.97 100.00 99.26 78.57 100.00 100.00 73.00 59.78 60.79 60.80 (3) mean 0.65 823* 51.66 31.65 9.79 6.91 1.42 1.16 9.20 3.16 8.52 60.82 15.61 46.67 37.71 47.97 23.70 7.42 16.09 6.54 17.65 10.86 44.85 2.21 2.77 7.80 0.46 (4) Sd 2.38 1490* 19.02 17.34 11.35 18.03 2.82 2.56 16.43 5.14 18.78 33.58 7.31 8.51 10.00 34.07 23.45 13.64 18.91 13.21 13.10 9.96 24.23 6.35 6.03 8.81 2.13 (5) sk -0.61 5 -0.51 0.78 12.79 -0.73 9.54 6.89 3.51 5.17 3.63 -0.51 0.95 0.43 0.55 -0.25 1.43 1.88 1.91 3.40 1.03 1.84 0.28 4.67 3.61 2.38 11.43 (6) kurt 6.59 34 3.57 5.88 492.09 96.30 203.70 92.71 16.27 48.64 15.72 1.89 6.19 8.52 5.72 1.59 4.77 5.54 6.77 17.66 4.48 9.58 2.82 29.83 18.97 9.95 207.59 (7) Obs 22816 16839 12687 12687 12686 12686 12517 12517 12517 12517 12517 12866 12514 12514 12514 11760 5754 11760 11760 11760 4658 4658 4658 4658 4658 4658 4658
Panel B: time period from last crisis (July 2007 - December 2010) (8) min -19.02 0 -2.96 -32.15 -543.79 -165.36 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -2.83 0.00 0.00 0.00 0.00 (9) max 15.80 37300* 458.23 598.85 111.18 94.00 22.78 33.89 100.00 61.47 100.00 100.00 52.68 100.00 100.00 100.00 100.00 69.61 100.00 84.64 100.00 61.97 100.00 56.75 48.91 56.60 55.12 (10) mean 0.20 1380* 51.78 36.18 8.24 3.85 1.71 2.34 11.44 3.28 10.79 44.81 15.63 44.26 40.10 50.50 26.58 8.38 18.47 8.52 14.09 11.98 45.75 4.43 4.71 10.26 1.68 (11) sd 3.21 2350* 20.06 20.40 13.38 10.32 2.08 3.29 17.60 4.89 21.16 31.11 5.95 8.98 9.55 33.96 25.23 14.30 19.13 12.65 11.35 11.72 25.76 7.62 7.27 9.37 3.62 (12) sk -0.90 6 0.33 4.24 -9.88 2.54 1.86 2.71 3.15 5.40 3.16 0.09 0.96 0.90 0.53 -0.35 1.25 1.65 1.69 2.12 1.66 1.28 -0.08 2.43 2.37 1.42 5.22 (13) kurt 5.89 47 18.79 80.64 378.62 45.53 8.18 14.37 13.61 49.56 12.15 1.74 6.76 9.05 6.63 1.54 3.84 4.45 6.18 8.23 9.37 4.39 2.28 9.64 9.90 5.65 52.84 (14) Obs 11161 8837 10507 10504 10504 10508 10463 10463 10463 10463 10463 5312 10452 10452 10452 9668 2251 9668 9668 9668 2467 2467 2467 2467 2467 2467 2467
Source: Elaboration on Morningstar data
Table 2 Summary statistics for sub-sample periods (January 2005-June 2007; July 2007-December 2010)
2007, and the period when the crisis had become visible, between July 2007 and December 2010. It is possible to observe from Table 2 how the distribution of the returns changes between the first and the second period of our analysis. The average return is 0.65 percent [Table 2, column (3)] in the first period, whereas it decreases to .2 percent [column (10)]. The fat tails phenomenon is more pronounced in the pre-crisis sample: 6.59 percent versus 5.89 percent [columns (6) and (13) respectively]. Furthermore, volatility was greater in the second sample, which covered the post-crisis period [standard deviation was 3.21 percent, column (4)], than the first sample [standard deviation was 2.38 percent, column (11)]. Looking at the four principal asset allocation variables, we find no significantly differences. Asset allocation equity, bond, and cash show a slight increase from the first to the second period, while the asset allocation
other falls strongly falls during the last crisis [see columns (3) and (10), variables AAe, AAb, AAc, AAo respectively]. It seems obvious to think of the investors flight to safety when the crisis took hold.
Estimations results
In this section we implement the estimation model with the sample previously described. The empirical analysis that follows the agent i described above will be proxied by the fund i, while the time t will be with a monthly frequency. As mentioned in the introduction, our analysis consists of three steps that aim to capture different aspects of the financial risk related with the diversified investments. The first two steps evaluate the portfolio risk, decomposed into their two fundamental components. However, the main contribution of the paper is the third step, where a return estimation model stresses the impact of agents diversification strategies and portfolio heterogeneity on the risk of a simultaneous collapse of a number of investors. This effect is also tested by augmenting the classical 201
market model with two additional explanatory factors, diversification and heterogeneity. To check the robustness of our model we also consider a multi-factor model [Fama and French (1993); Carhart (1997)].
considered as mutually independent. The results show that the largest portion of all funds has a beta factor close to the the market line (b = 1). Only few funds show extreme values, less than 0.5 or more than 1.5. More precisely, 47 funds (20 percent of the sample) show these values, whereas eight funds out of ten have a beta factor of between .5 and 1.5. 88 funds (more than 33 percent of the entire sample) show a beta factor value of between 0.8 and 1.2. 89 funds can be defined aggressive with a beta factor of more than one. The largest portion of funds consists of defensive with a beta factor of less than one.8
Portfolio return and the b factor

The first step in our analysis deals with the investigation of the relationship between funds returns and market proxy returns, to estimate the b factor. We estimate the b factor in equation (1) for both sample periods and a pre-sample five-year period. For the sample period, we perform
6
a random effects panel regression as in equation (1) using the following specification: Rit = i + biRmt + d d Cidt + it where Cidt are the set of control variables (with d, going from 1 to g, being the number of the controls), and it is the error term. We check for year effect (year dummies), and region effect (North America). We also define dummies to control for the effect of the crisis (taking 1 if t is between July 2007 and December 2008, zero otherwise), the pre-crisis (January 2006June 2007) and the post-crisis periods (the time periods subsequent to the crises of 2001 and 2007-2009: November 2001/April 2003, January 2009/June 2010, respectively). Other control variables are constructed to take into account the impact of fund size (log fund size), to control for the inception date (inception date; that is 1 if the fund has been incepted after the first month of the time window of analysis (November 2001), and zero otherwise), and to check for the nature of the fund (speculative; takes the value of 1 when the fund is speculative and zero otherwise).7 As Table 3 shows, the sensitivity of funds returns (return) to market returns (Global market) is close to 1 in all the estimates (1) to (10) suggesting high integration between the market and all financial agents. In other words, the co-movements of the funds and market returns are very synchronized over time. The North America dummy is significant and positive, highlighting the overall better performance of North American funds over the past decade. The crises dummies show the expected sign: negative impact on funds return for the post-crises and the crisis [estimates (3) and (5)], and positive impact for the pre-crisis. All the results remain the same when we add the funds characteristics [estimates (6) to (10)]. The results described above stem from a panel regression of equation (1), which returns one beta factor for all funds. However, it may be interesting to estimate different beta for each time series using the following model: Rt = + bRmt + t (8)
6 We use the same approach of Black et al. (1972) in estimating the pre-sample b. We estimate the b factor for the pre-sample period (November 1996 - October 2001) to compare with the b estimation in the sample period. We do not report the pre-sample beta estimation results because we do not rank funds by b factors. 7 In our dataset we define a fund speculative when the overall mean of speculative bonds (sum of the bonds below the triple B) owned by the fund over the time of the analysis is greater than the overall mean of the entire sample. 8 Details on this time series estimation are available from the authors upon request. 9 In computing the idiosyncratic risk as the standard deviation of residuals of equations (1) and (8) with the moving average approach we know that the wider the time window the more significant is the estimation and the higher the influence of older observations. The narrower the time window, the higher the weight of recent observations, the lower the significance of the estimates. To achieve an adequate compromise, we will use a time window of 36 six months, that is the same time window chosen by Morningstar in computing the standard deviation of portfolio returns. 10 Performing the Dorby-Whatson-Hausman test and the Breusch-Pagan/Lagrangian multiplier test, the results suggested that we run a fixed effects panel estimation model. However, we also double checked with a random effects panel estimation model which confirmed the results in table 4.
The impact of portfolio diversification, portfolio asset allocation, and diversification of the other agents on portfolio risk
In this section, we evaluate the relationship between the funds portfolio diversification strategy and the measures of idiosyncratic risk. We consider two proxies for idiosyncratic risk: i) the standard deviation of the panel b factor combined residuals, and ii) the standard deviation of the b factor time series estimation residuals.9 We perform the estimation of equation (3) using the following specification: it(it) = 0 + 1DIVit + 2DIVt + y y Viyt + d d Cidt + it where Cidt are the set of control variables (with d, going from 1 to g, being the number of the controls), and it being the error term. In the first set of regressions [Table 4, columns (1a) and (1b)] the proxy for idiosyncratic risk (RSD) for both panel and time series model is regressed against the funds diversification index DIVit.10 The DIVit coefficients [Diversification in table 4] are both negative and statistically significant, suggesting that diversification negatively affects idiosyncratic risk. In column (2a) and (2b), the estimation takes into account the average diversification (Av. diversification in Table 4) of the entire financial system. The findings may seem counterintuitive because of the positive
where equation (8) is estimated for each fund i. This estimation mod202 el implies, contrary to the hypothesis of panel, that the time series are

