Do currency managers add value?
Currency Performance Analytics set about answering this question by examining the performance of 152 individual overlay programs managed by 11 firms relative to each of their respective, client-assigned benchmarks.
Comparing the performance of currency managers is devilishly difficult. Each manager has a different set of currency exposures, constraints, base currencies and benchmarks with which to work. It is rare for two accounts to have the exact same characteristics and, therefore, it is impossible to establish one common performance measure for all accounts that fairly ranks performance. However, as all managers are given benchmarks reflecting these varying mandates, we can determine overall if managers are producing value for their clients by comparing each manager's performance relative to the assigned benchmarks for each account.
In other words, while it is virtually impossible to compare one manager's value added with another's, and contrast the magnitudes of their value added for different mandates, it certainly is possible to see whether value has been created for any single account and for a broad survey of accounts.
Additionally, we looked at the correlations of managers' value added and studied the effects of different types of mandates and various types of market environments on outperformance. Specifically, we studied effects of cross-hedging, symmetrical mandates (e.g. 50% hedged) and asymmetrical mandates (e.g. unhedged and fully hedged) on excess returns.
SUMMARY OF RESULTS
Currency overlay managers have added considerable value -- about 1.9% per year on average.
Value has been added fairly consistently and over a long period of time. The first mandates were given out about 10 years ago, although the number of assignments grew significantly during the past five years. On average, 80% of the accounts outperformed their benchmarks, with an annualized average value added of 2.4%; the 20% that underperformed their benchmarks did so by an annualized average underperformance of 1.3%. The standard deviation of the average value added was 3.5%, resulting in an information ratio of 0.54. This is greater than those produced by managers in many other asset classes.
The correlation of managers' excess returns was low, 0.2 on average, although some managers had results that were completely uncorrelated with others. Those most correlated had a tendency to follow trends.
Symmetrical mandates (i.e., those that allowed managers to both increase and reduce their hedge ratios relative to the benchmark) had more consistent value added than asymmetrical mandates. The more skewed the benchmarks, the less consistent the value added. For example, when accounts managed against polar benchmarks (0% or 100% hedged) are compared with those with a benchmark hedging ratio somewhere in between (50% hedged being the most frequently used), the polar group shows a 1.6% higher standard deviation of excess return.
Less constrained mandates (e.g. those allowing cross-hedging or exposure creation) resulted in more consistent outperformance than more constrained accounts. For example, managers allowed a greater degree of freedom to cross-hedge produced about the same 2% excess return as those with more restrictive mandates, but did so more consistently -- a smaller standard deviation.
Currency overlay managers reduced risk. Total currency return of the actively managed accounts studied had lower standard deviations on average than the performance benchmarks the currency overlay managers were given.
Seventeen well-established currency overlay management firms, each with more than $1 billion of currency overlay risk under management, were asked to provide quarterly returns for all of their overlay management accounts. The $1 billion cutoff mark was established in order to confine the study to the managers most active in the institutional investment market. It was preferable to study all of the individual accounts of the managers instead of composite information, as the individual account data are both more comprehensive and reliable.
Of the 17 managers polled, 11 submitted detailed performance numbers for all of their accounts, three submitted composites, and three declined to participate. Account data were given from inception of the account through the fourth quarter of 1997, or an earlier account termination date.
Concurrently, 47 pension plans known to have currency overlay programs were polled concerning their level of satisfaction with their programs. Eleven responded, representing 25 mandates, although there surely is considerable commonality of managers. Twenty-two of the 25 mandates were reported to be beating the client-assigned benchmark, and satisfaction with the programs was universally expressed.
The results of the manager survey are presented in two parts -- that covering the individual accounts and that covering manager averages, which include the composite data provided by several managers.
The 11 firms that provided comprehensive details for all of their actively managed accounts are: BEA Associates; FX Concepts; Pareto Partners; Bridgewater Associates; GK Capital Management; Record Treasury Management; BT Australia; Goldma n Sachs; State Street Global Advisors; First Quadrant; and Panagora Asset Management.
Each firm disclosed for each quarter the: total return of the managed account, including the returns from translating the underlying investments at changing currency rates, plus the manager's returns from hedging activities; the benchmark return, including the returns from translating the underlying investments at changing currency rates, plus the returns from hedging activities produced by following the benchmark rule; and the value added by the manager vs. the benchmark for each quarter.
They further detailed each account's currency base, benchmark hedge ratio and hedging rules; no clients were identified.
A total of 152 accounts were reported on, representing more than $40 billion of currency risk under management. Account quarters totaled 1,783. (An account quarter is a quarter during which an account was managed -- thus, a manager with t wo accounts, the first run for five quarters, the other for 20 quarters would produce 25 account quarters; another manager with a single account run for 25 quarters would also represent 25 account quarters.)
