The Impact of Collateralization on Swaps Rates Under Clearing

Freddy A. Rojas Cama

Rutgers University

November 21, 2017

[Last version available here]

Abstract

I study the relationship between the price of derivatives and clearing practices in a the-
oretical framework. Specifically, I setup this connection by measuring the total exposure
(the loss upon default of a contract) registered in a clearinghouse and its respective amount
of collateral requirement. I find that netting through novation has significant gains in re-
ducing exposures and therefore making a clearinghouse more competitive in terms of prices
and collateral requirements thus clearing turns out to be appealing to more participants.
Additionally, I also find above gains are large when comparing a financial structure of one
clearinghouse respect to other with two specialized clearinghouses. In the case of interest rate
swaps I find a relationship between netting and the Libor rate that may potentially aect the
dierence in prices among clearinghouses; in other words, a linear correlation calculated over
time-series data and term structure seems to validate the appearance of a widening basis -a
price dierential - between London Clearing House and Chicago Mercantile Exchange.

Keywords: Clearinghouse; price discovery; financial economics; Interest rate swaps; credit
default swaps; collateralization; netting; novation.
JEL codes: G10; G12; G13.

I want to thank Bruce Mizrach for his advice and support throughout this project. I acknowledge Todd
Keister for comments and suggestions to early versions of this paper. I also acknowledge Roberto Chang, John
Landon-Lane, Ryuichiro Izumi and all participants in the macroeconomic theory workshop at Rutgers University
for all their comments. All remaining errors are mine. Electronic address: frojas@economics.rutgers.edu

1
Contents

1 Introduction 3

2 Literature review 7

3 A model of swap determination under clearing 8

3.1 Determination of the interest-rate swap rate . . . . . . . . . . . . . . . . . . . . . 10

3.1.1 Swap rate with no collateralization . . . . . . . . . . . . . . . . . . . . . . 11

3.1.2 Swap rate with collateralization . . . . . . . . . . . . . . . . . . . . . . . . 11

3.1.3 Dierence in swap rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.2 Determination of the credit default swap premium . . . . . . . . . . . . . . . . . 15

3.2.1 Trading Frictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2.2 Bilateral arrangement: Payos by state of nature . . . . . . . . . . . . . . 19

3.2.3 Clearing arrangement: Payos by state . . . . . . . . . . . . . . . . . . . 22

3.2.4 Dierence in premium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Calculation of the exposure in clearinghouses 26

4.1 Assumptions of the joint distribution among assets . . . . . . . . . . . . . . . . . 27

4.2 Assumption about heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.3 Expectation of exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Analysis and discussion 29

5.1 The dierence in CDS premiums . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.2 The basis between LCH and CME . . . . . . . . . . . . . . . . . . . . . . . . . . 32

6 Conclusions 36

A Details of derivations 38

2
 1 1 Introduction

2 This paper contributes to understand the interplay between price of derivatives and clearing-
3 houses, relationship that is still not well understood in literature. In the light of the recent
4 financial crisis and its aftermath, clearing of derivatives has become central to the modern fi-
5 nancial system. In practice, netting of positions and other clearing procedures question the
6 standard practice of valuation aected by hedging and collateralization. Particularly, I explain
7 in a simple theoretical framework how much the swap rate is aected by clearing practices. I
8 thoroughly examine two type of swaps contracts: interest-rate and credit default swaps.

9 The traditional approach for the valuation of swaps1 uses information about the current
10 market conditions such as liquidity, supply-demand factors, and spreads between short-term
11 repo rates. More important, latest literature includes the eect of the default risk of multiple
12 counter-parties on these contracts (see Leung and Kwok [23], Johannes and Sundaresan [20] and
13 Duffie and Zhou [12]); it is mostly accepted that credit worthiness of counter-parties significantly
14 aects the fair-market swap rates. Literature seems to deliver a good understanding of the
15 association between the swap rates and its most pertinent underlying factors. However, other
16 factors that are more related to the structure of the financial market remain a pending subject in
17 the literature of swap valuation; financial regulation and their eects on asset pricing are hardly
18 formalized in the literature. In the light of the recent financial crisis and its aftermath, dynamics
19 of swaps spreads have recently received particular attention not actually being observed since
20 the end of nineties2 . The Dodd-Frank Act, a key piece of financial reform legislation passed by
21 Obama administration in 2010, opens the discussion regarding the eect of clearing practices
22 on price of trading derivatives; clearing is becoming an interesting proposal inside the recent
23 regulatory framework. However, the link between price and clearing practice is still not well
24 understood. This paper intends to fill out this gap in the literature.

25 Interest rate and credit default swaps are contingent claims that are massively traded in
26 clearinghouses. An interest rate swap (IRS) is a derivative contract through which two parties
27 exchange fixed and floating rate coupon payments; usually literature presents the structure of
28 swaps as simply aected by credit worthiness, the LIBOR rate and the spreads over repo (see He

1
A swap -that is a class of derivative- is a contract between parties whose value is based on an underlying
financial asset, index, or security. Source: Investopedia
2
Around those years, noticeable volatility over these contingent claims contributed to the financial turmoil
that led the US Federal Reserve to modify the path of interest rates (see He [14])

3
29 [14]). On the other hand, a credit default swap (CDS) is another derivative contract whereby
30 the buyer seeks protection from the loss arising from a credit event. In exchange, the seller
31 (typically a financial institution) absorbs the risk of arranging the conditional payment once
32 the credit event occurs. In the year 2016, according to Bank of International Settlement (BIS),
33 the market for interest rate swaps reached the notional3 value of 275 US trillion dollars, while
34 10 US trillion dollars of credit default swaps were negotiated in the same year (see figure 1a).
35 Thus, these two assets comprise around 90% of the total market value of derivatives. The most
36 important clearinghouses are Chicago Mercantile Exchange (CME) and London Clearing House
37 (LCH); their trading-volumes shares in 2016 reached to 10 and 82 percent respectively (see figure
38 1b). Recent analysis of empirical data from clearinghouses show noticeable spikes or persistent
39 unusual behavior of the trading swap rate that raise questions regarding the dynamics of the
40 price determinants of these financial instruments. Specifically, the basis -that shows dierence
41 of swap rates traded at two clearinghouses- should not theoretically show a significant wide size
42 since the instrument and its related characteristics as maturity and risk are the same.

(a) Outstanding market value (b) Clearing participation

Figure 1: Swaps Clearing Market

43 The clearinghouse does the settlement of any contract when novation takes place. Novation
44 is the inherent feature of the clearing process: the clearinghouse is the seller for any buyer and

3
In swaps, interest payments are computed based on a notional amount, which acts as if it were the principal
amount of a bond, hence the term notional principal amount, abbreviated to notional.

4
45 the buyer for any seller. Also, the clearinghouse establishes initial and variation margins or
46 collateral on contracts as well as collects default funds for mutualizing losses among market
47 participants. The two most common forms of collateral are cash and treasuries bonds since
48 they are default-free, they can easily be invested or loaned out4 (Johannes and Sundaresan [20];
49 ISDA [17]). Since posting collateral is generally costly, these payments induce economic costs
50 (benefits) to the payer (receiver). More significantly, the clearinghouse osets positions among
51 participants by performing compression and multilateral netting; precisely I study in this paper
52 the eects on swap rates of performing netting among all positions hold by participants in the
53 clearinghouse. The result of osetting positions is called exposure which needs to be hedged
54 by imposing collateral. Thus, the swap rate can be expressed as a function of the exposure
55 ultimately.

56 I explain in the next lines the relevance of studying the price determination of derivatives
57 in clearinghouses beyond the standard price discovery5 . The Dodd-Frank Act is the current
58 regulation framework for trading financial instruments and constitutes the most comprehensive
59 set of mandatory limits, exceptions and rules made by US government since great depression6 .
60 The magnitude of the last financial crisis made the previous administration to react properly by
61 proposing a regulation framework for the financial system. However, a potentially mandatory
62 regulation of derivatives poses a new mechanism to take into account that would potentially
63 aect price of derivatives; clearinghouses through novation can see positions of participants
64 and reduce significantly requirement of collateral by applying multilateral netting over class of
65 assets. The impact on prices can be significant and financially beneficial in comparison to other
66 proposals that relies on rising higher capital requirements as macro-prudential policies suggest
67 nowadays7 or other that may inherently be associated to spillovers such as agency problems

4
The most popular form of collateral is cash. In 2005 ISDA indicated that US dollars and euros cash accounted
for 73% of collateral assets.
5
The pricing of the swap or price discovery is a method of determining the price for a specific commodity or
security through basic supply and demand factors related to the market.
6
The preliminary Glass-Steagall Act was passed by the United States Congress on February 27, 1932, prior
to the inclusion of more comprehensive measures in the Banking Act of 1933, which is now more commonly known
as the Glass-Steagall Act. Source: Wikipedia.
7
See Tobias [31] for a quick refresh of macro-prudential policies and challenges. On the other hand, Ji-
had Dagher and Tong [19] assess the benefits of bank capital in terms of resilience; authors found that a high
capital requirement around 15-23 percent of risk-weighted assets would have been sufficient to absorb losses in
the majority of past banking crises. The basel rules and further details of extension III can be found in for Inter-
national Settlements [13], of International Settlements [25] and of International Settlements [26]. Other initiative

5
68 (see Chami et al. [4]8 ). Change in swap prices are generally associated to variations in the
69 implicit risk as standard theory predicts, but under an eective clearing practice i.e. netting,
70 the determination of swap prices needs a dierent and suitable framework to analyze.

