Investment Research General Market Conditions

29 September 2010

FX Research
A G10 FX Financial Scorecard
In this paper we present our G10 FX Financial Scorecard. The scorecard is designed mainly for tactical investors who have exposure to the G10 currency universe. The G10 FX Financial Scorecard is based upon four input factors: 1) Technical analysis; 2) Interest rates; 3) Risk aversion; and 4) Option market positioning. By applying our own Z-score model, we can see whether exchange rates undershoot or overshoot this information. Following the G10 FX Financial Scorecards signals since 2006, has yielded 67.6% i.e. annualised return 14.3% (excluding carry and trading costs). A model based on the G10 scorecard has an information ratio of 1.43. Going forward, we will publish the G10 FX Financial Scorecards indications each Monday morning. We recommend clients who are interested in actively following or trading the G10 FX Financial Scorecard to contact our sales team or research team for details.

Key points
The vision of our G10 FX Financial Scorecard is to be able to spot important changes in financial variables that affect exchange rates and take advantage of unjustified exchange rate movements. Following the Scorecards indications has yielded an annualised return of 14.3%. A model based on the Scorecard has an information ratio of 1.43. Please contact the author of this research note if you have interest in receiving the indications of the Scorecard on a weekly basis, or our sales team if you want to trade the signals.
G10 FX Financial Scorecard performance since 2006
80 70 60 50 40 30 20 10 0 -10

A tough nut to crack: Exchange rate determination in the short run

In 2004, former Fed Chairman, Alan Greenspan, made the following statement about currency forecasting in a speech before the Economic Club of New York: Despite extensive efforts on the part of analysts, to my knowledge, no model projecting directional movements in exchange rates is significantly superior to tossing a coin. I am aware that of the thousands who try, some are quite successful. So are winners of coin-tossing contests. In other words, Alan Greenspan warns against having too high expectations of exchange rate determination. We respectfully note, however, that Greenspan has been proven wrong on other occasions and that the quote was made before our G10 FX Financial Scorecard was launched. Several factors affect exchange rates in the short run: risk appetite; releases of economic data; M&A flows; investor sentiment; market positioning; trend-following behaviour to name a few. However, the difficult thing is not to list the factors that affect exchange rates. The trick is to measure and apply these in an intuitive way and be able to profit from the information. We believe we have come a long way in systemising relevant short-run information for exchange rate forecasting purposes with our G10 FX Financial Scorecard. We are not claiming that the approach is perfect and we do not believe anyone ever will be able to present an ideal short-run exchange rate model as the currency markets are driven by irrational movements, generated for example by order flows. But we argue that the G10 FX Financial Scorecard offers value as it treats relevant information systematically and has performed well even in difficult times such as autumn 2008.