When manager data are assembled on an account basis, each quarterly return data point is given an equal weighting when compiling averages.
J.P. Morgan, OSV Partners and Rothschild provided only composite data, covering a portion of their accounts. These firms reported a total of five composites, said to be representative of their aggregate management results. These composite s were excluded from calculations involving the individual accounts, but included in the manager average data in Table 6.
CPA then classified each account/composite as U.S. dollar or nondollar based; having a polar benchmark hedge ratio (0% or 100 % hedged -asymmetrical) or a mixed benchmark hedge ratio (something in between -- symmetrical); and having a res trictive (hedge only back to the home currency in the amount of the exposures) or permissive mandate (cross hedging, net shorting or exposure creation allowed).
An attempt was made to group managers according to what type of research drove their hedging decisions and examine the performance by style. However, this proved difficult as many managers did not follow a pure, single style. For example,
most managers that one might think of as being technical, also incorporated fundamental research in their process and vice versa. Some managers using options and trading on forecasts of volatility couldn't fairly be grouped with "dynami c hedgers" who methodically replicate the performance of options.
The accompanying tables refer to results calculated from the data submitted by managers for all of their accounts; no composite results were included, except for Table 6. The information displayed includes: the average quarterly return an d standard deviation for the entire time period of the included accounts; and the number of accounts in the category and the number which, on a cumulative basis, showed positive value added. That then is used to define a success ratio -- percentage of outperforming accounts for each time period. The average outperformance (win) and underperformance (loss) is computed, also producing a ratio of the average win to the average loss.
Also listed is the number of account quarters for the category and the number of account quarters in which the results were positive, again being used to determine a success ratio.
Table 1 has additional data on the total currency return and the volatility of both the managed account and the benchmark to help determine if managers contributed to currency risk reduction.
Table 1 shows the data on the broadest basis possible -- for all accounts that were submitted. The most encompassing definition of whether currency managers are doing their job is given by the overall quarterly average, which is positive and is annualized to about 1.9% per year. It stays fairly consistent in all time periods, but the volatility of the value added lessens considerably in nearer time periods -- i.e., managers are more consistently adding value recently. Two factors -- the recent path of the U.S. dollar and the increasing popularity of 50% hedged benchmarks -- contribute to this. Additionally, managers' skill levels might be improving, and the greater number of accounts in existence in recent years would weight their performance more. Currency overlay has been a rapidly expanding activity; only 24 of the total of 152 accounts existed for more than five years.
The portion of accounts with cumulative outperformance against benchmarks stays in the 80% to 90% range (the success ratio), irrespective of the length of time the account was managed. Average excess performance (the average win) is almost double or more the average underperformance or loss, although it does vary considerably.
The range of the win/loss ratio is 1.9 to 4.8. The percentage of quarters of an account's existence during which managers outperformed stays consistently around 60%, increasing slightly in nearer periods.
An important determination to make is whether the average quarterly value added of 0.47% could possibly have been due to luck. We found the chance was almost nil.
With no foreknowledge of managers' excess returns, we expect the average excess return to be 0% and can look at the probability the luck would produce an average outperformance of 0.47%. Assuming the returns are normally distributed, this probability will depend upon the actual average excess return, the standard deviation of those returns, and the sample size of data points. While there are 1,783 quarterly data points within this sample, there exists a bias for certain data points to be correlated, because multiple accounts were submitted by each manager. For example, if a manager submitted returns for 10 accounts in a particular quarter, those returns would tend to show some positive correlation, as the manager presumably makes similar bets across all of its accounts. Therefore, it could become necessary to count each quarter in which a manager submitted returns as a data point.
Each data point is essentially the average value added for all of the accounts of a manager in a particular quarter. A test of this data set indicated the probability that the results of the study are due to luck is below 0.001%. As another test, we looked at the number of positive manager quarters relative to the number of negative manager quarters, with the null hypothesis that the expected proportion of positive to negative quarters is 50%. However, through the study, we found 70% of the manager quarters were positive and 30% of the quarters were negative. The probability that these results are due to luck is less than 1%, as the observed 70% portion of positive quarters is far greater than 3 standard deviations away from the expected 50%.