71 The contribution of my paper in policy terms is as follows. First, trading in a clearinghouse

72 could make prices of derivatives competitive enough in comparison to bilateral agreements or
73 other specialized framework, for instance, the ones supporting more than one clearinghouse.
74 Moreover, other clearing practices as mutualization of losses among participants may reduce
75 further the costs of default. Second, price arbitrage -in a financial structure that allows multiple
76 clearinghouses- would produce shifts in the direction of the demand for a particular class of
77 assets among clearinghouses. Whether the price depends of the eort of reducing the overall
78 exposure, then a clearinghouse that treats risk properly would be efficient. A clear evidence of
79 the former is the behavior of the basis; two swap contracts may have exactly the same features
80 and they will probably be priced dierently due to dierent costs of funding in general. Finally,
81 in this setup, price depends on the regulatory policy; this relationship may aect the decisions
82 of agents in participating in clearinghouses.

83 The paper is organized as follows. The following section presents the literature related to
84 valuation of swaps and the progress made so far by including counter-party risk in the valuation
85 models. The third section explain the steps for achieving an analytical expression for the interest
86 rate and credit default swap as a function of the exposure under a clearing arrangement; I also
87 make a comparison of this exposure with bilateral agreements that currently are negotiated in
88 over-the-counter markets. The fourth section explains how to calculate the exposure using data,
89 I further explain the respective assumptions behind the formulas. The fifth section explains
90 the quantitative exercise in order to provide insights regarding what drives the dierence of
91 swap rates between clearinghouses. The calibration of parameters in the exercises is thoroughly
92 explained and discussed; the baseline values are mostly taken from recent literature. The last
93 section gathers the conclusion of this research and provide further questions to pursuing in a
94 future research.

is The U.S. House of Representatives passed the Financial CHOICE Act (FCA), it was put forward in 2016 by
the House Financial Services Committee, and it comprises key elements of the original DFA, leaving certain other
DFA elements out. According to Chami et al. [4], a key argument is the introduction of a regulatory o-ramp,
thus providing a relief in demanding capital requirements for so-called qualifying bank holding companies.
8
Even clearing -via mutualization of losses among participants- does not circumvent the usual problem of
commons associated to public goods, see Stephens and Thompson [30] and Cama [2] for a discussion.

6
95 2 Literature review

96 An extensive literature has developed that studies price determination for derivative contracts.
97 In the case of the credit default swaps, Leung and Kwok [23] and Jarrow and Yu [18] analyze the
98 eects of a change in the joint probability of default on spreads (prior to maturity). Duffie and
99 Singleton [8, 9] developed a methodology that derives reduced-form models of the valuation of
100 contingent claims subject to risk; more interesting, this paper introduces the eect of dierent
101 recovery rates on swap valuation. In a dierent approximation Acharya and Bisin [1] show -using
102 a theoretical model- the eect of releasing information on CDS premium; this analysis would
103 be equivalent to the eect of a explicit clearing method -or transparency of positions- on price
104 of derivatives. On the same approach, Stephens and Thompson [30] study price competition
105 with dierent type of insurers and show that mutualization may increase counterparty risk as
106 responsible insurers leave the market. Other contribution for price determination is found in
107 Koeppl [22]; in order to extracting benefits the contracts will rise prices in current contracts
108 under a clearing mechanism that not involves fulfillment of promises.

109 Collateral requirement is the cornerstone in the lender-borrower literature. Johannes and
110 Sundaresan [20] study the use of marking-to-market (MTM) and collateralization on swap rates
111 as they modify the flow of cash in the contract whereas risk of default rises. Johannes and
112 Sundaresan [20] derives an analytical expression where MTM and time-varying net costly col-
113 lateral alter the discount factor. The authors discusses limitations in evaluating the importance
114 of collateral. Precisely, comparing market swap rates with a par representation -constructed
115 from LIBOR bond prices- would be misleading since the par representation needs the market
116 swap rates. Finally, Johannes and Sundaresan [20] setups a zero-coupon structure made from
117 eurodollar futures, a strategy that does not require assumption regarding collateralization and
118 counter-party credit risk. On other hand, Duffie et al. [11] show, using pre-reform exposure
119 data set, that demand of collateral is increased significantly by the application of initial margin
120 requirements even if CDS are cleared or not. Most importantly, Duffie et al. [11] state that
121 mandatory central clearing is shown to lower collateral demand only when there is no significant
122 proliferation of clearinghouses. Duffie et al. [11]s work is closely related to Heller and Vause
123 [15], Sidanius and Zikes [29] and Johannes and Sundaresan [20] that simulate exposure data.
124 I extend their work establishing a measurable and theoretical relationship between exposure,
125 collateral needs and swap rates.

7
126 The clearinghouses can potentially reduce the size of the exposures through netting when
127 novation is performed. Specifically, Duffie and Zhu [10] and Cont and Kokholm [6] show the
128 tradeo between multilateral and bilateral netting when a particular asset from individual port-
129 folio is moved to central clearing reducing the net exposure calculated among trading partners
130 and assets. Duffie and Zhu [10] assume normality and no correlation between assets when clear-
131 ing, the foregoing assumption is relaxed by Cont and Kokholm [6] as showing that number of
132 optimal participants for reduction of the exposure is significant low. Also Cont and Kokholm
133 [6] shows that dierent methods do deliver dierent sizes of the credit exposure; however take
134 into account a particular distribution seem to be irrelevant when comparing results against a
135 gaussian distribution. Other important result in Cont and Kokholm [6] shows the exposure is
136 shrank significantly for cleared interest-rate swap contracts.

137 The empirical treatment of identifying the determinants of swap spreads is mostly standard
138 and relies on term structure models. For instance, He [14] uses a multi-factor term structure
139 framework assuming swaps are default-free and show that this structure is driven by market
140 expectations, risk premium and liquidity dierentials. Johannes and Sundaresan [20] model
141 the short rate using a ad-hoc two-factor model from Collin-Dufresne and Solnik [5]; previously
142 this author used calibration of models9 to compute hypothetical swap rates assuming swaps are
143 priced as a portfolio of forwards or futures. Thus, any dierence between actual swaps would
144 be attributed to collateral or margin strategies.

145 3 A model of swap determination under clearing

146 In this section I determine the swap rate for the exchange of flows between counter-parties
147 i.e. interest rate swap contract, and for the insurance contract signed among participants in the
148 credit default swap market. In each contract, by using novation, the clearinghouse nets positions
149 and calculates the size of collateral required to hedge them. The first two subsections deal with
150 the determination of the swap rate as a function of the size of the exposure under clearing. The
151 last section constructs the size of the exposure following the method developed in Duffie and
152 Zhu [10] and Cont and Kokholm [6].

153 The process of netting exposures in central clearing of OTC trades can lead to a decrease in

9
Authors use some ad-hoc adjustments following the treatment in Vasicek [34] and John Cox and Ross [21]
as well as calibration procedures made in Hull and White [16].

8
154 the sum of total bilateral exposures. For instance, as shown in Cont and Kokholm [6], consider a
155 market of four participants (A,B...) with bilateral exposures like shown in the left panel of figure
156 (2). In this market the sum of bilateral exposures amounts to 350. Introducing a clearinghouse
157 (or CCP) enables netting of the exposures across all participants which reduces the total net
158 exposure to 180.

(a) Bilateral exposure (b) Centralization

Figure 2: How clearing works

159 The arrows above represent the positive or negative (net bilateral) position of the contract
160 among participants; the direction of the arrow may be read as ... has an exposure to...
161 and amount above them indicates the size of the exposure. The above representation of a
162 trading market highlights the benefits of multilateral netting through a clearinghouse; positions
163 are oset among participants since clearinghouse centralizes them in any direction. However,
164 there are also situations where central clearing of a single asset class may actually increase
165 overall net exposures. For instance, Duffie and Zhu [10] and Cont and Kokholm [6] show that
166 multilateral clearing may actually increase the size of exposure when number of participants
167 are fewer enough. Besides, given a fixed number of participants, the size of exposure -according
168 to Cont and Kokholm [6]- would be highly sensitive to the assumptions of correlation and
169 distribution amid asset classes. In other words, performing novation is not a guarantee of an
170 eective reduction of the exposure.