(1) Variables Global market North America Post-crises Pre-crisis 2007-2009 Crisis 2007-2009 Year dummies Inception date Speculative Log fund size constant Observations R-squared return 0.92*** (190) Yes 0.07** (2.4) 24272 0.65
(2) return 0.92*** (192.51) 0.13*** (5.05) Yes -0.03 (-0.75) 24272 0.65
(3) return 0.91*** (188.76) 0.13*** (5.08) -0.38*** (-6.32) Yes 0.17*** (3.97) 24272 0.65
(4) return 0.92*** (192.38) 0.13*** (5.05) 0.29*** (-9.4) Yes -0.03 (-0.74) 24272 0.65
(5) return 0.92*** (192.38) 0.13*** (5.05) -0.29*** (-4.96) Yes -0.03 (-0.74) 24272 0.65
(6) return 0.94*** (173.73) Yes -0.01 (-0.2) 0.05** (2.18) 0.00 (0.21) 0.03 (0.22) 18342 0.67
(7) return 0.92*** (192.51) 0.13*** (5.05) Yes 0.01 (0.2) 0.02 (0.69) 0.00 (0.36) -0.10 (-0.67) 18342 0.67
(8) return 0.94*** (170.51) 0.16*** (5.41) -0.34*** (-6.65) Yes 0.00 (0.13) 0.02 (0.7) 0.00 (0.26) 0.08 (0.54) 18342 0.67
(9) return 0.94*** (173.75) 0.16*** (5.4) 0.3*** (4.66) Yes 0.01 (0.24) 0.02 (0.7) 0.00 (0.42) -0.11 (-0.73) 18342 0.67
(10) return 0.94*** (173.75) 0.16*** (5.4) -0.3*** (-4.66) Yes 0.01 (0.24) 0.02 (0.7) 0.00 (0.42) -0.11 (-0.73) 18342 0.67
Legend: (in parentheses: robust t statistics) p***<0.01; ** p<0.05; * p<0.1 Source: Elaboration on Morningstar data.
Table 3 Funds return and market return: the (beta factor)
sign of the coefficient [2.04 and 2.00, columns (2a) and (2b) respectively]. Despite that, the positive sign means that the higher the diversification of all funds the higher the idiosyncratic risk of the single fund. If we consider the two measures at the same time, the fund diversification and the average diversification of all funds, the results described above do not change [columns (3a) and (3b)]. Columns (4a) and (4b) include equity asset allocation among six different geographic regions: Asia (Easia), Emerging markets (Eem), Eurozone (Eeuro), Japan (Ejapan), U.K. (Euk), and North America (Ena). Emerging markets, U.K. and North America have a positive coefficient while Asia, Japan and Eurozone are negative. This difference may be associated with the different degrees by which these regions were imapcted by the recent crisis. North America and the U.K. were both severly affected by the last crises, especially their financial systems. In the U.K., a dramatic example of bank-run risk occurred during the collapse of Northern Rock bank, while both the bursting of the Internet bubble in 2001 and the subprime mortgages crisis of 2007 began in U.S. before spreading to other regions of the world. The negative value for the Emerging Markets coefficient may be instead associated with the high volatility and fragility of these markets, such that an investment in these economies may be
rightly assessed as strongly speculative and risky. By contrast, investments in developed Asian and Japanese markets, also physically far from the last crises centers, appear to reduce portfolio riskiness. In the same way, investments in Eurozone markets reduce idiosyncratic risk. This is probably due to the different structure of European financial system, where financial markets are less volatile and the institutional architecture appears to be more consolidated. The other outward counterintuitive finding is the significantly negative sign for the mortgage bond coefficients [columns (4a) and (4b)]. This could be related to two factors: the good performance of the real estate market during the years preceding the crisis and the credit quality of these assets. There is no information in fact about this specific category of assets, which may be low risk. In the same vein, the cash bonds (with a maturity of lower than one year) have a negative sign, probably because short-term investments are more liquid and generally less risky. The rest of the estimations [from column (5a) to column (8a), and from column (5b) to column (8b) respectively] seem to have a negative impact on idiosyncratic risk before and for a few months during the last crisis. The portfolio idiosyncratic risk is positive from the last crisis onwards. 203
Idiosyncratic risk panel estimation Variables Diversification Average diversification EAsia Eem Eeuro Ejapan Euk Ena Bmort Bcash Post-crises Pre-crisis Crisis Constant Obs R-squared (1a) RSD -0.41 (-10.64) 1.75 (41.11) 21496 0.06 (2a) RSD 2.04 (25.2) -0.19 (-2.91) 24442 0.02 (3a) RSD -0.39 (-11.64) 0.94 (9.33) 0.99 (11.56) 21424 0.05 (4a) RSD -0.43 (-2.21) -0.02 (-0.06) 0.00 (-0.26) 0.01 (2.66) -0.01 (-4.64) -0.04 (-7.59) 0.02 (6.13) 0.00 (3.75) 0.00 (-2.73) -0.03 (-10.38) 1.82 (4.99) 1239 0.22 (5a) RSD -0.38 (-2.11) 0.82 (1.99) 0.00 (-0.49) 0.01 (3.03) -0.01 (-4.48) -0.03 (-6.45) 0.02 (6.61) 0.00 (3.39) 0.00 (-2.34) -0.03 (-9.71) 0.22 (13.59) 1.01 (2.93) 1239 0.33 (6a) RSD -0.44 (-2.36) 0.06 (0.15) 0.01 (0.79) 0.01 (2.09) -0.01 (-3.95) -0.03 (-7) 0.01 (5.5) 0.00 (3.17) 0.00 (-2.06) -0.03 (-9.89) -0.21 (-10.86) 1.80 (5.18) 1239 0.29 (7a) RSD -0.48 (-2.52) 0.31 (0.71) -0.01 (-0.78) 0.01 (2.45) -0.01 (-3.34) -0.03 (-6.96) 0.02 (6.05) 0.01 (4.41) 0.00 (-3.21) -0.03 (-9.29) -0.16 (-8.12) 1.55 (4.35) 1239 0.26 (8a) RSD -0.48 (-2.86) 0.96 (2.52) 0.00 (0.36) 0.01 (1.73) 0.00 (-1.66) -0.02 (-5.23) 0.01 (5.4) 0.00 (3.74) 0.00 (-2.29) -0.02 (-7.81) 0.11 (6.4) -0.27 (-13.69) -0.21 (-9.78) 1.02 (3.22) 1239 0.43 (1b) RSD -0.28 (-10.75) 1.48 (72.65) 21101 0.03 (2b) RSD 2.00 (30.94) -0.34 (-6.49) 23995 0.05
Idiosyncratic risk OLS estimation (3b) RSD -0.26 (-10.11) 1.11 (14.11) 0.58 (8.63) 21101 0.09 (4b) RSD -0.35 (-2.28) 0.55 (1.57) -0.01 (-0.86) 0.00 (0.6) -0.02 (-7.01) -0.03 (-6.65) 0.01 (6.66) 0.01 (5.17) 0.00 (-2.64) -0.02 (-7.03) 1.07 (3.69) 1239 0.13 (5b) RSD -0.32 (-2.18) 1.18 (3.57) -0.01 (-1.12) 0.00 (0.79) -0.01 (-7) -0.02 (-5.49) 0.01 (7.12) 0.00 (4.91) 0.00 (-2.25) -0.01 (-6.14) 0.16 (12.69) 1.07 (3.69) 1239 0.29 (6b) RSD -0.36 (-2.4) 0.61 (1.81) 0.00 (0.02) 0.00 (0.02) -0.01 (-6.47) -0.02 (-6.06) 0.01 (6.1) 0.00 (4.69) 0.00 (-2.05) -0.01 (-6.44) -0.15 (-9.32) 0.46 (1.67) 1239 0.25 (7b) RSD -0.39 (-2.57) 0.79 (2.33) -0.01 (-1.36) 0.00 (0.34) -0.01 (-5.8) -0.02 (-6.02) 0.01 (6.59) 0.01 (5.82) 0.00 (-3.09) -0.01 (-5.94) -0.12 (-7.66) 1.05 (3.77) 1239 0.23 (8b) RSD -0.38 (-2.8) 1.28 (4.11) 0.00 (-0.45) 0.00 (-0.48) -0.01 (-4.58) -0.01 (-4.33) 0.01 (6.04) 0.00 (5.28) 0.00 (-2.21) -0.01 (-4.24) 0.09 (6.18) -0.19 (-11.53) -0.15 (-8.48) 0.86 (3.05) 1239 0.37
Legend: (in parentheses: robust t statistics) p***<0.01; ** p<0.05; * p<0.1 Source: Elaboration on Morningstar data.
Table 4 The effect of diversification and funds asset allocation choices on idiosyncratic risk
The relationship between homogeneous diversification of economic agents and systemic risk
If we accept that systemic risk is the risk that the entire financial system experiences a simultaneous distress when a given event occurs, the term simultaneous plays a prominent role in this concept. Two conditions must be met for an event to impact the entire financial system at the same time: i) the event itself must be able to affect the entire system (systemic event) and ii) the level of similarity (or homogeneity) among agents and institutions must be sufficiently high. As stated in the estimation model see equations (4) and (5) we calculate the heterogeneity index HETt 204 from November 2001 to December 2010.
When we plot an overall view of the simultaneous effects of diversification on idiosyncratic risk, the heterogeneity level of the financial system, and the simultaneous downturn rate (not shown here),11 we find that when diversification increases (until the beginning of the 2007 recession) the
11 We need to choose a threshold (TS) of equation (6) that, if surpassed, expresses the experience of a systemic event. Following the Brownlees and Engles approach (2010) the threshold (TS) may be fixed at -2%. In addition to Brownlees and Engles approach, using the same threshold (2%), we assess the SDrate of the number of funds that experience this loss (higher than 2%) - in the months that market loss surpasses this threshold. The indicator has its highest values during the 2007 financial crisis, and its peak corresponds to the month when Lehman Brothers collapsed. The SDrate index may thus be considered a good proxy for systemic crises and to measure the risk of a simultaneous collapse of