A frequent focus of investors when establishing currency overlay management programs is the desire to reduce the risk that currencies present to their international portfolios. Thus, a measure of performance ought to be whether the program provided excess return at concomitantly higher risk, possibly producing an equivalent risk-adjusted return. This can be assessed by comparing the total currency returns of the managed overlay accounts with the benchmark currency returns, rather than focusing simply on the value added. Table 1 shows, in all time periods, the higher returns of managed accounts were produced at lower risk levels than those yielded by the benchmark strategy. The standard deviation of those accounts consistently was lower than the same measure for the currency benchmarks given to the managers by their clients. The managed accounts produce a consistent return/risk ratio around 0.21, which is about three times greater than that of the benchmark, again in all time periods. Active currency overlay management appears to have produced not only higher returns, but less risk, and thus a significantly higher risk adjusted currency return.
In addition to looking at all accounts together, we broke them into groups to learn more. This gave us a greater appreciation for the performance results that could be expected, based on both the market environments and the construction of mandates. Because many factors intertwine to influence the performance of currency overlay managers, it is dangerous to attempt to come to strong conclusions without controlling for each factor.
One theory for explaining the success of currency managers is that the dollar has been decidedly stronger since 1995. A majority of the overlay accounts are dollar based, and the greater portion of the accounts have only been active during this period. As the dollar rises and currencies fall, hedging has the opportunity to add greater value more consistently for U.S. dollar accounts than when the opposite is true. This is especially true for accounts managed against unhedged benchmarks, as the propensity to hedge back to the dollar is greatest in these markets. Because four factors account for the main differences in the mandates (base currency, exposures, benchmarks and hedging restrictions -- with each of the latter three having infinite variations), at a minimum, 16 combinations could be studied to assess such biases. Of course, the sample size of each subset is drastically reduced and, hence, so also is the statistical reliability of conclusions pertaining to each of these subsets.
U.S. $ VS. NON-U.S. $
Table 2 examines the difference between the performance of U.S. dollar-based accounts and nondollar accounts. Note that value added was created for both groups, implying the direction of the dollar did not account for value being added.
However, while the average quarterly excess return of each category is similar in all periods, the volatility of that excess return narrows considerably for the dollar-based accounts in the nearer time periods as the dollar's rise progressed. The standard deviation fell to 1.04% from 1.75% for those accounts, while the same measure for nondollar accounts remained quite stable, hovering around 1.6% to 1.7%. While the account data provide little additional insight because the success ratios vary widely, the numbers for the account quarters reveal the success ratio of the dollar-based accounts rises in the nearer years and the ratio for nondollar accounts remains stable. Also, managers of dollar accounts were able to produce a greater relative number of winning account quarters during the period of dollar strength.
Therefore, the relatively steady gains of the dollar during this period appear to have resulted in some degree of greater consistency. This also could be caused by other factors, such as the manager mix of dollar accounts being different or the rising popularity of symmetrical benchmarks (i.e. 50% hedged).
When accounts are managed under asymmetrical mandates, a performance bias often can arise where the ability of a manager to add value is influenced by the direction of the markets. For example, when a manager is operating against an unhedged benchmark, it is difficult to add value when foreign currencies are rising because the manager can generally not take on additional foreign exposure. Generally, under such circumstances, the best a manager can do is avoid all hedging and hold the unhedged benchmark exposures. There is little chance to add value. Conversely, when operating against a fully hedged benchmark, it is difficult for a manager to add value when foreign currencies are falling.
In order to study this relationship between the chosen benchmark and the direction of the markets, we decided to look at U.S. dollar-based accounts that were managed against either fully hedged or unhedged benchmarks. When specific market environments are isolated, a performance bias for each group clearly emerges.
Table 3 discloses the performance of accounts with fully hedged and unhedged benchmarks during quarters of dollar strength and weakness. Specifically, a quarter in which the dollar rose against a basket of currencies based on the Morgan Stanley Capital International Europe Australasia Far East index was considered a period of dollar strength, and vice versa.
The validity of our aforementioned conjecturing became apparent. For accounts that are managed to polar benchmarks (i.e., 0% hedged or 100% hedged), it indicates, on average, value can only be added when the market happens to be moving in a particular direction. Ultimately, for accounts managed against unhedged benchmarks, managers were able to consistently add a substantial amount of value during quarters of dollar strength, but generally underperformed and detracted from returns during periods of dollar weakness.
The same relationship holds true for accounts managed against fully hedged benchmarks, only in the opposite direction. Largely for this reason, the consistency of managers with symmetrical benchmark mandates is greater than those with asymmetrical ones.
Early overlay accounts were typically compared against the results from being unhedged. If a manager's maximum stray from the benchmark was 20%, then the range of hedging was from 0% to 20%. Investors began to see not enough hedging was done when it was really needed, and realized their policy position made a big bet on the direction of the U.S. dollar. The 50% hedged benchmark began to attract attention, because it was neutral on the direction of currencies and it widened the range of hedging by allowing managers to behave with some symmetry in making their hedging decisions. The same 20% deviation from a 50% hedged benchmark could produce twice the hedge ratio range -- from 30% to 70%.