9
171 3.1 Determination of the interest-rate swap rate

172 In this section I follow the theory behind the determination of the interest rate swap rate as
173 shown in McDonald [24] and Johannes and Sundaresan [20]. Specifically, I provide a swap valu-
174 ation theory under marked-to-market10 and costly collateral and examine the theorys empirical
175 implications. A interest rate swap (henceforth swap) is a contract calling for an exchange of
176 payments over time. Specifically, companies use swaps to modify their interest rate exposures.
177 Thus, the swap makes payments -under contract- as if there were an exchange of payments
178 between a fixed-rate and a floating-rate bond. For instance, a fund manager might own floating-
179 rate bonds and wish to have fixed-rate exposure while continuing to own the bonds. Thus,
180 investors may change the structure of payment flows and hedge risk whether their balance sheet
181 face uncertainty or just they engage into this asset market due to just merely speculative reasons.
182 A swap generally has less credit risk than the bond in reference since only net swap payments
183 are at risk whereas the principal is not. Figure (3) illustrates the cash flows for a company that
184 borrows at LIBOR and swaps to fixed-rate. If one party defaults, it owes to the other party at
185 most the present value of net swap payments at current prices.

Figure 3: Cash flows for company swapping rates

186 Hedging counter-party risk makes the market value of the swap an important variable under
187 consideration. At inception of the swap the market value is zero, meaning that either party could
188 enter or exit the swap without having to pay anything to the other party. Once the flows are
189 eectively interchanged, however, its market value will generally no longer be zero. McDonald
10
Mark to market is a measure of the fair value of accounts that can change over time, such as assets and
liabilities. This measure aims to provide a realistic appraisal of an institution current financial situation. Source:
investopedia.

10
190 [24] mentions at least two reasons for the foregoing: first, forward and zero-coupon bonds rates
191 change over time, and second, once the first payment is made there would be a dierence of new
192 swaps relative to the forward rate; hence, in order to exit the swap this counter-party needs to
193 be either compensated or aected by a convenient fee (see McDonald [24] for details).

194 3.1.1 Swap rate with no collateralization

I carefully examine the calculation of the swap rate under regular conditions i.e. no collateral-
ization. I assume T swap settlements occurring on dates ti , i = 1, ..., T . The floating interest
rate from date ti 1 to date ti known at date 0 is r0 (ti 1 , ti ) and the swap rate is denoted by sw .
The price of a zero-coupon bond maturing on date ti is P (0, ti ). Following McDonald [24], the
requirement that the swap have a zero net present value is:
T
X
P (0, ti )[sw r0 (ti 1 , ti )] = V0 0
i=1

195 Above expression can be rewritten for an easy interpretation (see expression 1); this preliminary
196 result will help later to construct a swap rate that incorporates collateralization under clearing
197 as a consequence of counter-party risk.
T
" #
X P (0, ti )
w Pn
198 s = r0 (ti 1 , ti ) (1)
j=1 P (0, tj )
i=1

199 The only risk in this transaction is associated to the uncertainty of the floating rate; this risk
200 is hedged by entering into a forward rate agreement. This is a result of netting the payment on
201 forward and an unhedged net swap payment (see details in McDonald [24]).

202 The expression in square brackets sum to one; thus, the meaning of this expression dierently,
203 the (fixed) swap rate is a weighted average of the implied forward rates where zero-coupon bond
204 prices are used to determine the weights. An alternative and popular expression can be derived
205 using the fact that the implicit forward rate can be calculated from bond prices, see McDonald
206 [24] for details.

207 3.1.2 Swap rate with collateralization

208 The trading of swaps or options in the over-the-counter (OTC) market can create counter-party
209 credit exposures; the party that was in-the-money would have to replace the deal at current
210 market prices. Thus, the positive MTM value is a credit exposure. One way to reduce the

11
211 credit risk is to use a break clause i.e. replacement of the whole contract and a final payment
212 must be made each period, and then the parties can enter a new contract the following period.
213 However, there exist the option of collateral management. Basically, collateralization involves
214 that party with the negative MTM on the trade portfolio delivers collateral to the party with
215 the positive MTM. As prices move and new deals are added the valuation of the trade portfolio
216 will change.11

217 First, I start with a motivation by using a discrete-time model based on Duffie and Singleton
218 [9] and Johannes and Sundaresan [20] and then I formally extend the model to continuous-time
219 including costly collateral. I defined the contract from the side of the counter-party that holds
220 the fixed-rate leg and I assume that counter-party who holds the floating-rate leg is potentially
221 subject to defaulting; in terms of van Egmond [33] the marked-to-market value of a bilateral
222 swap is negative to the agent who holds the fixed-rate leg, therefore it has to post collateral.

223 Formally, I consider a defaultable contract that exchanges fixed and floating interest rates.
224 h is the conditional probability under a risk neutral probability measure Q of default between
225 periods; it is defined by a set of information given a state of nature of non-default at each period
226 i.e. I assume that this probability is constant for sake of simplicity12 . The amount of collateral
227 posted is denoted as c' where ' denotes the exposure and c is the fraction of the exposure that
228 is collateralized. I assume the foregoing -for sake of tractability of the solution- is equivalent to
229 the fraction s of the market value of the contract at time s (Vs ), see Duffie and Singleton [9].
230 In Johannes and Sundaresan [20] the amount of collateral is assumed to be free of default13 .
231 The discount factor in the contract is the default-free short rate (rs ). If the contract has not
232 defaulted by time t its market value Vt would be the present value of receiving 't+1 in the event
233 of default between t and t + 1 plus the present value of receiving Vt+1 in the event of no default,
234 this is as follows,
235 Vt = he rt
EtQ (c't+1 ) + (1 h)e rt
EtQ (Vt+1 ) (2)

236 where EtQ denotes expectations under a martingale measure conditional on information available
237 at period t. Above expression represents a joint distribution between ' and the discount factor
238 (r) over various horizons. According to Duffie and Singleton [9] the problem simplifies when
239 the expected collateral at time s is a fraction of the risk-neutral expected survival-contingent
11
The valuation is repeated at frequent intervals-typically daily. Thus, the collateral position is then adjusted
to reflect the new valuation.
12
This is relaxed in the empirical chapter of Cama [3].
13
They also introduce cost of collateral as a benefit/cost of holding it.

12
240 market value at time s + 1. Taking into account the foregoing, I define the amount of collateral
241 as follows;

Definition 1 (Market Value of Collateral - MVC) Under a risk-neutral probability mea-

sure Q the market value of collateral is defined as;

EsQ (c's+1 ) = Q
s Es (Vs+1 )

242 Thus, using the definition (1), the expression (2) can be expressed as follows;

Vt = s he
rt
EtQ (Vt+1 ) + (1 h)e rt EtQ (Vt+1 )
8 9 !
243 < X1 = (3)
EtQ exp Rt+j Xt+
: ;
j=0

244 where t is set before default time i.e. t < T d . The above expression is obtained by recursively
245 solving (2) forward over the life of the bond, see appendix for details. Then, a period before
246 default, the promised payo (Xt+ ) is default-free. is the number of periods immediately
247 ending before default time (t + ) and finally R is obtained as follows;

Rt rt rt
248 e = (1 h)e + t he (4)

249 As in Duffie and Singleton [9] for time periods of small length, the former can be seen as
250 Rt ' rt + h(1 t ). So, the fraction of market value posted as collateral positively influences the
251 interest rate. In other words, the swap spread, i.e. Rt rt , is a function of t. As an informal
252 corollary, if h = 0 i.e. there is no counter-party risk, then the swap contract, that originally
253 gathers multiple short-term contracts of swapping payments i.e. Xt = rf,t sw , will be as
254 expressed as in (1) where rf,t is the floating rate calculated at time t. In the calculation of sw I
255 will refer to the floating rate as Libor with maturity T ; the foregoing is denoted by L(T ). In the
256 next lines I add the collateral requirements into the model. The solution requires a continuous
257 setup since the default time ( ) lies on the set (0, T ). Since posting collateral is costly but adds
258 more value to the contract then I calculate the current value of keeping collateral up to . The
259 solution also requires to evaluate Vt under the probability de default over the maturity of the
260 claim. Thus, the expected value under a martingale measure Q includes the indices 1{ >T } and
261 1{ T } , those expressions denote a dummy variable or binary result relying on default time ( ).
262 Finally, all expected streams of paymets up to maturity T are discounted at interest rate r.

13
 Formally, in a continuous solution and following the setup in Johannes and Sundaresan [20]
that includes a costly collateral, the market value is
h RT R i
Vt = EtQ e t rs ds T 1{ >T } + e t rs ds ' 1{ T }
h Z T Z s Z Z i
Q
+ Et 1{ >T } exp ru du ys 's + 1{ T } exp ru du ys 's ds
t t t t

263 where ys is the benefit of posting collateral at time period s that increases the value of the
264 contract Vt ; T is equal to the dierence between the swap rate and the annualized Libor
265 applied to the contract up to maturity. The solution of the market value of the contract when
266 's is equal to Vs is
h Z T i
267 Vt = EtQ exp (rs + h(1 s) ys s )ds (L(T ) sw ) (5)
t

268 L(T ) is the Libor for the contract under maturity T ; this floating rate will be eectively swapped
269 according to the contract14 . Details of the derivation in the appendix. Since the market value
270 of the contract at the inception is equal to zero then the swap rate under collateralization is
271 defined in the following lemma (1). This result implicitly assumes non-full recovery of the value
272 of the contract after default which is a variation of Johannes and Sundaresan [20]s main result.