Variables Global Market Diversification Heterogeneity (t-6) Year dummies North America Post-crises Pre-crisis Crisis Inception date Speculative Log Fund size Constant Obs R-squared
(1) return 0.92*** (179) -0.19* (-1.67) -0.40*** (-6.31) Yes
(2) return 0.92*** (179) -0.24** (-2.11) -0.40*** (-6.32) Yes 0.087*** (3.24)
(3) return 0.92*** (175) -0.27** (-2.31) -0.34*** (-5.36) Yes 0.088*** (3.29) -0.31*** (-6.42)
(4) return 0.92*** (179) -0.24** (-2.08) -0.40*** (-6.32) Yes 0.087*** (3.22)
(5) return 0.92*** (179) -0.24** (-2.08) -0.40*** (-6.32) Yes 0.087*** (3.22)
(6) return 0.95*** (163) -0.17 (-1.36) -0.42*** (-5.86) Yes
(7) return 0.95*** (163) -0.25** (-2.01) -0.43*** (-5.88) Yes 0.16*** (4.70)
(8) return 0.95*** (160) -0.27** (-2.15) -0.38*** (-5.18) Yes 0.16*** (4.73) -0.28*** (-4.93)
(9) return 0.95*** (163) -0.25** (-1.96) -0.43*** (-5.87) Yes 0.16*** (4.70)
(10) return 0.95*** (163) -0.25** (-1.96) -0.43*** (-5.87) Yes 0.16*** (4.70)
0.27*** (4.32) -0.27*** (-4.32) 0.0056 (0.18) 0.053** (1.97) 0.0075 (0.97) 0.84*** (6.04) 20757 0.66 0.82*** (5.86) 20757 0.66 0.91*** (6.51) 20757 0.66 0.81*** (5.84) 20757 0.66 0.81*** (5.84) 20757 0.66 0.71*** (3.20) 15818 0.67 0.017 (0.54) 0.020 (0.71) 0.0082 (1.07) 0.64*** (2.92) 15818 0.68 0.015 (0.47) 0.020 (0.74) 0.0078 (1.01) 0.74*** (3.33) 15818 0.68
0.27*** (3.94) -0.27*** (-3.94) 0.018 (0.57) 0.020 (0.74) 0.0086 (1.12) 0.63*** (2.85) 15818 0.68 0.018 (0.57) 0.020 (0.74) 0.0086 (1.12) 0.63*** (2.85) 15818 0.68
Legend: (In parentheses: robust t statistics) p***<0.01; ** p<0.05; * p<0.1 Source: Elaboration on Morningstar data.
Table 5 Short-term impact of diversification and heterogeneity on funds returns using market model
idiosyncratic risk decreases, the level of heterogeneity within the financial system falls, and the simultaneous downturn rate increases.12 When a systemic event occurs, diversification reduces the portfolio-specific risk while increasing the likelihood of a simultaneous collapse of financial institutions. We also find that the relationship between these two factors and portfolio return is characterized by a lagged effect. This hypothesis may be tested through a return estimation model in which these two variables (considering different lags for diversification and heterogeneity at one, six, and twelve months) are taken into account as proposed in equation (7). Consequently, we run panel regression models where the dependent variable is always the monthly fund returns. Different combinations of contemporaneous and lagged variables of the heterogeneity index HETt and the diversification index DIVt are proposed in the estimated model. Among all these regressions, only three models have significant values for both the diversification and heterogeneity variables. Three of them refer to short-run effects of lagged and contemporaneous heterogeneity
(at time t-6, t-1, and t) and contemporaneous diversification on funds returns; the other refers to the long-run effect of lagged heterogeneity (at t-12) and contemporaneous diversification. The short-run results are shown in Table 5. Heterogeneity has a negative impact on funds returns for all the lagged time and the contemporaneous one. Thus, the degree of similarity among portfolios asset allocations produces a positive effect on funds returns in the short-run. This condition is perceived as a mitigator factor for systemic risk. A different way to interpret this result is that agents feel that homogeneity generates
financial system given the market proxy building strategy. The market returns proxy used in this analysis is, in fact, constructed as a weighted average of North American and European market proxies. Moreover, the market proxy is constructed reflecting the funds asset allocation, which is balanced among different asset classes (i.e., equity, cash, bond, other). Hence, the market proxy is a balanced index where at least one component (cash) is substantially less volatile than the others. Consequently, a strong drop in this index may be considered a good signal of a systemic distress. 12 This effect can be described with the expression the two faces of the same coin.
205
Variables Global Market Diversification Heterogeneity (t-12) Year Dummies North America Post-crises Pre-crisis Crisis Inception date Speculative Log Fund size Constant Observations R-squared
(1) Return 0.93*** (177) -0.18 (-1.47) 0.49*** (9.48) Yes
(2) Return 0.93*** (176) -0.24** (-2.00) 0.49*** (9.52) Yes 0.098*** (3.70)
(3) Return 0.93*** (174) -0.26** (-2.16) 0.45*** (8.51) Yes 0.099*** (3.73) -0.25*** (-5.20)
(4) Return 0.93*** (176) -0.23** (-1.97) 0.49*** (9.47) Yes 0.097*** (3.68)
(5) Return 0.93*** (176) -0.23** (-1.97) 0.49*** (9.47) Yes 0.097*** (3.68)
(6) Return 0.97*** (161) -0.14 (-1.06) 0.43*** (7.18) Yes
(7) return 0.97*** (161) -0.23* (-1.75) 0.43*** (7.21) Yes 0.16*** (4.82)
(8) return 0.96*** (159) -0.24* (-1.86) 0.39*** (6.45) Yes 0.16*** (4.84) -0.22*** (-3.94)
(9) Return 0.97*** (161) -0.22* (-1.70) 0.43*** (7.17) Yes 0.16*** (4.81)
(10) return 0.97*** (161) -0.22* (-1.70) 0.43*** (7.17) Yes 0.16*** (4.81)
0.25*** (4.19) -0.25*** (-4.19) 0.011 (0.38) 0.063** (2.35) 0.0088 (1.15) -0.63*** (-5.00) 20137 0.66 -0.66*** (-5.28) 20137 0.66 -0.43*** (-3.27) 20137 0.66 -0.66*** (-5.27) 20137 0.66 -0.66*** (-5.27) 20137 0.66 -0.76*** (-3.63) 15272 0.67 0.022 (0.74) 0.028 (1.02) 0.0096 (1.25) -0.82*** (-3.92) 15272 0.68 0.021 (0.68) 0.029 (1.05) 0.0092 (1.20) -0.62*** (-2.87) 15272 0.68
0.26*** (3.86) -0.26*** (-3.86) 0.023 (0.77) 0.029 (1.05) 0.010 (1.31) -0.83*** (-3.97) 15272 0.68 0.023 (0.77) 0.029 (1.05) 0.010 (1.31) -0.83*** (-3.97) 15272 0.68
Legend: (In parentheses: robust t statistics) p***<0.01; ** p<0.05; * p<0.1 Source: Elaboration on Morningstar
Table 6 Long-term impact of diversification and heterogeneity on funds returns using market model
a positive perception of their asset allocation choices.13 Contemporaneous diversification has a negative impact on monthly funds returns meaning that the portfolio return decreases when the portfolio diversification increases. The long run-effect (Table 6) is characterized by the persistence in the sign for diversification but it switched to positive for heterogeneity. We can interpret this change in the sign as a change in perception of the degree of homogeneity in the asset allocation choices. In the long run, the similarities among funds asset allocation choices is perceived by the market as a potential risk, while in the short-run it was a mitigating factor. In other words, the increasing degree of the homogeneity in the financial system has a negative impact on the funds returns only after a certain time lag. Looking at both short- and long-run effects on diversification and heterogeneity we can conclude that agents in the market perceive asset allocation choices differently based on the lag in time. In 206 the short run, the market seems to appreciate the homogeneity in agents
allocation choice to reduce both portfolio and systemic risk. This homogeneity is perceived by the market as a condition that makes agents more prone to suffer the consequences of a systemic event in the long run. Finally, for the long-run results described above, it is possible to estimate for each fund the coefficients associated with the parameters described in equation (7) through an OLS time series estimation where the dependent variable is the single fund return. The independent variables are: the market proxy returns, the lagged heterogeneity level, and the fund diversification index. Once the coefficients have been estimated, it is possible to measure the predictive powers of the estimates through a panel t-statistic test. Over the entire time window, the model predicts returns
13 We decide to report only results for heterogeneity at t-6 in Table 6. All other lagged measures show the same impact on funds return.

(1) Variables Global Market Diversification Heterogeneity (t-6) Year dummies North America Post-crises Pre-crisis 2007-2009 Crisis 2007-2009 Inception date Speculative Log Fund size SMB HML MOM Constant Observations R-squared Return 0.90*** (152) -0.20* (-1.78) -0.46*** (-7.31) Yes 0.11*** (22.6) -0.00093 (-0.20) 0.011*** (4.68) 0.85*** (6.13) 20757 0.66
(2) Return 0.90*** (152) -0.25** (-2.23) -0.46*** (-7.32) Yes 0.088*** (3.30) 0.11*** (22.6) -0.00080 (-0.17) 0.011*** (4.70) 0.82*** (5.94) 20757 0.66
(3) Return 0.89*** (148) -0.28** (-2.42) -0.40*** (-6.35) Yes 0.089*** (3.36) -0.30*** (-6.07) 0.11*** (22.6) 0.00057 (0.12) 0.0096*** (4.21) 0.91*** (6.54) 20757 0.66
(4) Return 0.90*** (152) -0.25** (-2.20) -0.46*** (-7.34) Yes 0.088*** (3.29) 0.17*** (2.83) 0.11*** (22.3) -0.0017 (-0.36) 0.011*** (4.94) 0.82*** (5.94) 20757 0.66
(5) Return 0.90*** (152) -0.25** (-2.20) -0.46*** (-7.34) Yes 0.088*** (3.29) -0.17*** (-2.83) 0.11*** (22.3) -0.0017 (-0.36) 0.011*** (4.94) 0.82*** (5.94) 20757 0.66
(6) Return 0.92*** (139) -0.18 (-1.46) -0.47*** (-6.57) Yes 0.0088 (0.29) 0.054** (2.06) 0.0073 (0.96) 0.11*** (20.6) 0.0093* (1.75) 0.012*** (4.74) 0.69*** (3.17) 15818 0.67
(7) Return 0.92*** (139) -0.26** (-2.12) -0.47*** (-6.60) Yes 0.16*** (4.78) 0.020 (0.65) 0.021 (0.78) 0.0080 (1.06) 0.11*** (20.6) 0.0094* (1.79) 0.012*** (4.75) 0.63*** (2.89) 15818 0.68
(8) Return 0.92*** (136) -0.28** (-2.25) -0.43*** (-5.88) Yes 0.16*** (4.80) -0.26*** (-4.69) 0.018 (0.59) 0.022 (0.80) 0.0076 (1.00) 0.11*** (20.6) 0.011** (2.02) 0.011*** (4.39) 0.71*** (3.27) 15818 0.68
(9) Return 0.93*** (139) -0.26** (-2.09) -0.47*** (-6.60) Yes 0.16*** (4.77) 0.16** (2.33) 0.020 (0.67) 0.022 (0.79) 0.0083 (1.09) 0.11*** (20.3) 0.0086 (1.63) 0.013*** (4.95) 0.62*** (2.85) 15818 0.68
(10) Return 0.93*** (139) -0.26** (-2.09) -0.47*** (-6.60) Yes 0.16*** (4.77) -0.16** (-2.33) 0.020 (0.67) 0.022 (0.79) 0.0083 (1.09) 0.11*** (20.3) 0.0086 (1.63) 0.013*** (4.95) 0.62*** (2.85) 15818 0.68
Table 7 Short-term impact of diversification and heterogeneity on funds returns using multi-factor model
values that are not statistically different from effective values.14
difference between the returns on diversified portfolios of high and low book to market stocks.15 The model by Carhart (1997) extends the FamaFrench model by incorporating an additional fourth factor that considers the momentum anomaly. The four factors model can be explained as a performance attribution model where the coefficients and premia on the
The multi-factors model

The return estimation model described in equation (7) is tested here using the multi-factor approach presented by Fama-French (1993), and by Carhart (1997):
^ Rit = i + b 1i Rmt + b 2iHETt + b 3iDIVit+ b 4iSMBt + b 5iHMLt + b 6iMOMt + ^ ^ ^ ^ ^ ^
it
(9)
where SMBt (small minus big) is the difference between the returns on diversified portfolios of small and big stocks, HMLt (high minus low) is the
14 We perform the t-test on panel data as follows: t = [ Rit R it ) (N2g/2N)] ^ ^ (N-1) 2R it +(N-1) 2Rit, where Rit and R it are the cross-sectional average values of monthly effective returns and estimated returns respectively, N is the number of cross-sectional observations, g is the number of degrees of freedom, 2 is the relative variance. 15 The book-to-market ratio attempts to identify undervalued or overvalued securities by taking the book value and dividing it by the market value. In basic terms, if the ratio is above 1 then the stock is undervalued; if it is less than 1, the stock is overvalued.
207
(1) Variables Global market Diversification Heterogeneity (t-12) Year dummies North America Post-crises Pre-crisis Crisis Inception date Speculative Log fund size SMB HML MOM Constant Observations R-squared 0.099*** (19.3) -0.018*** (-3.48) 0.0052** (2.15) -0.35*** (-2.78) 20137 0.66 return 0.91*** (149) -0.17 (-1.48) 0.26*** (4.78) Yes
(2) return 0.91*** (149) -0.24** (-2.02) 0.26*** (4.81) Yes 0.097*** (3.75)
(3) return 0.91*** (146) -0.26** (-2.21) 0.21*** (3.71) Yes 0.098*** (3.79) -0.28*** (-5.81)
(4) return 0.92*** (149) -0.23** (-2.00) 0.26*** (4.82) Yes 0.097*** (3.73)
(5) return 0.92*** (149) -0.23** (-2.00) 0.26*** (4.82) Yes 0.097*** (3.73)
(6) Return 0.94*** (136) -0.14 (-1.09) 0.20*** (3.07) Yes
(7) return 0.94*** (136) -0.23* (-1.78) 0.20*** (3.11) Yes 0.16*** (4.81)
(8) Return 0.94*** (134) -0.24* (-1.91) 0.15** (2.26) Yes 0.16*** (4.84) -0.25*** (-4.58)
(9) return 0.94*** (136) -0.22* (-1.74) 0.20*** (3.11) Yes 0.16*** (4.81)
(10) return 0.94*** (136) -0.22* (-1.74) 0.20*** (3.11) Yes 0.16*** (4.81)
0.19*** (3.06) -0.19*** (-3.06) 0.014 (0.45) 0.0090 (1.18) 0.064** (2.43) 0.099*** (19.3) -0.018*** (-3.46) 0.0052** (2.15) -0.38*** (-3.03) 20137 0.66 0.10*** (19.6) -0.018*** (-3.48) 0.0039 (1.59) -0.12 (-0.90) 20137 0.66 0.098*** (19.0) -0.019*** (-3.65) 0.0059** (2.41) -0.39*** (-3.05) 20137 0.66 0.098*** (19.0) -0.019*** (-3.65) 0.0059** (2.41) -0.39*** (-3.05) 20137 0.66 0.11*** (18.0) -0.014** (-2.32) 0.0048* (1.74) -0.48** (-2.28) 15272 0.67 0.024 (0.81) 0.0098 (1.29) 0.030 (1.11) 0.11*** (18.0) -0.014** (-2.30) 0.0049* (1.75) -0.54*** (-2.58) 15272 0.68 0.023 (0.75) 0.0093 (1.23) 0.031 (1.13) 0.11*** (18.3) -0.014** (-2.30) 0.0037 (1.33) -0.30 (-1.38) 15272 0.68
0.17*** (2.58) -0.17*** (-2.58) 0.025 (0.83) 0.010 (1.33) 0.031 (1.12) 0.10*** (17.7) -0.014** (-2.45) 0.0055** (1.98) -0.55*** (-2.63) 15272 0.68 0.025 (0.83) 0.010 (1.33) 0.031 (1.12) 0.10*** (17.7) -0.014** (-2.45) 0.0055** (1.98) -0.55*** (-2.63) 15272 0.68
Table 8 Long-term impact of diversification and heterogeneity on funds returns using multi-factor model
factor-mimicking portfolios aim to explain the proportion of mean return attributable to four elementary strategies, where MOMt represents the one-year momentum in stock returns. it is the error term. The multi-factor model confirms (Tables 7 and 8) the short- and long-term impact of diversification and heterogeneity on funds returns. As well as in the previous results, the short-run negative effect of heterogeneity appears at the same lagged (t-6, and t-1) and contemporaneous (t) time period. Model (9) also confirms the shift to the positive impact heterogeneity effect in the long run (t-12). Finally, moving from the market model to the four-factor model, we still support the persistence in contemporaneous diversification negative effects on funds returns for both the short and 208 long run.
Concluding remarks
The recent financial crisis has highlighted systemic risk as a possible, and a very important, variable that can, and should, play a role in the decisions taken by policymakers. This paper investigates two aspects of agents portfolio heterogeneity in terms of asset allocation: within a single portfolio and across investors portfolios. The latter may be considered one possible source of a systemic distress. The rationale behind this idea is that if agents portfolios become more similar to each other, the likelihood of a simultaneous collapse increases. The analysis has been implemented through a sample of investment funds over the last decade, in three steps. The first two steps relate to the impact of diversification on the two portfolio risk components,