2006 2007 2008 2009 2010

Source: Bloomberg, Danske Markets calculations

Senior Analyst John M. Hydeskov +45 45 12 84 97 johy@danskebank.dk

www.danskeresearch.com

FX Research

G10 FX Financial Scorecard choice of input factors

Necessary characteristics for input factors have been: i) Accessibility; ii) Reliability; and iii) Accuracy. Accessibility is required as, for practical reasons, it would delay the process if collecting data was cumbersome and time consuming and potentially also increase the risk of errors. Reliability is crucial especially for back-testing purposes if revisions are frequent it means that we cannot count on data when published, decreasing forecasting abilities. Accuracy of data is almost trivial, but it is nevertheless important to check that inputs to the model are measured with the highest degree of punctiliousness. Finally, we have chosen to use data from the same source, in order to ensure that data is collected at the same time. We use Bloomberg as data provider and have not experienced any problems with data quality during the past three quarters in which we have worked with the Scorecard. Our G10 FX Financial Scorecard uses four inputs: 1) Technical analysis; 2) Interest rates; 3) Risk aversion; and 4) Option market positioning. We are not claiming that these factors encompass all relevant information for forecasting purposes, but they catch changes in important drivers for exchange rates in the short run. Ad 1) Technical analysis: Many forms of technical analysis designed to capture trading signals exist and it is not easy to see one having a distinct advantage over another. We have chosen the Moving Average Convergence Divergence (MACD) methodology, adopting important theory from the oscillator the principle of convergence and divergence between the price and its underlying indicator. MACD comprises two exponential moving averages (EMA) a 12-period EMA minus a 26-period EMA. This difference (the MACD) is further smoothed by a 9-period EMA to form the Signal Line. Specifically, we use the MACD2-line as input, the difference between the MACD1-line and the Signal line. (see more info, see e.g. www.onlinetradingconcepts.com) Ad 2) Interest rates: We assume that rising interest rates are supporting currencies in the short run. This conflicts of course to the uncovered interest rate parity (UIP), saying that higher-yielding currencies will depreciate in order to compensate for the rate differential. But as it has been shown on numerous occasions that the UIP doesnt hold in the short run and in fact often works with opposite sign ( = -1), we feel confident in claiming that currencies with interest rates rising often faces tailwinds. Our approach in the short run stems more from a carry argument where investors buy higher yielding currencies in a forward rate bias strategy. Of course, we test this hypothesis. Specifically, we use changes in one-year swap rates as these are liquid and comparable across currencies. Alternatively, we could have used shorter-dated libor rates, but these are not fixed at same time and it is not evident that exchange rates are more influenced by interest rates with shorter maturities. Ad 3) Risk aversion: Investors suddenly attaching a higher risk premium on one currency relatively to the others is perceived negatively in our Scorecard. The assumption is that investors react by selling a given currency if they get scared for some reason. We use changes in one-month implied volatilities (measured against the US dollar) subtracted by the average level of G10 FX volatility (also measured against the dollar). This measure enables us to capture a rise in risk on a currency and take action on this at an early stage. The disadvantage is that since we need to have a common frame of reference and measure everything against the US dollar, the US currency will always end up with a zero value in this input. It doesnt mean much as everything is measured on a relative basis anyway. Ad 4) Options market positioning: We believe that we can use change in option traders behaviour actively for exchange rate forecasting purposes. For example, we interpret option traders increased preference for call options to put options in one currency as

2|

29 September 2010

www.danskeresearch.com

FX Research

belief in an impending appreciation of this currency. Specifically, we track the changes in one-month risk reversals.

G10 FX Financial Scorecard use of Z-scores and adjusting for exchange rate movements
For all four input factors on each currency we calculate the change in the past four weeks, (with 40% weight to the most recent week, 30% to the week two weeks ago, 20% to the week three weeks ago and 10% to the week four weeks ago). We could have optimised the weighting for example with respect to annualised return of the scorecard, but preferred intuition to return in this case. In order to construct an operational framework, we calculate Z-scores for each input factor. That is to make all input factors comparable so we dont compare apples and oranges. Z-scores have furthermore the advantage that the sum is zero, implying that we cannot be net short or net long the G10 currencies only neutral. In other words, we get some interpretable results that we could ideally use for forecasting purposes. We also calculate Z-scores for weekly exchange rate movements. This is done in order to capture the initial reaction to changes in our input factors. As the scorecard is intended to capture the information for our input factors that isnt instantaneously converted into exchange rate movements we need to subtract the actual currency change on a weekly basis . If exchange rate movements were entirely driven by the input factors presented above, we would end up with a series of zeros each week. The world is, however, more complex sometimes exchange rates undershoot, sometimes they overshoot. The scorecard is designed to capture where exchange rates either havent reacted enough to news in other markets, or have reacted too much. We over-compensate for the actual exchange rate movement if we subtract the full move from each input factor and then add the remaining together we simply cannot demand that all input factors move in the same direction. We therefore need to find the optimal adjustment ratio for exchange rate movements.

Fact box: Z-scores

In statistics, a Z-score or standard score indicates how many standard deviations an observation or datum is above or below the mean. It is a dimensionless quantity derived by subtracting the population mean from an individual raw score and then dividing the difference by the population standard deviation. The standard deviation is the unit of measurement of the z-score. It allows comparison of observations from different normal distributions, which is done frequently in research. The formula for the Z-score is z=(x-)/ where x is a raw score to be standardised, is the mean of the population and is the standard deviation of the population.