273 Lemma 1 (Swap rate under collateralization) . The swap rate including collateralization
274 and cost of posting collateral as in Johannes and Sundaresan [20] is
" n R o #
T
exp t (r s + h(1 s ) y s s )ds L(T )
w Q
s = Et
p(0, T )
" # (6)
275 n R o
T
exp t (rs + h(1 s) ys s )ds , L(T )
= E0Q [L(T )] + cov0Q
p(0, T )

276 Being p(0, T ) the discount factor up to maturity date (T ).

277 Above lemma state that the swap rate depends on the expectation over the measure Q of the
278 Libor15 and the linear association between R and the Libor. More important, the parameter of
279 interest is is going to be critical under clearing practices; once novation facilitates netting
280 shrinks and consequently the swap rate decreases since there is less requirement of collateral.
14
For instance, whether the contract previously requires swapping the rates after 6 months then the Libor
under consideration will be the expected rate to 6 months at the inception of the contract.
15
This contract is not hedged; however the treatment with forwards will be the same as in section under no
collateralization.

14
281 A negative correlation between libor and above adjusted interest rate is required for having the
282 foregoing statement true. Formally, the definition of is as follows.

283 Definition 2 ( ) s denotes the exposure ('s ) in terms of units of the value of the contract
c's
284 (Vs ); i.e. s = Vs .

285 The fraction varies over time since the exposure can change due to clearing methods. The
286 exposure calculation is discussed in section (4).

287 3.1.3 Dierence in swap rates

288 The final objective of this paper is to bring up discussion about the determinants behind the
289 dierence between swap rates among clearinghouses; thus the expression in discussion for an-
290 alyzing is as follows being superscripts on , sw and ys related to clearinghouses. I formally
291 specify the expression under interest for this paper in the following corollary;

Corollary 1 (Basis) The dierence in swap rates or basis among clearinghouses A and B is
as follows;
" #
nR o
T A
exp t (rs + h(1 s ) ysA A
s )ds , L(T )
s w,A
s w,B
= cov0Q
p(0, T )
" #
nR o
T B
exp t (rs + h(1 s ) ysB sB )ds , L(T )
cov0Q (7)
p(0, T )

292 In the next section I follow the same approach for determining an analytical expression that
293 relates clearing practices with the premium of credit default swaps.

294 3.2 Determination of the credit default swap premium

295 In this model there is an agent that is susceptible to some loss of wealth. There is also other
296 agent that sell insurance against that loss. Finally, there is another agent that have access to
297 some investment project but requires some funding in order to eectively undertake the project.
298 Summarizing and formalizing the setup of the model: there are three agents, the buyer of the
299 protection (henceforth the buyer or B), the seller of the protection (henceforth the seller or S)
300 and a third party (henceforth the investor or I) that has access to some investment project.

15
301 The investment and protection have a maturity of two periods (t = {0, 1}); the contract is set
302 at the first period. I describe in the following lines the timing of the actions involved, states of
303 nature and availability of endowments with uncertainty.

304 At t = 0, the endowment for the buyer !B is known with certainty; a quantity qm is
305 transferred to the seller with the promise of receiving m if the event happens (the loss of wealth);
306 q is the price of the contract per unit of m. Once qm is received then z is transferred to the third-
307 party with the promise to return z(1 + 2r ). A counter-party risk arises also from the foregoing
308 contract since this third party may default. The following restriction applies: !B qm z. I
309 assume that the buyer engages in purchasing insurance; otherwise it will receive a penalty or
310 cost L (see Duffie and Zhou [12] for details of this setup16 ).

311 At t = 1, the endowment for the seller !s is unveiled at no cost17 ; ex-ante the availability
312 of this endowment happens with some probability and this determines the default or not of the
313 original insurance contract. The third-party can seize both return r and notional amount z of
314 the project and declares default with some probability greater than zero. This action may be
315 verifiable but it is costly. The variable that remains indeterminate in the model is z. I assume
316 that contingent claim related to the amount z is exogenous given18 . Since seller has a short
317 position in CDS (and potentially over other contingent claim), the seller must post collateral.
318 Also, cross subsidization of costs of collateralization among contingent claims is possible and
319 realistic, I will give more details in the section of payos by arrangement on the foregoing.

320 The basic structure under discussion is depicted in the figures (4) and (5). These figures
321 show participants, the size of the exposures among them and the type of arrangement. The
322 bilateral arrangement supposes netting amid class of assets for each partner separately. Instead,
323 the clearing (or multilateral) arrangement supposes netting of dierent class of assets amid all
324 participants. The formalization of these arrangements -i.e. the calculation of exposure- is left
325 to section 4. Also in this structure each pair seller-buyer and seller-investor only trade one
326 asset; it is possible to have a more complicated structure that allows trading of a dierent class
327 of assets amid same pair of participants. However, any arrangement would produce the same
328 size of exposures due to netting ultimately. Finally, relative to the bilateral arrangement, the

16
Duffie and Zhou [12] assumes that if banks experiences a loss of principal on a loan, it incurs an additional
deadweight cost. Moreover, if the bank does not purchase insurance, then it also incurs some multiple of above
deadweight cost.
17
The quantity !s is verifiable
18
Actually it could be related to a contract of interest-rate swaps in general terms for instance

16
329 clearing arrangement produces a shrinkage of the exposure of the seller due to opposite positions
330 among dierent participants. Even though the total exposures in the multilateral arrangement
331 increases, the seller reduces her exposure significantly thus aecting the price of the contingent
332 claim.

333 The price determination is standard as asset valuation suggests. The method needs a dis-
334 counted premium and protection legs. The former takes into account the premium payments
335 and possible accrue value if the credit event occurs. The protection leg calculates the protection
336 amount when credit event happens. Then, after setting the values of these two legs, the premium
337 is determined by equalizing these two terms to each other.

Figure 4: Bilateral arrangement

Figure 5: Clearing arrangement

338 The figure (6), that shows the timming of actions and resolution of the contracts, summarizes
339 above discussion.

17
 Figure 6: Timming of the model

340 3.2.1 Trading Frictions

341 In this section I describe the three frictions in the environment described by the model. First,
342 the event -the loss of wealth (L)- encourages buyers to get a CDS contract. I assume that an
343 agent made choices (of investment or production) before the realization of this shock of size
344 L. They could insure against the shock by demanding full payment, however the former is
345 costly. For instance, intermediate producers -that expect a demand shock after production-
346 would demand to pay the whole bill o at stage 0. Thus, the shock would potentially halt the
347 production process or discourage the demand for the intermediate good. In order to avoid the
348 foregoing result, the producers buy a CDS contract in which promises a compensation if the
349 event happens.

350 Second, sellers may declare default. Thus buyers that trade with a specific seller in the first
351 period face a defaulting exposure. In other words, while being insured against the related event,
352 the buyer who writes a CDS contract now faces a default risk. Why not get insurance against
353 the default of the CDS seller? Buyers would need to trade another CDS contract with other
354 sellers; the new contract should hedge the counter party risk at some particular state of nature,
355 however the shock is aggregated in nature i.e. other sellers also have !s = 0. The only way to
356 hedging is by getting collateral. This setup reveals the incompleteness of the market.

357 Third, as Townsend [32] and Radner [28] suggests, output or result of projects may be
358 verifiable to some cost; in the model, the agent I is informed on the actual state of nature
359 of the project and in which this information may be transmitted to the rest of agents only at
360 some costs. In this paper, I assume that these costs are important and additionally I invoke the
361 conditions established in Ordonez [27] that allow to observing the borrowing of the amount z

18
362 and the report of some result of the project. Thus, there some states of nature where the agent
363 I only walks away from terms of contract.

364 Summarizing, the environment formalizes the fundamental frictions that will allow us to
365 endogenize the need for a CDS contract and setup a proper clearing arrangement. CDS contract
366 between a buyer and a seller can partially insure against the main event but exposes the buyer
367 to counter-party default. Collateral in form of pre-payment is available from endowments but
368 it is costly. This setup provides a rationale for clearing arrangements that can provide cheaper
369 and better insurance against default risk.

370 3.2.2 Bilateral arrangement: Payos by state of nature

371 Figure (4) shows the financial structure for a bilateral arrangement; buyer (B) has an exposure
372 of m to seller and this same seller (S) has an exposure of z to third-party (I). Requirement
373 of collateral is exogenous given and denoted by c as a fraction of the exposure. The nature of
374 the exposure of the seller to the agent I may be related to the investment of the whole or some
375 fraction of the total premium19 (qm) into a technology that returns 1 + r; seller and third-party
376 share r equally. The availability of the (observable) endowment ! s makes the seller to default
377 or not default. I also take into account that costs per unit of collateral i.e. 1 can be split
378 for calculation of benefits amid markets. In other words, the seller would consider a fraction !1
379 of these costs for calculation of benefits when sell CDS; and 1 !1 is earmarked when benefits
380 are calculated for other contingent claim. This feature adds the fact that competition in prices
381 make sellers to subsidizing among markets.