respectively: systematic and idiosyncratic. The last one focuses on the impact of portfolio diversification on systemic risk. The findings appear to suggest that diversification, even if is confirmed to be useful for reducing portfolio specific risk, could result in an increase in the degree of homogeneity among investors. This condition increases the risk that a negative systemic event produces a simultaneous collapse. If the agents allocated their wealth to the same assets, a negative event impacts all agents in the same way and at the same time. Our results corroborate Wagner (2006), who argues that total diversification is undesirable because while it reduces risk within an individual institution it increases the risk of a systemic crisis, and De Vries (2005), who argues that diversification reduces the frequency of failure of individual institutions when a shock is small and easily borne by the system but increases the likelihood of a systemic failure when a stronger shock occurs. Further strands of research may follow the investigation implemented in this paper. For example, it is possible to cluster the sample by regions (i.e., North America, Europe, Asia, Emerging Markets), rank portfolios by funds beta factors, or improving the model forecasting ability for heterogeneity, diversification, and systemic risk. These research proposals are beyond the scope of the paper, which may also be a warning in opposition to the recent provisions of financial authorities. Common capital adequacy rules, indeed, while increasing transparency, also encourage homogeneity in investment strategy and undertaking of risk, leading to a high concentration of risk. That means that global regulations can be dangerous because they may increase the amplitude of global credit cycles [TaxPayers Alliance (2010)].
References

De Nicolo, G. and M. L. Kwast, 2002, Systemic risk and financial consolidation: are they related?, Journal of Banking and Finance, 26:5, 861-880 De Vries, C. G., 2005, The simple economics of bank fragility, Journal of Banking and Finance, 29:4, 803-825 Dow, J., 2000, What is systemic risk? Moral hazard, initial shocks and propagation, Monetary and Economic Studies, Institute for Monetary and Economic Studies, Bank of Japan, 18:2, 1-24 Eisenberg, L. and T. H. Noe, 2001, Systemic risk in financial system, Management Science, 47:2, 236-249 Fama, E., 1972, Components of investment performance, Journal of Finance, 27:2, 551-67 Fama, E. and K. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics, 33:1, 3-56 Fama, E. and D. MacBeth, 1973, Risk, return, and equilibrium: empirical tests, Journal of Political Economy, 81:3, 607-36 Federal Reserve Bank of St. Louis, 2011, GDP time series data, http://research.stlouisfed. org/fred2/ Goetzmann, W. and A. Kumar, 2008, Equity portfolio diversification, Review of Finance, 12(3):433-463. Kambhu, J., K. Schuermann, and T. Stiroh, 2007, Hedge funds, financial intermediation and systemic risk, Federal Reserve Bank of New York staff reports N. 291 Kaufman, G., 1996, Bank failures, systemic risk and bank regulation, The Cato Journal, 16:1, 17-45 Kupiec N. L. and D. Nickerson, 2004, Assessing systemic risk exposure form banks and GSEs under alternative approaches to capital regulation, Journal of Real Estate Finance and Economics, 28:2/3, 123-145 Lang, L. and R. Stulz, 1994, Tobins q, corporate diversification and firm performance, Journal of Political Economy, 102:6, 1248-1280 Lehar, A., 2004, Measuring systemic risk: a risk management approach, Journal of Banking and Finance, 29:10, 2577-2603 Merton, R., 1974, On the pricing of corporate debt: the risk structure of interest rates, Journal of Finance, 29:2, 449-70 Schinasi, G. and T. Smith, 2000, Portfolio diversification, leverage and financial contagion, IMF staff papers, 47:2, 159-176 Schwarcz, L., 2008, Systemic risk, Georgetown Law Journal, 97:1, 193-249 Sheldon, G. and M. Maurer, 2008, Interbank lending and systemic risk: an empirical analysis for Switzerland, Swiss Journal of Economics and Statistics, pp. 685-704. Stern School of Business, New York University, NYU, 2011, Systemic risk rankings, http:// vlab.stern.nyu.edu/welcome/risk/ Summers L., 2000, International Financial Crises: Causes, Prevention, and Cures, The American Economic Review, Volume 90, N2 TPA, 2010, Financial regulation goes global: risks for the world economy, Tax Payers Alliance Tirole, J., 2002, Financial crises liquidity and the international monetary system, Princeton University, Princeton Wagner, W., 2006, Diversification at financial institutions and systemic crises, Journal of Financial Intermediation, Volume 19, Issue 3, pp. 373-386 Woerheide, W. and D. Pearson, 1993, An index of portfolio diversification, Financial Services Review, 2:2, 73-85
Acharya, V. V., 2009, A theory of systemic risk and design of prudential bank regulation, Journal of Financial Stability, 5:3, 224255 Acharya, V. V., L. H. Pedersen, T. Philippon, and M. Richardson, 2010, Measuring systemic risk, Mimeo, Stern School of Business, New York University Allen, F., A. Babus, and E. Carletti, 2010, Financial connections and systemic risk, EUI working paper series N. 16177 Allenspach, N. and P. Monnin, 2007, International integration, common exposure and systemic risk in the banking sector: an empirical investigation, Swiss National Bank Working Paper BIS, 1994, Systemic risk, 64th annual report, Bank for International Settlement Bartram, S., G. Brown, and J. Hund, 2007, Estimating systemic risk in the international financial system, Journal of Financial Economics, 86:3, 835869 Becker, S. and M. Hoffmann, 2008, Equity fund ownership and the cross-regional diversification of household risk, Stirling economics discussion paper N. 25 Black, F., M. Jensen, and M. Scholes, 1972, The Capital Asset Pricing Model: some empirical tests, in Jensen, M. C., Studies in the Theory of Capital Markets, Praeger, 79-121 Brownlees, C. and R. Engle, 2010, Volatility, correlation and tails for systemic risk measurement, Mimeo, Stern School of Business, New York University Byrne, P. and S. Lee, 2003 An exploration of the relationship between size, diversification and risk in UK real estate portfolios: 1989-1999, Journal of Property Research, 20:2, 191-206. Carhart, M. M., 1997, On persistence in mutual fund performance, Journal of Finance, 52:1, 57-82 CFTC, 2008, Systemic risk, U.S. Commodity Futures Trading Commission De Bandt, O. and P. Hartmann, 2000, Systemic risk: a survey, ECB Working papers N. 35
209
210
PART 2
Breaking Through Risk Management, a Derivative for the Leasing Industry

Sylvain M. Prado Consultant, Capco Ram Ananth Principal Consultant, Capco1
Abstract
In the leasing industry the lessor faces the risk, at the end of the contract, of not being able to recover sufficient capital value from the resale of the asset. We propose a financial product to hedge residual value risk. Furthermore, we discuss the contributions of derivatives to risk management.
1 We would like to thank Manish Mandalia of Capco for his comments on the first version of the paper. The views expressed in this article are those of the authors and are in no way representative of the views of Capco or any of its partners.
211
The academic approach to risk management accepts fantasy rather than reality
In 2000, a group of Parisian economics students funded the movement for Real-world Economics to contest the usual approach of economic orthodoxy: a limited concern for the relationship between the economic theory and the real world. The students argued that concepts like the representative agent, the maximization of a utility curve, or the Efficient Market Hypothesis (EMH) do not help in getting a better understanding of the economy. In 2001, the movement was supported by Cambridge
2
The leasing industry provides an essential source of financing for small companies but appears to be a little known area, sparking off a limited interest from financial markets and academics
The International Accounting Standard for Leases, or IAS17,3 defines a lease as an agreement whereby the lessor conveys to the lessee, in return for a payment or series of payments, the right to use an asset for an agreed period of time. Legislation may be different across countries, contracts can be adjusted to meet the needs of a customer, and any sort of good can, in theory, be leased. Consequently, the term lease includes a large variety of contracts. On the whole, a lease is a financial instrument for the procurement of equipment. In a leasing contract, a lessor provides equipment for usage on a defined period of time to a lessee for specified payments. The asset leased could be any kind of equipment (i.e., printers, trucks, aircrafts) used by the lessee for business purposes. The lessor purchases the equipment and has the legal ownership of the asset. To use the equipment the lessee makes periodic payments throughout the contract to the lessor. The entire global leasing market [White (2008)] was estimated to be more than U.S.$633 billion in 2006. Europe and North America accounted for 41 percent and 38 percent, respectively, of the world market. The World Leasing Yearbook 2009 reported a total global amount of equipment leased to be more than U.S.$760 billion for 2007. The European leasing associations reported new leasing volumes of above 330 billion and a contribution to financing, on average, of 18% of total European investments for 2008. Furthermore, leasing involves more than 16 million cars in Europe [Lease Europe Association website]. Broadly speaking, almost all academic contributions to the subject of leasing are related to the comparison of leasing with lending and purchasing. Researches can be divided into three groups. The first group analyze the tax advantages leasing over regular debt [Smith and Wakeman (1985); Fung and Subrahmanyam (1987); MackieMason (1990); Goodacre (2003)]. The second group discuss the leasing puzzle. Theories in finance and economics suggest that leases and debt are substitutes; an increase in leasing should lead to a decrease in debt. Moreover, standard finance theories view cash flows from lease obligations as being equivalent to cash from debt. Ang and Peterson (1984) found empirically that lease and debt appear as complements for companies. They argued that greater leasing is associated with greater debt. They could not find a conclusive explanation and called their finding the
students; they published a manifesto, Opening up economics, which was signed by 797 economists. Forums were organized and new reviews were created all over the world to propose a different approach to economics. Within economic institutions, however, nothing happened. Shojai and Feiger (2011) showed that the 2008 crisis had no effect on the economic mainstream despite the fact that it dramatically highlighted the irrelevance of the economic theories. The main reviews were still publishing articles with no value in the understanding of the real world; they were still using oversimplified models and irrelevant concepts like the EMH. Bernard Guerrien [Guerrien and Gun (2011)], a Sorbonne economist and one of the founders of the Post-Autistic movement, concluded that unfortunately, discussions focused on the validity of EMH will probably continue, and confusion will persist, with time and energy lost in vain. Following Mark Planck, do we have to conclude that a scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die and a new generation grows up that is familiar with it? And, all things considered, do we really have to worry about these theoretical problems? Unfortunately, we should be genuinely worried about serious failures in a specific field of applied economics: risk management. According to Shojai and Feiger (2010), risk tools and concepts are of no use to improve the survival of companies or the economic system. Built over non-relevant assumptions, they do not really manage risks. Mainstream economists can reply that it is easy to criticize, but to propose new practices and innovative solutions is something altogether different. We aim to illustrate another way to practice risk management. We propose a new financial derivative allowing for the management of residual value risk and providing a real value to the economy, not only a new mechanism for speculation or a fee generator for investment banks. Moreover, this article is intended for people within the leasing industry interested in an innovative financial product, as well as people from the financial markets concerned with leasing risk opportunities. To be more specific, we aim to hedge residual risk on operating lease contracts. 212
2 3
At that time, one of the authors was an economics student in Paris. All listed companies in the European Union are obliged to adopt the International Standards. The equivalent standard in the U.S. is FAS 13.