Chart 1: Relationship between information ratio and exchange rate adjustment factor (2006 mid-August 2010)
1.50

Information ratio

1.40 1.30 1.20 1.10

1.00 0.90

0.25 0.31 0.38 0.44 0.50 0.56 0.63 0.69 0.75 0.81 0.88 0.94 1.00

Exchange rate adjustment ratio (%)

Source: Bloomberg, Danske Markets calculations

Chart 1 shows an exchange rate adjustment ratio of little less than 70% generates the highest information ratio (1.46). As can be seen, increasing the adjustment ratio leads generally to better model performance up to this point. However, the performance isnt particularly sensitive to this ratio; everything from 50% to 90% delivers good performance.

3|

29 September 2010

www.danskeresearch.com

FX Research

G10 FX Financial Scorecard construction of weekly portfolio

We use Z-scores to construct the optimal G10 FX portfolio each week. The currency with the highest Z-score gets the highest weight, the second highest the second-highest weight and so forth. It is not pre-determined how many currencies we get long signals for and how many we get short signals for. We can in one week be long a weighted basket of three currencies and short a basket of seven currencies while long five and short five the following week if signals have changed. Chart 2 shows the average weight in our portfolio. As can be seen, the scorecard on average attaches around 40% of the long portfolio to the highest-scoring currency and the same in the short portfolio to the lowestscoring currency.
Chart 2: Average position weight in G10 FX portfolio, 2006-2010
0.5 % 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 -0.4 -0.5

Average weight

10

Higest scoring position

Source: Bloomberg, Danske Markets calculations

lowest scoring position

We observe that our way each week of picking the currencies likely to appreciate works the highest-scoring have performed better than the lowest-scoring currencies in general. The distribution is not perfect ideally we would like to see a picture similar to chart 2 but that would perhaps have been too good to be true. Second, we note that our weighting works the position we should believe the most in and therefore have attached the highest weight to, delivers the best performance. The currency we have shorted the most and given the largest numerical negative weight delivers the second-best performance. The positions between are where we are less certain on future exchange rate movements and therefore have only attached small weights to deliver modest returns, except for the fifth-highest scoring.

Chart 3: Total FX returns and weighted FX returns, 2006-2010

70 %
50

30 10 -10
-30

Weighted FX returns FX returns 1 2 3 4 5 6 7 8 9 10

-50 Higest scoring position

Source: Bloomberg, Danske Markets calculations

lowest scoring position

4|

29 September 2010

www.danskeresearch.com

FX Research

G10 FX Financial Scorecard test of individual input factors

To check whether all input factors work as intended and add information, we have developed single-input models. Table 1 summarises the results.

Table 1: Testing input factors individually

Technical analysis 27.9 6.1 8.3 0.73 -14.5 Interest rates 56.6 12.3 9.7 1.26 -11.1 Risk premium 43.8 9.5 10.2 0.93 -9.2 Options market positioning Full model 21.9 67.6 4.8 14.3 10.2 10.0 0.47 1.43 -12.5 -10.9

Total performance, 2006 - today (%) Annualised return (%) Annualised volatility (%) Information ratio Maximum drawdown (%)

Note: The same exchange rate adjustment ratio as for the full model, 0.69, has been used in testing singleinput models. Higher information ratios could have been achieved if adjustment ratios were optimised for each single-input model. Source: Bloomberg, Danske Markets calculations

As can be seen from table 1, all input factors work well by themselves and deliver decent performance some better than others. Perhaps unsurprisingly, changes in interest rates deliver the best performance also when adjusting for volatility. An information ratio of 1.26 is quite good and indicates a very robust model. A single-input model using risk premiums generates an information ratio just below 1, which also is good. A model based on technical analysis generates the lowest volatility among our input factors but has nevertheless the highest maximum drawdown and an information ratio of only about 0.7. Options market positioning is by a first glance the least valuable input factor with an information ratio just below 0.5, but we note that the performance is being generated in other periods than the other input factors and the correlation with these is low (see table 2).