382 In the following table (1) I show the size of exposure for each participant in the basic structure
383 depicted in figure (4). As defined early, amount z is related to some class of derivatives, for
384 instance interest-rate swaps or loans; while m is related to CDS asset specifically.

Participant b cl

S z 0
B m m
I 0 0

Table 1: Exposure for the basic structure

19
It also may be related to the purchase of insurance - the seller buying insurance from others-.

19
385 The symbols b and cl denote exposures in a bilateral and clearing arrangement. In the
386 case of clearing practices, as shown in above table the exposure of the seller is equal to zero since
387 z m < 0. The foregoing is a result of netting practices. This exposure must not be confused
388 with the exposure of the clearinghouse to the seller which is equal to m z as shown in figure
389 (5). The following table (2) shows the cost of posting collateral.

Participant Bilateral clearing

S cm( 1) c(m z)( 1)
B 0 0
I cz( 1) cz( 1)

Table 2: Cost of posting collateral

390 I calculate the cost of posting collateral and the value of the collateral in each arrangement,
391 I denote this value as (m, z) which represent these costs per unit of c. The value of the
392 collateral is expressed in netting terms since the seller receives and posts collateral.

Participant Bilateral clearing

z
S (m ) (m z)
m m
B

I z z

Table 3: Function '(m, z)

393 In order to determine the premium of the CDS I explicitly state the payos of each partici-
394 pant. Thus, I consider the following events;

395 The (credit) event occurs 1; Otherwise 0.

396 Seller defaults 1; Otherwise 0.

397 Investor defaults 1; Otherwise 0.

398 For example the triplet (1; 0; 0) means: the event occurs and the seller and third party keep
399 the promise to pay back. I also consider a dierent notation for the costs of posting collateral
400 in the case either the seller or investor do not default. I denote this as x c'x (m, z)( 1),
s

20
401 where x denotes the type of arrangement and s identifies the participant. I early mentioned the
402 possibility of cross-subsidizing costs of collateralization between markets. I denoted as !1 the
403 fraction of collateral costs that enters into calculation of the premium.

404 State 1: (1;1;1)

B =! B qm L + cm

405 S =qm c'b (m, z)!1 z (8)

I =z(1 + r) cz

406 State 2: (1;0;1)

B =! B qm L+m

407 S =qm m z(1 c) + ! s b

s !1
(9)

I =z(1 + r) cz

408 State 3: (1;0;0)

B =! B qm
L+m
r
409 S =qm m + z + ! s b
s !1 (10)
2
I r b
=!I + z I
2

410 State 4: (1;1;0)

B =! B qm L + cm
r
411 S =qm c'b (m, z)!1 + cz + z (11)
2
r
I =!I + z b
I
2

412 State 5: (0;1;1)

B =! B qm

413 S =qm z(1 c) b

s !1
(12)

I =z(1 + r) cz

414 State 6: (0;0;1)

21
 B =! B qm

415 S =qm z(1 c) + ! s b

s !1
(13)

I =z(1 + r) cz

416 State 7: (0;0;0)

B =! B qm
r
417 S =qm + z + ! s b
s !1 (14)
2
r
I =!I + z b
I
2
418 State 8: (0;1;0)

B =! B qm
r
419 S =qm + z b
s !1 (15)
2
r
I =!I + z b
I
2

420 3.2.3 Clearing arrangement: Payos by state

421 Below the payos under a clearing arrangement.

422 State 1: (1;1;1)

B =! B qm L + cm

423 S =qm c'cl (m, z)!1 z (16)

I =z(1 + r) cz

424 State 2: (1;0;1)

B =! B qm L+m

425 S =qm m z(1 c) + ! s cl

s !1
(17)

I =z(1 + r) cz

426 State 3: (1;0;0)

B =! B qm
L+m
r
427 S =qm m + z + ! s cl
s !1 (18)
2
I r cl
=!I + z I
2

22
428 State 4: (1;1;0)
r
B =! B qm L + c(m z) + z(1 + )
2
429 S =qm c'cl (m, z)!1 z (19)
r
I =!I + z b
I
2
430 State 5: (0;1;1)

B =! B qm

431 S =qm z(1 c) cl

s !1
(20)

I =z(1 + r) cz

432 State 6: (0;0;1)

B =! B qm

433 S =qm z(1 c) + ! s cl

s !1
(21)

I =z(1 + r) cz

434 State 7: (0;0;0)

B =! B qm
r
435 S =qm + z + ! s cl
s !1 (22)
2
I r cl
=!I + z I
2
436 State 8: (0;1;0)

B =! B qm
r
437 S =qm + z cl
s !1 (23)
2
r
I =!I + z cl
I
2
438 Since I assume that there is a separated profit function for trading CDS. Thus, I put aside
439 any settlement related to z into the profit function z .

440 s = CDS + z (24)

441 In the following lines I calculate the payos by state when trading CDS. It is worth men-
442 tioning that function CDS is always zero. Accepting trading CDS avoids the deadweight loss
443 L thus it is optimal for sellers to issue CDS.

23
444 State 1: (1;1;1) q x m c'x (m, z)!1

445 State 2: (1;0;1) ws + q x m m x!

s 1

446 State 3: (1;0;0) ws + q x m m x!

s 1

447 State 4: (1;1;0) q x m c'x (m, z)!1

448 State 5: (0;1;1) q x m x!

s 1

449 State 6: (0;0;1) ws + q x m x!

s 1

450 State 7: (0;0;0) ws + q x m x!

s 1

451 State 8: (0;1;0) q x m x!

s 1

452 Premium determination. As before, premium is obtained under the assumption of zero
453 profits. So, I state the zero profit condition for the seller. As before, for sake of notation, I
454 pinpoint the probability of the event or state i as pi . Since some states deliver same payo, the
455 probability for the foregoing payo will have a proper notation that gather the sum of respective
456 probabilities.

qxm p1,4 c'x (m, z)!s p2,3 m + p2,3,6,7 !s p2,3,5,7,8 x

s !1 =0

457 Thus the premium traded in clearinghouse or strategy x is;

!1 !s x !1
458 q x = p2,3 + p1,4 c'x (m, z) p2,3,6,7 + p2,3,5,7,8 s (25)
m m m

459

460 Since x = c'x (m, z)( 1), then above expression can be arranged as follows:
s

461

!1 h i !s
462 q x = p2,3 + c'x (m, z) 1+p p2,3,6,7 (26)
m m
463

464 Where p = p1 +p4 ; in other words it is the marginal probability of seeing the seller defaulting
465 when the main event occurs.

24
466 3.2.4 Dierence in premium

467 I compare the premium between a bilateral and clearing agreement. It is worth noticing that
468 collateral requirements may be potentially dierent under each arrangement; however, I assume
469 that collateral fraction c under a clearing and bilateral agreements are the same.

h i !
1
qb q cl = c 1+p 'b (m, z)
'cl (m, z) (27)
m
h ih z i!
1
=c 1+p m +z m
m
h i ( 1) z
=c 1+p !1
m

470 Below figure 7a show the dierence in premium between clearing and bilateral arrangements
z
471 under the following parameterization: c = 0.20; m = 0.8 and !1 = 0.20

(a) (b)

Figure 7: Dierence in premium

472 I can use expression (26) in order to have a dierence in premium as a function of dier-
473 ent policies aecting collateralization. Thus, I have the following expression that explains the
474 dierence of premium between clearinghouses A and B with dierent collateralization policies:

h i ( 1) z
qA q B = (cA cB ) 1+p !1
m
The closed-form expression derived above may be arranged to show the price dierence among

25
 two dierent clearinghouses;

h i ('A 'B
qA qB = c 1+p !1
m

475 4 Calculation of the exposure in clearinghouses

476 I closely follow Duffie and Zhu [10], Cont and Kokholm [6] and Derrico et al. [7]s method for
477 calculating bilateral and clearing exposures. I consider a finite N market participants which
478 trade some derivatives. I allow for K classes of derivatives. These classes could be defined by
479 the underlying asset classes, such as credit, interest rates, foreign exchange, commodities, or
480 equities. Since multiples long- and short-positions may exist in any possible direction then these
481 flows can be oset or netted across participants. Formally, netting as a measure of efficiency
482 can be achieved through either a bilateral or clearing process. The latter extracts additional
483 benefits by the usage of novation. For instance, if entities i and j have a CDS position by which
484 i buys protection from j, then both i and j can novate to a clearinghouse, who is then the seller
485 of protection to i and the buyer of protection from j.

486
k be the amount that j will owe i in some derivatives class k,
Amid participants, let Xi,j
487 before considering the benefits of netting across asset classes, collateral, and default recovery.
488
k , 0) because, by definition X k =
Similarly, the exposure of participant i to j is max(Xi,j k
Xj,i
i,j

489 The posting of collateral depends on the size of the exposure and therefore counter parties
490 incur in pay-up front payments. The unit cost of posting these collateral requirement is 1
491 where 1. The fraction of collateral per unit of exposure is denoted by c.