unsolved puzzle in finance or the leasing puzzle. Following their seminal article, many have tried to investigate the results [Marston and Harris (1988); Lewis and Schallheim (1992); Krishnan and Moyer (1994); Branson (1995); Beattie et al. (2000); Yan (2006)]. The puzzle leads to the third area of research covering every article related to the incentives and advantages of leasing. Lasfer and Levis (1998), for instance, rephrased the problem. In theory, leasing provides no added value to the purchase of equipment when a lessee or a purchasing firm, lend or borrow at the same rate of interest, have a similar tax status, and expect the same resale price of the asset at the end of the contract. In practice, however, these conditions are not satisfied and there are several motives for leasing.4 Their study suggest that leasing allows small firms to finance their growth, and/or survival while for large firms, leasing appears to be a financial instrument used by sophisticated financial managers to minimize the after-tax cost of their capital. Drakos and Goulas (2008a) mention the uncertainty in entrepreneurship exacerbated by the irreversibility of capital. By allowing the separation of usage to the commitment of equipment ownership,5 leasing facilitates a possible disengagement in unsuccessful projects. Schmit and Stuyck (2002)6 and Schmit (2003, 2004)7 demonstrate that leasing is a relatively low risk activity and that Basel II requirements should be reduced for leasing contracts. All in all, leasing appears as a safe financial instrument, furthermore, leasing can reduce environmental problems [Fishbein et al. (2000)]. The practice of leasing products, rather than selling them, prevents waste generation and creates a pattern of closed loop material use. For specific equipment and leasing contracts, manufacturers become more aware of operating problems and cost management for products at life end. Consequently, they may redesign their products accordingly. Staying in a commercial channel, the equipment is less likely to be stored or discarded when it becomes obsolete.
4
Value
Original equipment cost Depreciation
Residual value
Time
Lease period
Lease rental = [(original equipment price residual value) lease period] + interest rates + taxes + service charges
Figure 1 Depreciation in a leasing contract
The lessor has to set the end of contract market value of the equipment, called residual value. It is a forecasted value of the asset. The three parameters mainly define the level of depreciation (which can be seen as the variance between the original equipment price and the residual value all along the lease period) and the rental amount paid by the lessee to use the equipment. The depreciation part of the lease payment (usually the larger component) can be calculated by the total amount of depreciation divided by the number of periods.9 Additionally, several features are included in the payments made by the lessee during the contract; depreciation of the asset interests on the lessor investment, servicing charges (including operational costs, insurance, counseling, and repairs) and taxes. Figure 1 illustrates the mechanism involved in the calculation of the lease payments.10 As a result, when the residual value decreases, the variance with the original equipment cost decreases, as well as with the rental amount. A low residual value creates, for the lessee, an opportunity to pay a lower
Leasing companies are not only intermediaries; their expertise produces a real added value in the leasing process. They select the appropriate equipment based on its ability to improve the leasing cash flow through various parameters like equipment characteristics, economic life of the asset, taxes, or residual value risk. Leasing companies also have skills in finance, credit, equipment acquisition, and dealing. All things considered, they facilitate the transactions between equipment suppliers and equipment users.8 For leasing companies, the risk of loss on sales at the end of the contract term, as well as the pricing, are critically impacted by the forecasted resale price of the asset.
The lessor faces crucial challenge: the residual value risk

The three key parameters of a leasing contract are the original equipment price, the lease period, and the residual value. They drive the competitiveness and, at the same time, risk on the asset.
They listed three main motives for leasing: first, leasing can reduce financial taxes; second, it could be advantageous financially for distressed companies to get an agreement and access to the equipment because lessors have first claims over the asset; and finally, leasing can reduce agency costs because it does not constitute an investment for the lessee. 5 Capital heterogeneity would explain different choices of leasing by different degrees of irreversibility. 6 Schmit and Stuyck (2002) analyzed defaulting leasing contracts issued from 1976 to 2000 in six countries and originated from 12 major European companies and found that the recovery rate of defaulting leasing contracts are comparable to senior secured bonds. 7 The estimation of the probability density function of losses and the standard portfolio credit value at risk (VaR) measures in the leasing industry reveals a relatively low risk activity [Pirotte and Vaessen (2008)]. 8 Among the few contributions related to the advantages of leasing, we could also mention Kichler and Haiss (2009), showing the support that leasing has provided in economic growth of central and Eastern Europe, as well as the thesis of Brage and Eckerstm (2009), comparing the incentives to leasing in Sweden and Japan. 9 For instance, in an automotive lease contract with an original equipment price of $12000, a residual value of $7,200, and a lease period of 24 months, the monthly payment of the depreciation (excluding interest rate and taxes) should be $200 per month [(12,000 7,200)/24]. 10 For simplification, we did not mention another key element: the end of term options. At the end of the contract, there are options allowed to the lessee. Lease period can be extended; lease can be renewed; the equipment can be bought or returned.
213
Value
1987
1991
1997
2001
1988
1989
1990
1994
1995
1996
1998
1999
2000
2004
2005
1992
1993
2002
Loss on sale
Time
Global leasing volume (U.S.$ bln)
Global CDO issuance (U.S.$ bln)
Sources: White Clark Global Group 2008), World Leasing Yearbook 2009, SIFMA
Figure 2 Fair market value of an asset
Figure 3 Volume of equipment leased in the world
amount for the usage of the equipment. At the same time, it makes the lessor more price competitive. But being price competitive could lead to difficulties when market conditions change. The lessor faces the risk of not being able to recover sufficient capital value from the resale of the asset, the residual value risk. As illustrated in Figure 2, the fair market value curve implies a gain on sale, or a loss, depending on the level of residual value.
Simply speaking, the reader should have in mind a portfolio of cars leased for several years. The second-hand car market is the main source for residual value calculations. Residual values correlations, for instance, can be estimated using resale market statistics. In addition to resale prices, asset characteristics and price indices are also used to set model variables. Hedging residual value risk could be done through a security derivative.
So the lessor faces a dilemma: the lower the residual value, the lesser the risk of loss on sale, but the higher the rental payment and the worse the price competitiveness. Conversely the higher the residual value risk, the better the price competitiveness. As a matter of fact, few academic papers and developed models are dedicated to residual value risk. For instance, although leasing has a long history and the volumes involved are quite similar (Figure 3) to structured finance, the latter has attracted greater attention within academic circles. Sadly, leasing is still perceived as a niche part of the financial markets but we believe that the recent problems in the credit markets may boost the leasing sector.11
Security definition includes financial security (bond, stock) but also capital market securities (mortgage, long-term bonds). It is an investment instrument which offers benefits of debt or equity. A security derivative is a financial security whose value is derived in part from the value and characteristics of another security, the underlying asset. And securitization is the process of aggregating similar securities that can be transferred or delivered to another party. An innovative derivative would allow the lessor to transfer the risk to a fourth party (i.e., an insurance company). Residual value risk and credit risk have a clear analogy: portfolios are constituted of units that are more or less risky. A lease portfolio is similar to a loan portfolio; both could be divided into systematic and idiosyncratic risks; losses occur when certain events happen and the correlation risk has a huge impact. Consequently, applying a credit model to residual value risk is attractive. We propose to replicate a model from credit risk to residual value risk. The famous one factor model allows for the creation of a residual value derivative. Pykhtin and Dev (2003) already applied the one factor model to auto lease. They calculated the economic loss associated with residual risk, leading to an estimate of economic capital (the amount of capital required to support the level of risk and maintain a target debt rating). The model was constructed and modified for financial leases with the option
A solution to residual value risk management would be the use of an innovative financial product
To illustrate our approach, we focus on automotive leasing contracts. The high residual values of cars at the end of the contract, by comparison with other types of equipment, increases the probability of losses and represents a critical challenge for leasing companies. Two kinds of auto lease contracts can be defined: short-term auto leases include cars (or trucks) rented to private or professional clients for a relatively short period of time in order to meet their occasional transport needs; and long-term auto leases, which cover contracts for businesses outsourcing their vehicle fleet needs to a leasing company. In this paper, 214 we focus on long-term auto leases.
11 Companies turn to leasing to combat credit drought, Financial Times September 20, 2011.
2003
Residential value
2006
2007
Gain on sale
900 800 700 600 500 400 300 200 100 0