Table 2: Average correlations (2007-10) between input factors

Technical analysis Interest rates Technical analysis Interest rates Risk premium Positioning 1.00 0.15 0.31 0.26 1.00 0.40 0.03 Risk premium Positioning

1.00 -0.02

1.00

Source: Bloomberg, Danske Markets calculations

G10 FX Financial Scorecard the full model

Our hope with the scorecard was originally that each input factor would work for exchange rate forecasting purposes but that the sum of these would work even better. This was by no means given beforehand so we are pleased to see that this actually functions. Our G10 FX Financial Scorecard is able to generate a total performance of 67.6% from January 2006 to September 2010, cf. table 1 and chart 4. With annualised return as high as 14.3% we can live with an annualised volatility of 10%, giving us an information ratio of 1.43. The hit ratio is also satisfying at 0.58.

5|

29 September 2010

www.danskeresearch.com

FX Research

Chart 4: G10 FX Financial Scorecard total performance

70 60 50 40
30 Total scorecard performance

20
10

0 -10 2006 2007 2008 2009 2010

Source: Bloomberg, Danske Markets calculations

Since 2006, the model has performed well in all years (see table 3). The first year was the least profitable but volatility was also substantially lower than subsequent years. Annual information ratios since 2007 have been satisfying and hit ratios not bad either. Autumn 2008 (about the time of the Lehman Brothers collapse) was an extreme period with large currency swings. The Scorecard was caught on the wrong foot here and lost around 10% from beginning of September to mid-October. It adjusted quickly though and regained more than the lost in the subsequent weeks. In contrast to, for example, carry strategies that stubbornly cling to high-yielding currencies, the Scorecard managed to turn quickly which undoubtedly is one of the key advantages with this approach. 2008 ended as the best year for the scorecard.

Table 3: Annual model performances

Model performance Return Volatility Sharpe ratio Hit ratio Maximum drawdown 2006 3.3 5.7 0.58 0.60 -5.4 2007 13.5 7.6 1.78 0.62 -5.1 2008 20.4 15.0 1.36 0.62 -10.9 2009 18.7 10.2 1.83 0.58 -4.2 2010 11.8 9.1 1.30 0.49 -4.5

Source: Bloomberg, Danske Markets calculations

It is important to emphasise that the performance is calculated without taking carry and trading costs into account. Including carry would, however, not alter the results as the Scorecard over time is carry neutral by definition and the model does not favour for example higher-yielding or lower-yielding currencies. With regards to trading costs, one has to bear in mind that around 468 trades are required a year (52 weeks, 9 currencies traded against the US dollar) as currency shares are balanced weekly. Even though shares have to be balanced it only very seldom that positions are switched 180 degrees though.

G10 FX Financial Scorecard a closer look at currency pairs

We note with satisfaction that the scorecards signals apply for all selected currency pairs (see table 4). For example, doing a long/short strategy on EUR/USD based on the scorecards signals has yielded 48.2% in total since 2006. Applying a Z-score threshold (cut off ratio) on when we follow the signals of 1 meaning we only have a position when the combined signal is above 1 or below -1 reduces the number of trades from 247 to 168 and has only marginally worse performance. If trading costs are an issue, this makes sense.

6|

29 September 2010

www.danskeresearch.com

FX Research

Table 4: G10 FX Financial Scorecard performance, selected currency pairs

Z-score threshold EUR/USD Total performance, 2006 - today (%) EUR/USD # of trades, 2006 - today USD/JPY Total performance, 2006 - today (%) USD/JPY # of trades, 2006 - today EUR/GBP Total performance, 2006 - today (%) EUR/GBP # of trades, 2006 - today EUR/CHF Total performance, 2006 - today (%) EUR/CHF # of trades, 2006 - today EUR/SEK Total performance, 2006 - today (%) EUR/SEK # of trades, 2006 - today EUR/NOK Total performance, 2006 - today (%) EUR/NOK # of trades, 2006 - today AUD/JPY Total performance, 2006 - today (%) AUD/JPY # of trades, 2006 - today 0 48.2 247 60.7 247 16.7 247 12.5 247 32.5 247 33.7 247 92.1 247 1 46.7 168 60.1 188 16.1 171 11.7 153 28.1 186 26.7 185 79.2 210 2 35.5 114 68.0 147 0.4 108 6.0 86 27.8 133 27.0 122 99.0 176 3 11.8 68 40.9 102 -4.2 67 3.4 43 16.4 90 19.6 79 55.6 133

Source: Bloomberg, Danske Markets calculations

It is, however, important to emphasise that the threshold can be set too aggressively, which of course diminishes the number of trades, but also means that performance relies on a few lucky punches. Volatility of such strategies is quite high and can generally not be recommended.