492 Under bilateral netting, the exposure of participant i to any counter-party j, is netted across
493 all K derivative classes, but exposures to dierent counter-parties cannot be netted (see Duffie
494 and Zhu [10]). Formally, the total netting efficiency is:

Xh nX
K oi
b k
495 N,K = max Xi,j ,0 (28)
j6=i k

496 I consider the implications of a clearinghouse for all class of derivatives, this measure of
497 efficiency has a broader definition in comparison to the one specified in Duffie and Zhu [10]. In
498 this case, all positions across assets amid counter-parties are novated to the same clearinghouse:

26
 nXX
K o
cl,K k
499 N,K = max Xi,j ,0 (29)
j6=i k

500 In equation (28) netting is done across all asset classes for each counter-party; and the total
501 net exposure of dealer i is the sum of the individual exposure calculated with each counter-party
502 j. When a CCP is introduced in the market, as in equation (29), the total net exposure is the
503 exposure to the CCPs of party i which is calculated across assets and counterparties. As Cont
504 and Kokholm [6] states, whether the introduction of a clearinghouse increases or decreases net
505 exposures depends on the particular market, e.g. the notional sizes of the asset classes, riskyness
506 of the asset classes, correlation between the asset classes, the number of asset classes, the number
507 of participants etc. In the next section, I start deriving the size of the (variable) exposure (X)
508 for each participant assuming that it is proportional to the notional values. Then, I show the
509 analytical expectations of the net exposures specified in expressions (28) and (29).

510 The exposure cl,k1 assumes that asset class k1 is cleared through a central counter-party;

X nX o nX o
cl,k1 k1
511 N,K = max (1 wk )Xijk , 0 + max wk1 Xi,j ,0 (30)
j6=i K j6=i

512 The exposure cl,k1,2 assumes that two asset classes k1 and k2 are separately cleared through
513 a central counter-party

X nX o nX o nX o
cl,k1,2 k1 k2
514 N,K = max (1 wk )Xijk , 0 + max wk1 Xi,j , 0 + max wk2 Xi,j , 0 (31)
j6=i K j6=i j6=i

515 4.1 Assumptions of the joint distribution among assets

516 Lets consider that for a specific class of asset k, I assume Xijk N (0, 2
k) for all i 6= j and allowed
517 to be correlated across other assets. As in Duffie and Zhu [10] and Cont and Kokholm [6], I
518 assume that the standard deviation of X k is proportional to the credit exposure i.e. k = k X.

519 4.2 Assumption about heterogeneity

520 As in Cont and Kokholm [6], instead of credit exposures I use notional values to determine the
521 size of the Xk and assume that they are proportional to the notional values according to the

27
522 following expression:
Zjk
523 Xijk = k Zik P k
Yijk (32)
h6=i Zh

524 Where Yijk is distributed as N (0, 1) and Zik is the notional size of dealer i in asset k. In other
525 words, the exposure of i to j for asset k is a fraction of the notional value of party i in derivative
526 class k. This fraction is the notional value of counterparty j in comparison to the total amount
527 of all other remaining notional values for h dierent of i. The parameter k is related to the
528 risk of asset k. Putting the former dierently, dierent assets may dier in the valuation of the
529 risk and therefore this will lead to dierent collateral requirements. As in Cont and Kokholm
530 [6], I assume that this parameter is equal to 3.9e 3 and 9.8e 3 for interest rate and credit
531 default swaps respectively20 .

532 4.3 Expectation of exposures

533 Since Yijk is distributed as a normal, then analytical expression for the expected (net) exposure
534 is as follows;

v
u
1 X uX X Zjk mP j
Zm
535
b
E('i ) = p t km k Zik P Z (33)
k m i m
2 j6=i k=1 m=1 h6=i Zh h6=i Zh
536 v
uX X X
cl,f 1 u Zjk mP j
Zm
537 E('i ) = p t km k Zik P k
m Z i m (34)
2 j6=i k=1 m=1 h6=i Zh h6=i Zh

538 Thus, in the case of CDS, the relevant expression

that determines the swap dierence between
E 'b 'cl
539 two clearinghouses in expression (27) is Zk
which can be calculated from above expres-
c'i
540 sions. In the case of interest rate swaps, the ratio Ztk
is the expression that matter for the
541 determination of the dierence in swaps as expression (7) suggests. Here, the foregoing fraction
542 will aect the discount rate and would produce a correlation with the forward rate; thus the
543 swap rate is aected.

544 I also consider the calculation of exposures when a fraction wk of a contingent asset k is
545 cleared in a clearinghouse; this approach is the same as in Cont and Kokholm [6]. I intend to
546 replicate the results in Duffie and Zhu [10] and quantitatively show the size of exposure between
20
Cont and Kokholm [6] states that for the CDS, k is calculated as the mean of the standard deviation of the
daily profit-loss of 5-year credit default swaps on the names constituting the CDX NA IG HVOL series 12 in the
period July 1st, 2007 to July 1st, 2009 (page 13). In the case of interest rates, this parameter is calculated as the
standard deviation of the historical daily profit-loss from holding a 5-year with notional of 1.

28
547 a bilateral and full clearing strategies. Also, I consider the size of the exposure when there
548 are at least two clearinghouses. The idea is to show how dierent variation of market structure
549 would impact in the size of the exposures and consequently on price of assets.

550 The following expression gives the expected exposure when there is a fraction of one contin-
551 gent claim cleared in one clearinghouse.

v
u
1 X uX X Zjk Zjm
cl,1
E('i ) = p t k
(1 wk )(1 wm )km k Zi P Z m P
k m i m
2 j6=i k=1 m=1 h6=i Zh h6=i Zh
552 qP (35)
K 2
1 K j6=i (Zj )
+ p K w k Z i P K
2 h6=i Zh

553 Parameter wx = 0 unless x = {K} since the contingent claim K is the only one cleared in
554 the clearinghouse.The following expression gives the expected exposure when there is a fraction
555 of two contingent claim cleared in separated clearinghouses.

v
u
1 X uX X Zjk Zjm
cl,2
E('i ) = p t (1 wk )(1 wm )km k Zi Pk Z m P
k m i m
2 j6=i k=1 m=1 h6=i Zh h6=i Zh
qP
K 2
556 1 K j6=i (Zj ) (36)
+ p K w k Z i P K
2 s h 6= iZh
qP
K 1 2
1 K 1 j6=i (Zj )
+ p K 1 w k 1 Z i P K 1
2 h6=i Zh

557 Also above parameter wx = 0 unless x = {K, K 1}.

558 5 Analysis and discussion

559 In this section, I present discussion and implications of the analytical expressions (7) and (27)
560 those related to the swap-rate dierence between clearinghouses. At the end of this section I
561 will succinctly discuss the data that could be used in order to find evidence of the eect of
562 clearing methods on swap valuation. The full empirical strategy for the underlying model is
563 pushed into the chapter III in my dissertation. In this section I will describe assumptions and
564 possible scenarios where the theoretical model may explain some particular events in the range
565 of data. This approach has certainly the advantage of freely playing with assumptions over the

29
566 set of parameters; thus, below discussion is clearly simple and oers escenarios that support the
567 results of the underlying theoretical model more likely.

568 5.1 The dierence in CDS premiums

569 The expression in (27) shows that the sign of the swap rate is given by the dierence between
570 sizes of exposures relative to the amount of CDS trading. In order to evaluate the sign of this
E('i,A )
571 expression, I construct the ratio mi,A
for clearinghouse A and participant i, m is the notional
572 amount of CDS traded in that clearinghouse. The expected measure of exposure (') is specified
573 in section (4.3) and the exercise requires data of notional values; I follow Cont and Kokholm
574 [6] and use data of top 10 companies in USA from Office-of-the Comptroller of the Currency
575 (OCC) for calculation of exposures. The amounts and market participation of 10 top companies
576 are shown in table (4).