to buy out (the lessee has a purchase option at the end of the contract). The loss distribution was calculated for a fine grained portfolio (specific to large portfolios without significant individual exposure) and the model was only driven by the systematic factor. Because we aim to hedge residual risk using a financial product, our approach differs slightly. The new derivative should cover residual value risk in a portfolio of leases. The model, detailed in the Appendix, is built according to leasing contract characteristics. It is inspired by synthetic Collateralized Debt Obligations (CDO). Like a CDO, the residual value derivative can be sliced and diced, and tranches can be sold. Like a standard derivative, it would allow insurance or hedging for the lessor. In addition, it would allow the counterparty to take an opposite position. Through the selected residual value level and tranches limits, lessor and lessee may be able to select a level of risk. Finally, in case of negative correlation, the derivative could go against a downturn in the sectorial market and create opportunities for risk diversification.
By comparison, car sales provide a lot of information. Millions of cars are sold every year. The frequency of sales allows for an accurate calibration of residual value models. Abundant and relevant information will limit the irrational volatility of the derivative price. Additionally, the value of the residual derivative reflects the quality of a car, not the subjective credit quality of an institution or a customer. Cars (and equipment in general) have to follow high and concrete standards of quality. The volume of information and the external position of the automotive sector (manufacturers, experts, etc.) restrict the academic practice of creating a theoretical structure and impos[ing] it [on] to the data. Shojai and Feiger (2010) reveal another malpractice in risk management: the fallacy in academic approaches to risk management (enthusiastically adopted by the financial institutions themselves) is to assume that the techniques shown to be reasonably useful for the analysis of large samples of individual instruments can be of significant value in assessing the others level of risk. In other words the risk at the level of the individual financial instrument, the risk of a financial institution holding diverse instruments, and the risk of the system of financial institutions are not the same. The residual value derivative does not fall into this fallacy. The mechanisms behind residual value require the portfolio analysis to be performed at the asset level (the car). The calibration of the model has to be done on an individual level. Finally, Shojai and Feiger (2010) contest the assumption of a stationary risk distribution. The convergence of trading behavior, credit-fuelled valuation bubbles, and the fact that economic events sometimes have causes which bear rather sharply on the correlation among instruments prevent a stationary distribution in capital markets.
Ironically, the credit risk model appears more relevant for residual value risk management
An academic paper related to a new financial product has usually two main components: a review of literature and a lot of mathematics. We want, however, to provide a broader contribution to the risk management practices. We raise an essential question: what is the real value of our new derivative for risk management? Our model is inspired by CDOs. For this reason, the 2008 crisis shed light on our approach. The article of Shojai and Feiger (2010), explaining the CDO crisis and highlighting the failures of risk management, constitutes a basis for the discussion. They argue that CDOs created a problem of incentives within banks: the fact that the loan remained on the banks own books ensured that they monitored the activities of their brokers more closely, since any defaults would directly cost the bank itself; the bank is no longer a mortgage company but simply a financial engineer of mortgages. The incentive problem discloses another issue in credit risk modeling: the scarcity of relevant information. Defaults remain exceptional events.12 Relatively few banks and countries go into bankruptcy every year. There is insufficient information for a robust calculation of default probability. Credit ratings provided by credit agencies and widely used for model calibration, are subjective. The implied default probabilities, calculated from market prices, are not relevant (the historical behavior of prices in the CDO market remaining a perfect example). Uncertainty and vagueness around credit quality amplified the problem of incentives. They made it difficult to realize the extent of low quality assets within books. They hid indirect threats for financial institutions.
We believe, by comparison, that second hand car prices are stable. The main reason being the interaction between the new and the second hand market; high prices in the second market, for instance, bring back buyers to the new market [Prado (2010)]. Consequently, a stationary distribution in the used car market does not appear as a strong assumption. On the whole, the new derivative provides a solution to minimize causation and correlation problems around residual value. It will improve the liquidity and the geographical diversification in the leasing industry [Prado (2009)]. Ultimately, in case of an irrational increase in the derivative price, leasing companies could still leave financial markets and bear the risk again.
12 Except maybe on a microeconomic level, for example, credit cards.
215
Conclusion
Leasing supplies a critical means of financing for small companies. In a leasing contract, however, the risk of loss on sales and the pricing are critically impacted by the forecasted resale price of the asset. The lessor faces a residual value risk. Regrettably, leasing has sparked off only a limited level interest in the financial markets. There is almost no academic literature on residual value risk.
We transpose the one factor model, initially used for credit risk, onto residual risk. A new derivative product is created. Pooling together a large portfolio of equipment that has been leased, the derivative converts residual risks into an instrument that may be sold in the capital markets. It allows the lessor and lessee to select their degree of exposure to residual value risk and to improve price competitiveness. As a result, the model is a contribution geared for people from the leasing industry interested by an innovative financial product, as well as people from the financial markets interested in leasing risk opportunities. Through a comparison with CDOs, we assess the safety of the new derivative for the economic system and we highlight its contribution towards robust risk management practices.
References

216
An, Y., 2006, Leasing and debt financing: substitutes or complements? Journal of Financial and Quantitative Analysis, 41:3, 709- 732 Ang, J. and P. P. Peterson, 1984, The leasing puzzle, Journal of Finance, 39:4, 1055-65 Andersen, L., J. Sidenius, and S. Basu, 2003, All your hedges in one basket, Risk, 16, 67-72 Beattie, V., A. Goodacre, and S. Thomson S., 2000, Operating leases and the assessment of lease-debt substitutability, Journal of Banking and Finance, 24:3, 427-470 Bluhm, C., L. Overbeck, and C. Wagner, 2003, An introduction to credit risk modeling, Chapman and Hall/CRC Financial Mathematics Series Brage, V. and G. Eckerstrm, 2009, Leasing a comparative study of Swedish and Japanese retail firms, University essay from Gteborgs universitet Fretagsekonomiska institutionen [2009-03-12T10:30:41Z] Branson, B. C., 1995, An Empirical Reexamination of the Leasing Puzzle, Quarterly Journal of Business and Economics, 34:3, 3-18 Cherubini, U., E. Luciano, and W. Vecchiato, 2004, Copula methods in finance, New York: John Wiley and Sons Copeland, T. E. and J. F. Weston J., 1982, A note on the evaluation of cancellable operating leases, Financial Management, 11, 60-67 Drakos, K. and E. Goulas, 2008a, Irreversible investment under uncertainty: the message in leasing expenditures, International Journal of Monetary Economics and Finance, 1:4, 412-426 Drakos, K. and E. Goulas, 2008b, The relative importance of sector and country factors for leasing penetration rates across European industries, Applied Economics Letters, 15, 11971200 Fishbein, B. K., L. S. McGarry, and P. S. Dillon, 2000, Leasing: a step toward producer responsibility, INFORM report www.informinc.org Brick, I. E., W. Fung, and M. Subrahmanyam, 1987, Leasing and financial intermediation: comparative tax advantages, Financial Management, 16:1, 55-59 Gibson, M. S., 2004, Understanding the risk of synthetic CDOs, Federal Reserve Board. Finance and Economic Discussion Series Paper No. 36 Goodacre, A., 2003, Assessing the potential impact of lease accounting reform: a review of the empirical evidence, Journal of Property Research, 20:1, 4966 Guerrien, B. and O. Gun, 2011, Efficient Markets Hypothesis: what are we talking about? Real-World Economics Review, 56, 19-30 Hull, J. and A. White, 2004,Valuation of a CDO and an nth to default CDS without Monte Carlo simulation, Journal of Derivatives, 12, 8-23
Irotte, H., M. Schmit, and C. Vaessen, 2004, Credit risk mitigation evidence in auto leases: LGD and residual value risk, Working Paper. Retrieved from http://www.solvay.edu/EN/Research/ Bernheim/documents/wp04008.pdf Jost, A. and A. Franke, 2005, Residual value analysis, Paper presented at the 23rd International Conference of the System Dynamics Society Kichler, E. and P. Haiss, 2009, Leasing and economic growth evidence for Central and South Eastern Europe, Paper submitted for presentation at the 36th Annual European Finance Association, EFA, Meeting Krishnan, S. V. and C. R. Moyer, 1994, Bankruptcy costs and the financial leasing decision, Financial Management, 23:2, 31-42 Laurent, M. P. and M. Schmit, 2007, An empirical approach to estimate residual value risk and its interconnection with credit risk the case of automotive leases, Working Paper Lease Europe Website: http://www.leaseurope.org Levis M. and M. Lasfer, 1998, The determinants of leasing decisions of small and large companies, European Financial Management, 4:2 Lewis, C. M. and J. S. Schallheim, 1992, Are debt and leases substitutes? Journal of Financial and Quantitative Analysis, 27:4, 497-511 Li, D. X., 2000, On default correlation: a copula approach, Journal of Fixed Income, 9, 43-54 Mackie-Mason, J. K., 1990, Do taxes affect corporate financing decisions? Journal of Finance, 45:5, 1471-93 Marston, F. and R. S. Harris, 1988, Substitutability of leases and debt in corporate capital structures, Journal of Accounting, Auditing and Finance, 3, 147-70 Merton, R., 1974, On the pricing of corporate debt: the risk structure of interest rates, Journal of Finance, 29, 449-470 Nelsen R., 2006, An introduction to copulas, Second Edition, Springer, New York OConnor S. and G. Plimmer, 2011, Companies turn to leasing to combat credit drought, Financial Times September 20 Opening up economics: a proposal by Cambridge students, Manifesto, 2001, www.btinternet. com/~pae_news/Camproposal.htm Pirotte, H. and C. Vaessen, 2008, Residual value risk in the leasing industry: a European case, European Journal of Finance, 14:2, 157-177 Prado, S. M., 2009, Hedging residual value risk using derivatives, EconomiX Working Papers 2009-31, University of Paris West Nanterre la Dfense, EconomiX Prado, S. M., 2010, Macroeconomics of the new and the used car markets, Economics Bulletin, AccessEcon, 30:3, 1862-1884 Pykhtin, M. and A. Dev, 2003, Residual risk in auto leases, Risk, 16:10, 10-16 Smith, C. W. and M. L. Wakeman, 1985, Determinants of corporate leasing policy, Journal of Finance, 40:3, 895-908 Schmit, M., 2004, Credit risk in leasing industry, Journal of Banking and Finance, 28, 811-84 Schmit, M , 2004b, Is automotive leasing a risky business? Finance, 26, 35-66 Shojai, S. and G. Feiger, 2010, Economists hubris the case of risk management, Journal of Financial Transformation, 28, 25-35 Shojai, S. and G. Feiger, 2011, Economists hubris the case of award winning finance literature, Journal of Financial Transformation, 31, 9-17 Vershuere, B., 2006, On copula and their application to CDO pricing, Masters thesis, University of Chicago White, E., 2008, Global Leasing Report, White Clarke Group

Appendix
Our approach is inspired by the one factor standard Gaussian copula developed by Li (2000) and extended by Gibson (2004). Why do we use copula? In a portfolio, risks are non-independent. Copulas constitute a convenient tool to specify a joint distribution. Using a copula function, we are able to link the survival function of an obligor to the survival function of another obligor in a portfolio.
contract is qi(t). qi(t) is a variable with mean mFMVi, variance eFMVi, and distribution function Ei(mFMVi; eFMVi). The probability of loss depends on residual value: qi = Ei(Vi). So the default threshold xi is equal to Fi-1(qi). Conditional on the value of the sectorial factor M, the probability of default is therefore qi(M) = H-1[(F-1(qi) aiM)/(1-ai2)] The conditional probability is PN(l;M) = (lN) qi(M) and the unconditional probability can be calculated as
PN(l) = - PN(l) g(M)dM.
Homogeneous equipment type model

The initial idea is simple: we use the equipment resale value as the asset value-like xi and the probability of resale value below residual value (xi<xi) as the probability of default. In a reference portfolio of i = 1, ,N units (vehicles, equipment, etc.), for each obligor, losses occur when xi (reference unit normalized asset value) falls below xi (reference unit normalized residual value). xi = aiM + (1 - a2Zi
i
As a result, the recovery rate is equal to the probability of loss; by construction the recovery rate is R = qi. The loss on sale for any unit is (1-R) Lp. Finally, the resale price becomes RLp. Previous elements allow the creation of a financial product, inspired by CDOs:
The correlation of resale prices between units i and j is equal to aiaj. M is the sectorial factor affecting equipment units on the resale market and Zi is the risk of loss on resale on unit i xi, M, and Zi are mean-zero, unit variance random variables with distribution functions Fi(0;1), G(0;1), and Hi(0;1). The random variables are assumed to be independently distributed.
The expected loss is EL = N PN(l)*max(min(Lpl*(1 R), H) L, 0) i=0 The contingent leg is = n Di(EL) i=0 The premium leg is FEE = sn Di i[(H L) ELi] i=0 The spread is s = (Contingent) n Di i[(H L) ELi] i=0
At that point, the construction is similar to the credit model, but we include residual risk. Resale value can be lower than residual value, so there is a risk of loss on sale. Three new elements will have an impact on the leasing adjustment of the model. Vi is the residual value, or in other words the expected fair market value, mFMVi is the historical average fair market value, and eFMVi is the historical standard deviation. mFMVi, eFMVi, and Vi are set on a percentage of Lp, list price by unit. As an example, an asset bought for 10000 and leased for a residual value of 5000 has a Vi of 50%. V1i, V2i, V3i are residual values for groups 1, 2, and 3 Then the residual risk is added: probability of loss at the end of the 217 The construction is similar to the homogeneous equipment type model. Three representative vehicles constitute the model: Ex, Em, and Sm. Now there are different types of asset residual values, number of units, list prices, etc.
Heterogeneous equipment type model: a portfolio of three different assets