Why not just trade the highest- against the lowest-scoring currency?
A tempting thought is just to trade the highest-scoring against the lowest-scoring currency each week. More than 60% of the full models performance stems from the strongest vs the weakest signal (see chart 3). This is of course fantastic as these are signals the scorecard has most faith in, but it is also due to the fact that weights are highest for the highest- and lowest-scoring signals in the portfolio. Even though trading the highest-scoring against the lowest-scoring currency each week in total has delivered an annual performance of 22.7%, the strategy is not optimal (see chart 5). Volatility is simply too high, some 18.2%, and this returns in a lower information ratio, of 1.25, than the full model. Maximum drawdown of this strategy is 16.7% well above the 10% of the well-diversified, full model. Investors with higher risk appetite can however enter this strategy as it works well. We will also from time base trading recommendations on this signal but only when our fundamental view is aligned. Tight stop-loss levels are highly recommended.

Chart 5: Performance - highest-scoring currency against lowest-scoring currency

120 100
80 60

%
Highest against lowest scoring currency performance

40 20 0 -20
2006 2007 2008 2009 2010

Source: Bloomberg, Danske Markets calculations

7|

29 September 2010

www.danskeresearch.com

FX Research

Signals on a weekly basis

Going forward, we will publish the indications from the G10 FX Financial Scorecard each Monday morning for our clients. (An example of the output is presented in table 6.) The output will contain detailed information on how the signals are composed and include a suggestion as to how the G10 FX portfolio shall be balanced for the week. The table will also provide a wrap-up on last weeks performance. Please contact the author of this note, Senior Analyst John M. Hydeskov (johy@danskebank.dk +45 45 12 84 97), if you are interested in receiving the signals from the G10 FX Financial Scorecard. The ideal way to trade the signals from the scorecard would be via an index, for example on Bloomberg, or a total return swap. This is, however, not feasible at present and we recommend therefore to establish a G10 portfolio and balance it out each Monday, based on the signals from the scorecard. Our Sales Desk is ready to provide assistance. Please contact Senior Dealer Trine Kamph (tri@danskebank.dk +45 45 14 32 72) or Senior Dealer Henrik Wilcken (hfro@danskebank.dk +45 45 14 68 01) for details.

Table 6: Weekly scorecard output

G10 FX financial scorecard
24/09/2010 Techncial (Z-scores) Interest rates (Z-scores) Risk premium (Z-scores) Positioning (Z-scores) FX (Z-scores) Total Change from last week Suggested weight in portfolio (%) EUR 2.4 -0.2 -0.7 0.8 2.5 -0.2 1.1 -1 USD -1.0 -0.5 0.0 -1.0 -4.4 2.0 2.9 16 JPY -1.0 -0.3 2.0 -1.8 -0.4 -0.6 -4.5 -5 GBP 0.7 -0.3 0.8 -0.4 -1.9 2.8 3.5 22 CHF 0.2 -0.3 -0.8 -0.5 1.2 -2.6 1.5 -21 SEK -0.4 0.5 -0.5 -0.2 3.5 -4.1 -1.7 -33 NOK -0.4 -0.9 -1.0 1.1 3.6 -4.8 -2.4 -39 AUD 0.2 2.2 -0.6 1.2 0.7 2.4 -0.6 19 NZD -0.4 -1.2 -0.3 0.9 -2.0 1.0 -0.5 8 CAD -0.4 1.0 1.1 -0.2 -2.7 4.2 0.7 34 CAD GBP AUD USD NZD EUR JPY CHF SEK NOK
-10 -5 0 5

This week's signals (Z-scores)

Note: Table shows weekly Z-scores from each currency. A positive score indicates strength, a negative score indicates weakness. Risk premium Z-score for USD is zero par default. Total shows sum of technical, interest rates, risk premium and positioning subtracted by 2.75 times FX. Suggested weight in portfolio denotes scorecard's proposal to over- and underweighting currencies

Signals on selected currency pairs

Signal Value EUR/USD USD/JPY EUR/JPY GBP/USD EUR/GBP EUR/CHF EUR/SEK EUR/NOK NOK/SEK AUD/JPY SHORT LONG NEUTRAL NEUTRAL SHORT LONG LONG LONG NEUTRAL LONG -2.18 2.62 0.44 0.75 -2.93 2.40 3.91 4.66 -0.74 2.97 80 70 60 50 40 30 20 10 0 -10

Performance since 2006 (%) %

Note: Score below -1 triggers short position while score above 1 triggers long position. Scores between -1 and 1 suggest neutral stance.