577

Holding Total derivatives Swaps Credit

CITIGROUP INC. 47092.6 25141.0 1761.9

JPMORGAN CHASE & CO. 46992.3 25670.8 2028.4
GOLDMAN SACHS GROUP, INC., THE 41227.9 20837.7 1424.3
BANK OF AMERICA CORPORATION 33132.6 19044.2 1264.9
MORGAN STANLEY 28569.6 15660.9 904.3
WELLS FARGO & COMPANY 7099.0 4496.8 30.9
HSBC NORTH AMERICA HOLDINGS INC. 6342.5 4012.4 125.0
MIZUHO AMERICAS LLC 4755.2 4364.7 4.1
STATE STREET CORPORATION 1445.8 12.1 0
CREDIT SUISSE HOLDINGS (USA), INC. 989.4 82.2 45.6

Table 4: Notional amounts of derivative contracts (in US billions)

578 For sake of explanation and intuition I can start with a simple example; assuming the
579 following: i) there are only two participants in each clearinghouse, ii) two contingent claims k
580 and m (for instance m is CDS and k is an interest swap contract); iii) correlation between assets

30
581 () is zero, and (iv) same risk parameter () amid assets, then expression (34) is as follows:
q
1
582 'A (m, k) = p (Sik )2 + (Sim )2 (37)
2
'i
583 Where S k and S m is the notional value of k and m. Thus, the expression for fraction mi
in
584 terms of notionals is: s
'A (m, k) 1 S k 2
i
585 p 1+ (38)
mi,A 2 Sim
586 Thus, expression (35) for the above case is denoted as;
s
B
' (m, k) 1 S k 2 1
p i
587 1 w m + m + p wm (39)
mi,B 2 S i 2

588 where Sik + Sim = Si is the total notional of derivatives. Since there are two participants, the
589 clearinghouse which is not specialized is actually representing a bilateral arrangement; thus the
590 exercise resembles a comparison of price dierence between a bilateral arrangement (clearing-
591 house A) and a specialized clearing arrangement i.e only clearing CDS (clearinghouse B). The
592 results discussed in the next lines are mostly driven by the number of participants as Duffie and
593 Zhou [12]s findings suggest. Figure (8) shows the size of the dierence between rates under dif-
594 ferent sizes of the notional values for interest rate swaps respect to the total value of derivatives
595 market for agent i and also dierent fraction sizes (!m ) of the value of CDS that are cleared. I
596 provide a benchmark (red line) in figure (8) that resembles a high market participation of LCH
597 in credit derivatives.

598 In the following table I show three more scenarios and their characteristics

Scenario participants Risk parameters Correlation

0 2 m = k = 0; Other sizes of
1 10 m = k 2 ( 0.8, 0.8)
2 10 0.5m = k 2 ( 0.8, 0.8)
3 10 0.3m = k 2 ( 0.8, 0.8)
4 10 0.1m = k 2 ( 0.8, 0.8)

Table 5: Scenarios for CDS premium determination

599 Given the scenarios I can calculate the premium basis under dierent sizes of exposures that
600 hinge on the fraction of notional cleared and number or participants. For the scenarios 1 to 4, I
601 use expressions (34), (35) and (36) for calculation of exposures. I use (33) as a comparison case.

31
 (a) Dierent CDS Market-size under clearing (b) Dierent correlation amid assets

Figure 8: Dierence in premium between clearinghouses

602 Figure (8) shows two participants that trade CDS and other derivative as interest rate swaps.
603 In the panel (8a) the dierence in premium declines as long the market share of interest rate
k
604 swaps ( SS ) increases. In other words, for any market share, the seller that clears all assets in
605 one clearinghouse has a price (q cl,0 ) lower than seller that only clears CDS has. As a higher
606 percentage of the notional goes to the specialized clearinghouse the exposure is higher and the
607 premium is too. In the panel (8b) a higher negative correlation amid these two assets increase
608 the dierence in premium between these two sellers.

609 The following figure (9) shows the exposure ratio of an agents that choose to clear some
610 fraction of all assets in a clearinghouse respect to clear same assets separately. Each symbol
611 represents a holding as shown in table (4). The eect of netting is significant when the correlation
612 between positions is negative; as long the correlation goes to positive the gains are not large of
613 clearing in one clearinghouse.

614 Figure (10) shows the exposure ratio under dierent assumptions regarding the relative size
615 of the valuation of risk () for the CDS and IR,

616 5.2 The basis between LCH and CME

As described in expression (7) the dierences in swap rates hinge on the covariance between the
default-risk adjustment (R) and the LIBOR. In order to sign the dierence between sw,A and
sw,B I undertook a first-approximation to the exponential ex 1+x this implies for sw,A < sw,B

32
 (a) wm = 0.3 (b) wm = 0.5

(c) wm = 0.7 (d) wm = 0.9

Figure 9: Exposure ratio between one versus two clearinghouses

the following;
h Z T i h Z T i
cov0Q 1 (rs + h(1 A
s ) ys sA )ds, L(T ) < cov0Q 1 (rs + h(1 B
s ) ys B
s )ds, L(T )
t t

Assuming that benefits of collateral are constant i.e. ys = y, above expression is equivalent to:
h Z T i h Z T i
cov0Q 1 (rs + h(1 A
s ))ds, L(T ) < cov0Q 1 (rs + h(1 B
s ))ds, L(T )
t t

617 Finally, if I assume that clearinghouse A keeps constant any clearing strategy A across time and
s

618 term structure; then the swap rate in a clearinghouse is higher than other swap rate-traded in
619 another clearinghouse- that assumes no linear association with Libor. If I assume that A =0
s

620 and ysA = 0 then the result in Johannes and Sundaresan [20] arises. I define the concept of
621 strategy as the one pursued by clearinghouse that reduces the amount of collateralization
622 required in the contract. I formally show these results and the usage of concepts in the following
623 proposition and a corollary,

33
 (a) k = m (b) k = 0.5m

(c) k = 0.3m (d) k = 0.1m

Figure 10: Exposure ratio between one versus two clearinghouses

624 Proposition 1 The basis between clearinghouses A and B, sw,A sw,B , where clearinghouse
625 A keeps an invariant strategy of netting across time and term structure, is positive under the
626 following condition,
hZ T i
627 cov0Q h B
s ds, L(T ) 0 (40)
t

628 i.e. sw,B sw,A .

629 Proof. The result is easily obtained from above discussion .

hR i
T
630 Corollary 2 If A
s = 0 and cov0Q t
B
s (h + ysB )ds, L(T ) > 0; the swap rate in the presence
631 of costly collateral is higher than that one which assumes no default i.e. sw,B > sw,A .

632 Proof. The result follows from discussion in Johannes and Sundaresan [20] . The following
633 proposition shows the sign of the basis when there is an strategy of reducing the collateral
634 requirements i.e. netting across time keeping constant this requirement across term structure.

34
635 Proposition 2 Considering a strategy B (t, s) = a0 a1 t where t denotes time and it is not
636 aected by the term structure; if expression (40) holds then sw,B sw,A , Otherwise sw,B > sw,A .
RT B B
637 Proof. Expression t=0 h s ds is equivalent to h t T. Assuming that exposure ('t ) shrinks
638 across time but term structure i.e. B (t, s) = a0
a1 t, then the sign of the basis hinge on
h i
639 correlation of these lessening exposure with Libor rate i.e. cov0Q hT (a0 a1 t), L(T ) .

640 Above proposition concludes that the basis would depende of the linear association between
641 the netting strategy of the clearinghouse and LIBOR. Lets see the data. Figure (11) show the
642 behavior of six months Libor and the LCH-CME basis from the end of 2015 to October 2017.

(a) Libor rate (b) Basis

Figure 11: Daily Libor and Basis

643 Above figure shows that Libor trend was reversed starting 2015; before this year the Libor
644 rate had a long-run persistent negative trend. The libor was completely flat starting the financial
645 crisis up to 2014. Since 2016 the libor rate shows a persistent positive trend.

646 Figure (2) shows that LCH started gaining market participation in clearing at a significant
647 pace; whether assuming that this clearinghouses risk management strategy were accompanied
648 by reducing exposures through netting, then the covariance between t and libor would be more
649 likely negative. As a consequence and according to proposition (2) the basis would be positive
650 then i.e. sw,CM E sw,LCH > 0.

651 Spikes in basis data are also puzzling. Panel figure (11b) shows that basis is wide immedi-
652 ately before the end of 2015 and after July 2016 to the end of that year. The basis for claims
653 with a long-term maturity is visibly more aected. This may be associated to significant in-
654 crease of the libor rate under these ranges of data, see panel figure (11a). In those periods the

35
655 covariance between t and libor more likely would be increasing and as a consequence widening
656 the basis. I do not discard either the eects of reducing exposures by the other clearinghouse
657 and collateralization across term structure in order to explain the wide of the basis over this
658 period in analysis; this for sure would be clarified in a proper empirical strategy.

659 6 Conclusions

660 I show how the exposures are related to the swap rates in a theoretical framework. I analyzed
661 this relationship in a financial structure that allows clearing of positions. Clearing is potentially
662 a tool that reduces significantly exposures and thus lessen collateral requirements. The impact
663 on swap rates of these financial structure has not been broadly discussed in the literature. The
664 simple approach presented in this paper is not intended to capture every movement in swap rates,
665 and more likely changes not captured by the theoretical model are due to model misspecification,
666 market segmentation, or temporary miss-pricing as fairly suggested by Johannes and Sundaresan
667 [20].

668 I showed in section (3) that swap rates have a closed functional from hinging on the expo-
669 sure size. Sellers of these contingent claims may choose to clear these ones in clearinghouses
670 and taking further advantage of netting through novation. In the case of interest rate swaps,
671 I adjusted the standard valuation expression by incorporating the eect of collateral require-
672 ments in a clearing setup. In a full recovery assumption, interest rate swaps can be seen as
673 short contracts with terminal date at default. If only a fraction of the value of the contract is
674 recovered, then the swap would be increasing as collateral requirements are higher. Practices
675 as netting or compression that save collateral costs would reduce the swap rate. In section (4) I
676 discussed the calculation of the exposures and how they are aected by clearing through assets
677 and participants. I also consider the case where participants can clear their claims in separated
678 or specialized clearinghouses. The analytical expression for swap rates needs of information
679 about notional values, percentage of volume that is cleared, risk of the claim and relative size
680 of the market for each derivative.