The model is extended to a portfolio with non-similar units. A company fleet is commonly constituted of various car models. In a European leasing contract for a medium-sized European company, the lessee usually requests different categories of cars in an auto lease contract. The fleet is usually divided into three groups: executive cars (usually high brand car), employee cars (medium level cars), and small cars.
mFMV1i, mFMV2i, and mFMV3i are fair market value historical averages. eFMV1i, eFMV2i, and eFMV3i are historical standard deviation historical averages. mFMV1i, eFMV1i, and V1i are set on a percentage of list prices. The recovery rate for group 1 is R1 = q1i, the loss on sale for any unit is R1(1 Lp1), with Lp1 unit list price. Indeed, the resale price is R1Lp1. The principle is the same for others groups. For each vehicle, the asset value is still xi = aiM + (1 - a2Zi
i
Gaussian may not be the most relevant copula for risk management. Other families of copula could be tested: the Student copula, the double t copula, Marshall-Olkin copulas, the Clayton copulas, etc. [Vershuere (2006)]. The modeling could also be developed in various ways: the accuracy of the correlation parameter could be improved by a complete macroeconomic analysis, and the fair market value parameter could also be improved on at the equipment level.
For group 1, default threshold xi is equal to Fi-1(qi) with q1i = Ei(V1i) and Ei(mFMV1i; eFMV1i). The principle is the same for other groups. The distribution of the number of defaults, conditional on the common factor M, is computed for each group as PN1(u; M) = (N )q1i(M), PN2 (v; M) = (N )q2i(M), PN3 (w; M) = ( N1)q3i(M), 1 1 with N1, N2, and N3 being the number of units, PN1, PN2, and PN3 being the conditional probabilities, and u, v, and w as number of defaults for group 1, 2, and 3. Default distributions are usually computed through a recursion method [Andersen et al. (2003), Hull and White (2004)]. In the case of homogeneous units and for simplification purposes, a binomial function is simpler and leads to similar results. The probability of default is computed on the whole portfolio: PN(u; v; w; M) = PN1(u; M) PN2 (v; M) PN3 (w; M) The expected loss is: EL = N PN (u; v; w; M) * max(min((Lp1u * (1 R1)) + ((Lp2u * (1 R2)) + l=0 ((Lp3u * (1 R3)), H) L, 0) The premium leg, contingent leg, and spread are calculated like in the homogenous equipment type model. It is straightforward to generalize this approach to more than three vehicle types. For simplification purposes, we focused on a bivariate distribution function and the two dimensional Gaussian copula. The previous results, 218 however, can be extended to the multivariate case [Nelsen (2006)]. The
u v w
PART 2
A Financial Stress Index for the Analysis of XBRL Data

Amira Dridi LARODEC, ISG, University of Tunis Silvia Figini Department of Social Sciences, University of Pavia Paolo Giudici Department of Economics and Management, University of Pavia Mohamed Limam LARODEC, ISG, University of Tunis
Abstract
In this paper we present a novel nonparametric approach to measure financial stress index to analyze financial data arising from balance sheets. We compare the results achieved with our approach with classical methods based on variance equal weight in a cross validation exercise via Monte Carlo simulation. Empirical investigations achieved on a real dataset show that our proposed approach provides a better performance in terms of error rate. 219
Recently, researchers have become interested in stress periods that measure the vulnerability of a financial system in order to take actions to avoid potential crises. Appropriate methodologies, such as financial stress testing [Alexander and Sheedy (2008); Committee on the Global Financial System (2005); Kupiec (1998)] and financial stress index (FSI) [Dridi et al. (2010); Cardarelli et al. (2009); Balakrishnan et al. (2009)] are used to determine stress intensity of a given system. The stress level is measured on a scale ranging from tranquil situations, where stress is quasi-absent, to extreme distress, where the system goes through a severe crisis. The interaction between the shocks magnitude and the systems fragility determines the stress level. One relevant measure of the latter is the FSI.
Our proposal
The method proposed by Dridi et al. (2010) could be improved by using a more general approach based on the EDF. Our approach considers the K EDFs computed on the quantitative variables X1,X2,...,XK. More precisely, each variable is transformed into perk centiles based on its sample EDF(Fn ),k=1,...,K. The transformed variables are then averaged using arithmetic or geometric means. We choose the geometric mean as a central tendency measure of averaging, since it provides important results in terms of risk ordering and comparison in stochastic dominance criteria approach as shown in Figini and Giudici (2010). In order to demonstrate why our proposal based on EDF is appropri-
Related indices have been developed by Hanschel and Monnin (2003) for the Swiss National Bank and by Illing and Liu (2006) for Canada. Cardarelli et al. (2009) computed an FSI for developed economies, and Balakrishnan et al. (2009) proposed an FSI for emerging countries and compared it with developed ones. Dridi et al. (2010) proposed an FSI based on extreme value theory (EVT) to determine stress periods in the Tunisian banking system. They propose to merge the information available on a measure called FSIj, j=1,...,n, which underlines the relative stress index for the j statistical unit, using a variance equal weights approach (VEW).
th
ate to compute FSI, we consider theoretical results present in the literature. First, we remark that by the strong law of large numbers, the EDF converges almost surely to the cumulative distribution function (CDF) (F) and the estimator EDF is consistent (see e.g. Van der Vaart, 1998); and secondly, that following the Glivenko-Cantelli Theorem the convergence happens uniformly over t, t R: ||Fn-F|| = sup|Fn(t) F(t)|0,t R. (2)
The latter computes the FSI first by standardizing the variables and then averaging the standardized scores using identical weights. More formally, the index formula, for the jth company is the following:
k FSIi=(i=1)
In our opinion, when using the EDF we have the possibility of testing the goodness of fit between the Fn and the assumed true EDF F via the Kolmogorov Smirnov statistic under the sup-norm assumption or equivalently Cramer von Mises statistic under the L norm. With respect to VEW
(Xi-Xi) i ,
(1)
approach, EDF offers a further characterization of the asymptotic distribution in different ways. First, the central limit theorem states that the pointwise estimator Fn(t) has asymptotically a normal distribution with the standard n rate of convergence: n (Fn(t) F(t))N(0,F(t)(1-F(t))). (3)
where k is the number of variables in the index, Xi the mean of the variable Xi, and i its standard deviation. Note that FSI assumes all real values. Also, Dridi et al. (2010) propose to monitor FSI based on an EVT control chart, namely the FSI-EVT control chart relying on EVT to determine a threshold capable of detecingt critical periods especially in the banking sector. The aim of this paper is to extend the methodology proposed by Dridi et al. (2010) to a different context. More precisely, on the basis of a real dataset composed of XBRL balance sheets, we propose to measure FSI for the financial data of small- and medium-sized enterprises (SME). From a methodological perspective, the novelty of the proposed approach is the introduction of a nonparametric FSI based on the empirical distribution function (EDF). We also provide a Monte Carlo computational procedure to compute and assess the Value at Risk (VaR) for FSI models using the EDF and VEW methods. Empirical evidences are achieved on the basis of financial indicators based on liquidity and debt information extracted from XBRL balance sheets. 220
These results could be extended by using Donsker theorem, which as serts that the empirical process n (Fn(t) F(t)) converges in distribution in the Skorokhod space D = ]-,[ to a mean-zero Gaussian process. On the other hand, another interesting implication is that the rate of con vergence of n (Fn(t) F(t)) can also be quantified in terms of the asymptotic behavior of the sup-norm via Dvoretzky-Kiefer-Wolfowitz inequality [Shorak and Wellner (1986)]. In practice by using EDF, it will be easier to derive a threshold on the resulting distribution of FSI. For example, using the 99% percentiles as extreme high level of stress for all variables involved in the analysis [Illing and Liu (2006)]. Furthermore, with respect to the VEW approach the transformed variables

are unit-free and implicitly reflect all the moments of their distributions without assuming normality derived standardizing data. Finally, the interpretation of quantiles based on EDF is a good approximation of the cumulative distribution function and possibly bootstrap techniques that rely on the probability integral transformation can be applied to the EDF to simulate the repetitive drawing of samples from a probability distribution which is known only through a sample. We finally note that since the elements involved in the averaging are nonnormalized quantities with values in [0,1], the corresponding FSI results in a bounded index with values in [0,1], differently from VEW method which takes all real values. From a theoretical perspective, we are interested in estimating the distribution function Fu of the values of FSI above a certain threshold u. The distribution function Fu, called the conditional excess distribution function, is given by: Fu (y) = P((X u y)\ X > u), 0 y x F u. (4)
Finally, in order to assess the goodness of the VaR itself we need to carry out a cross validation exercise based on a loss function as described below. The cross validation exercise will determine the best VaR model among VEW- and EDF-based VaR models. The best model will be the one that leads to the least loss. Loss functions are mostly used in predictive classification problems. We will mainly refer to this context here, for a comprehensive review, see Giudici and Figini (2009).
Empirical study
A real dataset composed of financial information about a set of SMEs is used to illustrate our proposed approach. We analyze 416 companies belonging to five different sectors, namely construction, industry, nonfinancial, agriculture, and service. Financial ratios are useful indicators of a firms performance and financial situation. Most ratios can be calculated from information provided by the financial statements. The most frequently used financial ratios are:
Liquidity ratio gives a picture of a companys short term financial situation or solvency. Leverage ratio shows the extent of debt used in a companys capital structure. Operational ratio uses turnover measures to show how efficient a company is in its operations and use of assets. Profitability ratio applies margin analysis to show the return on sales and capital employed. Solvency ratio gives a picture of a companys ability to generate cash and pay it financial obligations.
Following Embrechts et al. (1998), we focus on parametric peaks-overthreshold models based on the generalized Pareto distribution (GPD). Let X1,X2,...,XT be statistically independent variables drawn from the same distribution FX, then the exact distribution of the maximum MT is given by: FM (x) = P(MT x) = P(X1 x, X2 x,,XT x) = (FX (x))T.
T