Last week's performance

17/09/2010 Weight (share) Spot return vs. USD (%) Weighted spot return (%) Total spot return (%) JPY 33 1.9 0.6 -1.27 CAD 29 0.8 0.2 AUD 25 2.4 0.6 NZD 13 1.1 0.1 GBP -6 1.2 -0.1 USD -7 0.0 0.0 EUR -11 3.3 -0.4 SEK -20 3.7 -0.8 NOK -21 3.8 -0.8 CHF -34 2.6 -0.9

2006 2007 2008 2009 2010

Note: Spot return vs. USD measures FX performance from previous Friday 17.00NYT to Friday 17.00NYT

Performance since 2006

Total performance, 2006 - today (%) Annualised return (%) 67.54 14.32 Sharpe ratio Annualised volatility (%) 1.42 10.05

Note: Please see " A G10 FX financial scorecard" (23 August 2010) for details

Source: Bloomberg, Danske Markets calculations

8|

29 September 2010

www.danskeresearch.com

FX Research

Disclosure
This research report has been prepared by Danske Research, which is part of Danske Markets, a division of Danske Bank. Danske Bank is under supervision by the Danish Financial Supervisory Authority. The author of this report is John Hydeskov, Senior Analyst.

Danske Bank has established procedures to prevent conflicts of interest and to ensure the provision of high quality research based on research objectivity and independence. These procedures are documented in the Danske Bank Research Policy. Employees within the Danske Bank Research Departments have been instructed that any request that might impair the objectivity and independence of research shall be referred to Research Management and to the Compliance Officer. Danske Bank Research departments are organised independently from and do not report to other Danske Bank business areas. Research analysts are remunerated in part based on the over-all profitability of Danske Bank, which includes investment banking revenues, but do not receive bonuses or other remuneration linked to specific corporate finance or debt capital transactions. Danske Bank research reports are prepared in accordance with the Danish Society of Investment Professionals Ethical rules and the Recommendations of the Danish Securities Dealers Association. Financial models and/or methodology used in this research report Calculations and presentations in this research report are based on standard econometric tools and methodology as well as publicly available statistics for each individual security, issuer and/or country. Documentation can be obtained from the authors upon request. Risk warning Major risks connected with recommendations or opinions in this research report, including as sensitivity analysis of relevant assumptions, are stated throughout the text. First date of publication Please see the front page of this research report for the first date of publication. Price-related data is calculated using the closing price from the day before publication.

Disclaimer
This publication has been prepared by Danske Markets for information purposes only. It has been prepared independently, solely from publicly available information and does not take into account the views of Danske Banks internal credit department. It is not an offer or solicitation of any offer to pur chase or sell any financial instrument. Whilst reasonable care has been taken to ensure that its contents are not untrue or misleading, no representation is made as to its accuracy or completeness and no liability is accepted for any loss arising from reliance on it. Danske Bank, its affiliates or staff may perform services for, solicit business from, hold long or short positions in, or otherwise be interested in the investments (including derivatives), of any issuer mentioned herein. The Equity and Corporate Bonds analysts are not permitted to invest in securities under coverage in their research sector. This publication is not intended for retail customers in the UK or any person in the US. Danske Markets is a division of Danske Bank A/S. Danske Bank A/S is authorized by the Danish Financial Supervisory Authority and is subject to provisions of relevant regulators in all other jurisdictions where Danske Bank A/S conducts operations. Moreover Danske Bank A/S is subject to limited regulation by the Financial Services Authority (UK). Details on the extent of our regulation by the Financial Services Authority are available from us on request. Copyright (C) Danske Bank A/S. All rights reserved. This publication is protected by copyright and may not be reproduced in whole or in part without permission.

9|

29 September 2010

www.danskeresearch.com