Table 4 segregates accounts by whether they use a benchmark at the extremes, 0% or 100% hedged, or somewhere between those poles, a mix of 50% and other partially hedged benchmarks. The 50% ratio was the most popular. They are designated polar and mixed, respectively. Some accounts changed from a polar benchmark to a mixed one somewhere along the way; these were treated accordingly in each respective time frame.
The growth in popularity of mixed benchmarks is evident in the rising number of those accounts in recent years, as well as the rising proportion they represented of all accounts. As managers are encouraged to go "both ways" with a midpoint benchmark, you could expect they might add value more consistently, although not necessarily add greater value.
This appears to be the case, as the risk of mixed accounts is consistently lower than that of the polar accounts, and very stable in all time periods. The risk hovers between 1% and 1.2%, while the risk for the polar benchmarks ranges more widely from 1.4% to 2%. Excess return is produced more consistently with mixed benchmarks than with polar benchmarks, but with lower average excess returns. Higher success ratios for accounts with mixed benchmarks in more recent years result partially from the growing number of these accounts. However, the information ratio suggests risk-adjusted returns might be similar for each class.
An additional influence on the ability of a currency manager to add value arises from the constraints and restrictions of the client's mandate. Hedging opportunities might be restricted back to the base currency and only in the amount of the exposures in the investment portfolio, although a proxy currency might be used to hedge the united European currencies. Alternatively, clients might give a manager more permissive instructions, allowing some cross-hedging into nonbase currencies, perhaps even encouraging the creation of currency exposure if managers believe the base currency is falling, or net shorting if the opposite. With fewer restrictions, a manager might be expected to more consistently add value, because the direction of currencies doesn't necessarily influence the results.
Table 5 examines that supposition by segregating the data according to whether the manager is given a restrictive or permissive mandate.
Permissive mandates yield more consistent results -- i.e., success ratios were better. This made sense as cross-hedging allowed more uncorrelated bets. However, the amount of value added arising from permissive mandates was somewhat lower. Theoretically, there is no reason this should be the case. It could be due to exogenous factors, such as the market opportunities being greater in dollar bets or those managers who used permissive mandates being less aggressive or less successful with them.
RESULTS BY MANAGERS
The analysis done thus far was based on the performance of the 152 individual mandates. To the extent some managers have more accounts than others, the prior analysis does not show how the managers did on average for their clients. So, we looked at the performance of the managers, including the composite results of the managers who did not provide the performance of their individual accounts. As mentioned from the outset, it is impossible to compare one manager against another because of differences in their mandates.
This alternative approach to analyzing the effectiveness of currency overlay programs views all of the accounts a manager has, by averaging the excess return of all accounts, either cumulatively or for each period. This very simply states whether, on an equal-weighted average basis, a manager has outperformed, or not, for all accounts, or for the period in question. These manager averages could be further averaged, giving each manager an equal weight in determining whether managers, as a whole, are adding value. This approach contrasts with the earlier "all account" approach, which is weighted to the performance of managers with greater numbers of account quarters.
Table 6 affords a look at these average results. The average value added per quarter is positive, about 1.75% a year, and not greatly different from the earlier calculations. The percentage of successful accounts on a cumulative and quarterly basis is 80% and 62%, respectively, and once again similar to values derived on an all-account basis. Likewise, the consistency of adding value increases in nearer years, witnessed by the drop in standard deviations, and slightly rising success ratios.
A look at overall performance using this approach corroborates the conclusions drawn earlier.
Despite the attempt to avoid selection bias, it should be observed that some amount of selection bias unavoidably remains in these results. For example, survivorship bias may exist because these survey results might not include the results of accounts or managers that have ceased to exist because of poor results, although managers were asked to also provide returns for terminated accounts. We do not believe this to be material because the number of mandates and managers included in this survey represents the overwhelming majority. Knowing of some mandates that were shut down, they would have had positive value added if they were continued and included in this survey.
Further, while the results are presumed to be reliable, they were not audited, so it is possible they are superior to the total universe of all accounts. As noted earlier, in an attempt to broadly verify results, we surveyed the institutional investors to ascertain whether they derived value added. Because not all replied, we could not obtain a total verification. However, 22 of the 25 that were covered in their replies showed positive added value and 100% of the investors surveyed were satisfied with the results.
There is no reason to believe those who replied to the survey were any more satisfied than those who did not reply; in fact, given the propensity of people to complain, one might be inclined to think that those who were most dissatisfied would have been most inclined to reply. Therefore, while we presume some amount of selection bias in these results, we believe it to be very small and not material to the conclusions summarized at the beginning of this study.