(5)
We underline that the distribution of extremes depends mainly on the properties of FX for large values of x. Indeed, for small values of x the effect of FX(x) decreases rapidly with T. Hence, the most important information about the extremes is contained in the tails of the distribution of X. Following Balkema and De Haan (1974), Fu(y) can be approximated by the Generalized Pareto Distribution (GPD), defined by the location parameter , scale parameter , and shape parameter R as follows: 1 (x,,,) =
In this paper, we will focus on the first two categories of financial ratios. In general, liquidity ratios provide information about a firms ability to meet its short-term financial obligations. In this paper, we analyze liquidity current ratio and liquidity quick ratio. The liquidity current ratio is the ratio between current assets and current liabilities. Short-term creditors prefer a high current ratio since it reduces
(1+ ( x- )(- -1)),
(6)
their risk. Shareholders may prefer a lower current ratio so that most of the firms assets are working to grow the business. Typical values for the current ratio vary by firm and sector. For example, firms in cyclical industries, in order to remain solvent during downturns, may maintain a higher current ratio. One drawback of the current ratio is that inventory may include many items that are difficult to liquidate quickly and that have uncertain liquidation values. The quick ratio is an alternative measure of liquidity that does not include inventory in the current assets. The quick ratio is defined as the ratio between the current assets minus the inventory and current liabilities. 221
where x and x (x- )/, < 0. In order to estimate the parameters for the FSI and on the basis of the above distribution we need to employ techniques to estimate the VaR. We randomly split the available data into two subsamples of training and testing sets. The training sample is used for model fitting and parameter estimation and the predictive effectiveness of the fitted model is evaluated in terms of the number of times that the losses are greater than the VaR.
On the other hand, financial leverage ratios provide an indication of the long-term solvency of the firm. Unlike liquidity ratios that are concerned with short-term assets and liabilities, financial leverage ratios measure the extent to which the firm is using long term debt. In this paper, we focus on the debt ratio, measured as total debt divided by total assets, and the debt to equity ratio, defined as total debt divided by total equity. Debt ratios depend on the classification of long-term leases and on the classification of some items as long term debt or equity. The first analysis considers FSI computation based on the financial ratios at hand. For this purpose, we compare two univariate models namely the VEW and EDF models, as explained above. Two FSIs are computed, FSI for liquidity which is composed of liquidity current ratio and liquidity quick ratio. The FSI for debt is composed ofthe debt ratio and the debt to equity ratio. We have estimate the GPD parameters for the FSI based on the VEW model. The results for are 0.3653 and 0.2009 for liquidity and debt ratios respectively. Furthermore, the estimates for are 1.5302 and 1.0890. In the parametric framework, the shape index gives an indication of the heaviness of the tail. Results indicate that FSI computed with the VEW method has a positive shape index, meaning that the distribution of FSI is a reparametrized version of the ordinary Pareto distribution. The latter has a long history in actuarial mathematics as a model for large losses. Once we have GPDs parameter estimations of the FSI for liquidity and debt, we proceed with the Monte Carlo simulation in order to estimate the VaR for different risk levels, as given in Table 1. Table 1 reports the VaR computed with respect to three different percentiles. We notice from Table 1 that the scales of the two results are different, because VEW gives real values for the FSI, while EDF gives values of between 0 and 1. To validate the models, we use the cross validation as backtesting tool for the VaR found by the Monte Carlo simulation. With 5,000 iterations we calculate the error rate as the number of times the FSI is greater than the VaR over the sample size and results are summarized in Table 2.
FSI Liquidity Liquidity Liquidity Debt Debt Debt
VaRs Level 1% 3% 5% 1% 3% 5%
VaR-VEW 22.3745 12.3502 9.2278 8.3708 5.5521 4.3747
VaR-EDF 0.8207 0.6236 0.6114 0,9009 0.8783 0.8613
Table 1 VaR estimation results for VEW and EDF for liquidity and debt
FSI Liquidity Liquidity Liquidity Debt Debt Debt
VaRs level 1% 3% 5% 1% 3% 5%
Error rate-VEW 1.2% 3.11% 5.02% 1.2% 3.11% 5.02%
Error rate-EDF 0.96% 3.11% 5.02% 0.96% 3.11% 4.55%
Table 2 Error rate: comparison between VEW measured in standard units and EDF measured in percentage points for liquidity and debt
1% 3% 5%
Error rate construction 11.8% 12.7% 12.4%
Error rate industry 28.03% 27.39% 26.75%
Error rate non-financial 16.67% 16.67% 16.67%
Table 3 Error rate for business sectors using VEW results for debt
1% 3% 5%
Error rate construction 1.03% 3.08% 5.13%
Error rate industry 1.27% 3.16% 5.06%
Error rate non-financial 0.00% 5.56% 5.56%
Table 4 Error rate for business sector using EDF results for debt
Tables 3 and 4 report the error rate derived for the sectors considered in Table 2 provides a clear overview of the cross validation results achieved by our approach. On the basis of the error rate, it shows that FSI computations based on the EDF method perform better than VEW, reaching an error rate minimum of 0.96%. The two approaches are not that different for the liquidity ratios, in terms of goodness, for specific values of equal to 3% and 5%. Considering debt ratios, EDF performs better than VEW, taking a value of equal to 1% and 5%. Finally, we have considered the described approach using the sectorial information. We have selected SMEs in the following sectors: construction, industry, and non-financial. For each sector, on the basis of the debt 222 ratios, we have estimated the VaR and we have computed the error rate. the analysis using the VEW and EDF approaches, respectively.
Conclusions
In this paper we have introduced a novel nonparametric approach to compute financial stress indicators. We believe that our approach could help financial institutions and rating agencies to analyze, in a coherent way, financial data arising from balance sheets. In the paper we have tested what was proposed on a real dataset. Both Monte Carlo-based simulations and cross-validation performance measures have been applied to the data at hand. The obtained evidence shows that our proposed method performs better than classical models, used in current best practices.

References

Alexander, C. and E. Sheedy, 2008, Developing a stress testing framework based on market risk models, Journal of Banking and Finance, 32:10, 2220-2236. Balakrishnan, R., S. Danninger, S. Elekdag, and I. Tytell, 2009, The transmission of financial stress from advanced to emerging economies, IMF Working Paper Balkema, A. A. and L. De Haan, 1974, Residual life time at great age, Annals of Probability, 2, 792-804 Beirlant, J., J. Teugels, and P. Vynckier, 1996, Practical analysis of extreme values, Leuven University Press, Leuven Cardarelli, R., S. Elekdag, and S. Lall, 2009, Financial stress, downturns, and recoveries, IMF Working Paper Committee on the Global Financial System, 2005, Stress testing at major financial institutions: survey results and practice, CGFS Publications, No. 24 Danielsson, J., P. Hartmann, and C. De Vries, 1998, The cost of conservatism, Risk, 11:1, 101-103 Dridi, A., M. El Ghourabi, and M. Limam, 2010, On monitoring the financial stress index with extreme value theory, Quantitative Finance, 0:0, 1-11, ahead of print Embrechts, P., S. Resnick, and G. Samorodnitsky, 1998, Living on the edge, Risk Magazine, 11:1, 96-100 Frye, J., 1997, Principals of risk: finding value-at-risk through factor-based interest rate scenarios, NationsBanc-CRT Giudici, P. and S. Figini, 2009, Applied data mining for business and industry, Wiley Glasserman, P., P. Heidelberger, and P. Shahabuddin,2000, Efficient Monte Carlo methods for value at risk, Working Paper, Columbia University Hanschel, E. and P. Monnin, 2003, Measuring and forecasting stress in the banking sector: evidence from Switzerland, BIS Papers, 22, 431449 Illing, M. and Y. Liu, 2006, Measuring financial stress in a developed country: an application to Canada, Journal of Financial Stability, 2:3, 243-65 Jamshidian, F. and Y. Zhu, 1997, Scenario simulation: theory and methodology. Finance and Stochastics, 1, 43-67 Kupiec, P., 1998, Stress testing in a value at risk framework, Journal of Derivatives, 6, 7-24 McNeil, A. J., 1999, Extreme value theory for risk managers,in Internal modelling and CAD II, Risk Books, 93-113 Pickands, J., 1975, Statistical inference using extreme value order statistics, Annals of Statistics, 3, 119-131 Shorak, G.R. and J. A. Wellner, 1986, Empirical processes with applications to statistics, Wiley Van Der Vaart, 1998, Asymptotic statistics, Cambridge University Press, Cambridge.
223
Layout, production and coordination: Cypres Daniel Brandt, Kris Van de Vijver and Pieter Vereertbrugghen Graphic design: Buro Proper Bob Goor Photographs: Bart Heynen 2012 The Capital Markets Company, N.V. All rights reserved. This journal may not be duplicated in any way without the express
224
written consent of the publisher except in the form of brief excerpts or quotations for review purposes. Making copies of this journal or any portion there of for any purpose other than your own is a violation of copyright law.
Cass Consulting
At Cass we believe that the purpose of research lies in its real-world application, which is why our consultants all have extensive hands-on experience in their chosen fields. Cass Consulting offers an unrivalled portfolio of expertise, services and solutions to enable you to grow and strengthen your business whether on a domestic or global scale. What sets us apart is being able to call on a range of insights and provide bespoke independent advice for any challenges you may face. To find out more about how Cass Consulting can help you, contact Dr Christina Makris at christina.makris.1@city.ac.uk, call on +44 (0)20 7040 3273 or visit www.cassknowledge.com
www.cassknowledge.com
amsterdam antwerp Bangalore Bratislava Chicago Frankfurt Geneva Johannesburg London New York Paris San Francisco Toronto Washington, D.C. Zurich
CaPCo.Com

Risk Journal 34

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Risk Journal 34

Încărcat de

Drepturi de autor:

Formate disponibile

Journal

The Capco Institute Journal of Financial Transformation

Recipient of the Apex Awards for Publication Excellence 2002-2011

Cass-Capco Institute Paper Series on Risk

Cass-Capco Institute Paper Series on Risk

IT Complexity: Model, Measure, Manage, and Master

IT complexity: a new metric to evaluate the impact of change

The power of quantitative metrics

The Capco Institute Journal of Financial Transformation

How a complexity metric fits into the CIO agenda

the following examples will be familiar to any CIO:

Consider IT complexity to make more informed decisions

Functional and interface complexity error analysis and resolution is another

Data complexity when implementing

Complexity metric - quarterly aggregated report

Number of weighted interfaces (QII1)

Infrastructure Requirements (QT1)

Percental coverage of QT1

Number of IT products per functional group/ cluster

Ratio QFI1 / QFB1: Implementation Quotient in percent

Architecture Cluster Functional Group

Figure 2 Complexity dashboard, a sample view

The Capco Institute Journal of Financial Transformation

How do you measure complexity?

Immediate application and benefits

partnerships and create new opportunities to achieve measurable results.

Expansions and possibilities

IT Complexity Metrics How Do You Measure Up?

Introduction and background

Commerzbank, with Capcos input, models complexity at the moment.

Each indicator is defined on the basis of a hypothesis advanced by Commerzbank and

improved, becoming more generally reusable within the industry.

FC FO FLU FU FUD FIS FIO FSC FDD

IIF II IIO IIT IEFR IER

DIO DBO DIO

The Complexity Model in detail

TIR TIP TIS

Table 1 Overview of complexity indicators

The Capco Institute Journal of Financial Transformation

the source of real, holistic insight). Technology

The Capco Institute Journal of Financial Transformation

#IT pro DQ ducts 73 231 138 44 8 21 25 40 162 243 21 124 992

5% 5% -6% -6% -10% 0 -6% 6% -20% 2% 6% -4% 7% 1%

6% -4% -4% -6% 0.01 0.01 0.01 -12% 6% 0.03 5% 0 -2%

6% -6% -3% -2% 0.01 0.01 0.1 -1% 6% 0.03 5% 0 -2%

20,646 8% 53,021 -1% 0 0 23,113 8% 6,579 12% 6,181 4% 2,091 8% 2,102 0

Transaction Processing Group Steering Data Warehouse Corporate services Total

8,420 3% 10,532 -6% 19,592 0.08

6,160 7% 48,301 0.01 26,522 -6% 19,046 0

8,706 -9% 0 0 199356 -8% 18 0

Interfaces complexity for business objects

IER IEFR IIT II IIF

IER IEFR IIT II IIF

Diversity Interdependency Diversity Interdependency

Figure 1 Commerzbank IT complexity landscape

Toward a Bottom-Up Approach to Assessing Sovereign Default Risk: An Update

The Capco Institute Journal of Financial Transformation

Modern history sovereign crises

140 120 100 80 60 40 20 0

Syndicated bank loans Sovereign bonds All

Figure 1 Number of sovereign defaults (1824-2004)

What do we know about predicting sovereign defaults?

The Capco Institute Journal of Financial Transformation

Euro -1.87 (0.50) 0.043 (0.014) 0.41 39

Swiss franc -1.85 (0.40) 0.130 (0.048) 0.44 39

Euro -3.17 (0.95) 0.073 (0.025) 0.39 39

Swiss franc -3.53 (0.99) 0.252 (0.097) 0.46 39

time zones, tzi , draw$i = b1 + b2s$i + b3dilfc + b4*tzi + i (7)