681 I quantitatively showed in section (5) that relative size of the CDS market and the percentage
682 of claim that is cleared aects the exposure and thus the CDS premium. I started with a simple
683 example of two participants and two assets with the purpose of showing the benefits of netting.
684 An increasing fraction of CDS earmarked to clearing as long as the CDS market size shrinks

36
685 would increase the exposure in the structure with a CCP that clear only CDS claims; thus
686 the premium determined in a CCP which clears CDS and additional claims will be relatively
687 cheaper. Additionally, a higher and positive linear association between positions of dierent
688 claims makes the premium competitive when participants choose to clear both claims.

689 In a second exercise I use data from ten top holdings that handle derivatives in their portfolio.
690 I compare the premium between structures with one and two specialized CCPs. There is a
691 significant dierence in exposures between these structures when the fraction of CDS earmarked
692 to clearing is high and also when the correlation amid positions is highly negative. Additionally,
693 a lower risk of default for other claims rather than CDS would reduce the size of the exposure in
694 the structure with one CCP, this reduction will be important at negative values of correlation
695 amid positions.

696 In the case of interest rate swaps I could establish a relationship between netting strategy
697 through the time and the Libor rate that is connected with the basis. The negative linear
698 association between above variables shown in the years 2015 and 2016 seems to validate the
699 appearance of a widening basis between LCH and CME. The basis for claims with a long-term
700 maturity is visibly more aected and precisely are more likely associated to significant increase
701 of the Libor rate.

702 The results of this research has the following implications in terms of policy. First, netting
703 through novation has significant gains in reducing exposures, further benefits may relies on corre-
704 lation between asset positions as theory of portfolio predicts. I assess these gains using data from
705 OCC, these gains are large when comparing a financial structure of one clearinghouse respect
706 to other with two specialized clearinghouses. A financial structure with a clearinghouse can be
707 competitive in terms of prices and collateral requirements thus turning out to be appealing to
708 more participants. Second, a macro prudential policy such as capital requirement is becoming
709 a central feature in financial structures in the aftermath of the recent financial crisis and more
710 clearly after Dodd-Frank act. Clearing of a fraction of portfolio of derivatives of banks could
711 oer a feasible way to reduce pressure on these requirements since counter-party risk can be re-
712 duced eectively inside clearinghouses. Furthermore, other strategies as mutualization (sharing
713 losses amid participants) and further risk management strategies can potentially be part of the
714 regulation in the next years.

37
715 A Details of derivations

Derivation of expression (3). Derivation of final expression starts with the following ex-
pression:

Vt = (1 h)e rt
EtQ (Vt+1 ) + he rt
EtQ (c't+1 )

(1 h)e rt
EtQ (Vt+1 ) + he rt
EtQ ( Vt+1 ) (41)
h i
EtQ Vt+1 (1 h)e rt
+h e rt

Considering the change in notation for expression (1 h)e rt +h e rt as e Rt and considering

a recursive solution for expression (41) up to period t + , I have the following expression,
h P 1
i
Vt EtQ Vt+ e j Rt+j
(42)

716 Evaluating the right hand expression a period before the default then I have the expression in
717 (3) i.e. Vt+ Xt+

718 References

719 [1] Viral Acharya and Alberto Bisin. Counterparty risk externality: Centralized versus over-
720 the-counter markets. Journal of Economic Theory, 149:153182, 2014.

721 [2] Freddy A. Rojas Cama. The eect of mutualization and collateralization on credit default
722 swaps premium. PhD thesis chapter II, 2016.

723 [3] Freddy A. Rojas Cama. Credit default swaps under clearing: What data say? PhD thesis
724 chapter III, 2017.

725 [4] R. Chami, E. Kopp T. Cosimano, and C. Rochon. Back to the future: The nature of
726 regulatory capital requirements. IMF working paper, 2017.

727 [5] Pierre Collin-Dufresne and Bruno Solnik. On the term structure of default premia in the
728 swap and libor markets. Journal of Finance, 56:10951115, 2001.

729 [6] Rama Cont and Thomas Kokholm. Central clearing of otc derivatives: bilateral vs multi-
730 lateral netting. mimeograph, 2012.

731 [7] Marco Derrico, Stefano Battiston, Tuomas Peltonen, and Martin Scheicher. How does risk
732 flow in the credit default swap market. Working Paper Series, 33, 2016.

38
733 [8] Darrell Duffie and Kenneth J. Singleton. An econometric model of the term structure of
734 interest-rate swap yields. The Journal of Finance, 52:12871321, 1997.

735 [9] Darrell Duffie and Kenneth J. Singleton. Modeling term structures of defaultable bonds.
736 The Review of Financial Studies, 12:687720, 1999.

737 [10] Darrell Duffie and Haoxiang Zhu. Does a central clearing counterparty reduce counterparty
738 risk. Stanford Working Papers, 2010.

739 [11] Greg Duffie, Martin Scheicher, and Guillaume Vuillemey. Central clearing and collateral
740 demand. Working Paper, 2013.

741 [12] Gregory R. Duffie and Chunsheng Zhou. Credit derivatives in banking: Useful tools for
742 managing risk? Journal of Monetary Economics, 48:2554, 2001.

743 [13] Bank for International Settlements. Revisions to the basel iii leverage ratio framework. BIS
744 Consultative Document, April 2016.

745 [14] Hua He. Modeling term structures of swap spreads. Working Paper, 2001.

746 [15] Daniel Heller and Nicholas Vause. Collateral requirement for mandatory central clearing of
747 over-the-counter derivatives. BIS working papers 373, 2012.

748 [16] John Hull and Alan White. Pricing interest rate sensitive securities. Review of Financial
749 Studies, 3:573592, 1990.

750 [17] ISDA. International swaps and derivatives association inc. 2003. ISDA Margin Survey,
751 2003, 2003.

752 [18] R. Jarrow and F. Yu. Counterparty risk and the pricing of defaultable securities. Journal
753 of Finance, 56:17651799, 2001.

754 [19] Luc Laeven Lev Ratnovski Jihad Dagher, Giovanni DellAriccia and Hui Tong. Benefist
755 and costs of bank of capital. International Monetary Fund, 2016.

756 [20] Michael Johannes and Suresh Sundaresan. The impact of collateralization on swap rates.
757 The Journal of Finance, 62:383410, 2007.

758 [21] Jonathan Ingersoll John Cox and Stephen Ross. A theory of the term structure of interest
759 rate. Econometrica, 53:385408, 1985.

39
760 [22] Thorsten Koeppl. The limits of counterparty clearing: Collusive moral hazard and market
761 liquidity. Queens Economic Department Working Paper, 2013.

762 [23] Seng Yuen Leung and Yue Kuen Kwok. Credit default swap valuation with counterparty
763 risk. The Kyoto Economic Review, 74(1):2545, 2005.

764 [24] Robert L. McDonald. Derivatives Markets. 2e. The Addison-Wesley Series in Finance,
765 2006. ISBN 0-321-28030-X.

766 [25] Bank of International Settlements. Basel iii: A global regulatory framework for more re-
767 silient banks and banking systems., June 2011. URL http://www.bis.org/publ/bcbs189.
768 pdf.

769 [26] Bank of International Settlements. Global systemically important banks: Updated as-
770 sessment methodology and the higher loss absorbency requirement., July 2013. URL
771 http://www.bis.org/publ/bcbs255.pdf.

772 [27] Guillermo Ordonez. The asymmetric eects of financial frictions. Mimeograph, 2012.

773 [28] R. Radner. Competitive equilibrium under uncertainty. Econometrica, 36:3158, 1968.

774 [29] Che Sidanius and Filip Zikes. Otc derivatives reform and collateral demand impact. Fi-
775 nancial Stability Paper, 18, 2012.

776 [30] Eric Stephens and James R. Thompson. Cds as insurance: Leaky lifeboats in stormy seas.
777 Working paper, 2011.

778 [31] Adrian Tobias. Macro-prudential policy and financial vulnerabilities, Septem-
779 ber 2017. URL https://www.imf.org/en/News/Articles/2017/09/22/
780 sp092217-macroprudential-policy-and-financial-vulnerabilities. [Online;
781 posted 22nd-September-2017].

782 [32] R.M. Townsend. Optimal contracts and competitive markets with costly state verification.
783 Journal of Economics, 21:129, 1979.

784 [33] Jens van Egmond. The impact of collateralization on swap curves and their users. Network
785 for Studies on Pensions, Aging and Retirement, 2011.

786 [34] Oldrich Vasicek. An equilibrium characterization of the term structure. Journal of Financial
787 Economics, 5:177188, 1977.

40