Luck or skill?
Arun Muraldihar aims to prove a point
One of the critical issues that has an impact on the asset management business is whether managers can outperform their benchmarks over long time horizons. Even when such positive performance is generated, there is a nagging doubt that the outperformance has been achieved by luck rather than skill. At a third level, one also wants to know whether outperformance persists when returns are adjusted for risk.
In the area of active currency overlay, managers are constantly faced with the comment that, over the long term, currency returns are zero and hence there are no excess returns to be had. This article sets out to answer the second question and to ask whether excess returns in currency are significantly different from zero on an unadjusted and risk-adjusted basis. In short, it seeks to answer many of the questions raised by Muysken and Collins (1998).
There has been some research to lay out a methodology that allows one to evaluate whether active managers have generated excess returns from skill or luck. Ambarish and Seigel (1996) argue that the minimum time to determine whether outperformance comes from skill is a function of the degree of certainty required, actual and benchmark portfolio return, risk and correlation. They conclude that the greater the certainty required and the lower the correlation between the returns of the benchmark and the portfolio, the greater the time required. They also demonstrate that, for a portfolio that has outperformed by an annualised 300 basis points, where the annualised standard deviation of the benchmark is 15%, the annualised standard deviation of the portfolio is 25% and the correlation is 0.9, one would need 175 years of data to be 84% sure!
Since it is rare to obtain so many data points in the asset management business, Muralidhar and U (1997) extend the analysis to ask the question in reverse – given a limited data set, can we determine what degree of certainty do we have that the outperformance comes from skill? In addition, Modigliani and Modigliani (1997) introduced a risk-adjusted performance measure, whereby portfolio performance is adjusted for “leverage” taken relative to a benchmark.
In this article we use all three approaches to examine currency overlay accounts. We examine the performance of 14 accounts managed by the currency overlay group at JP Morgan Investment Management (JPM). JPM uses a diversified strategy with respect to currencies – namely, it has a currency model that incorporates both fundamental and technical evaluations of the relative value of currency pairs. Different aspects of the literature have looked at the attractiveness of using fundamentals and technicals and also the advantages of diversified strategies.1
While JPM manages many more mandates, the rationale behind this selection was to include only those accounts with an inception period that provide us with at least four years of data (ie, inception prior to May 1995). One of the advantages is that some accounts have been managed for the past decade and hence there is sufficient data to perform this analysis. In Table 1, we provide some basic information about the mandates including inception date, hedge mandate (fully hedged, unhedged, etc.), and whether the mandate has been changed since inception.2
We will examine the performance of these accounts and not only determine whether the results have come from luck or skill, but also ask whether the nature of the mandate or the start date provide information on the final outcome. Equally important, we would like to know whether on a risk-adjusted basis currency has added to or detracted from total returns.
In Table 2, we provide performance information on the portfolios – namely return and standard deviation of the benchmark and the actual portfolio. In Table 3, the information is distilled in terms of the excess returns – namely, we provide annualised excess returns, the annualised tracking error and the correlation of the benchmark and the actual returns. It is interesting to observe that the annualised excess returns range from 4bps to 335bps, while the correlations range from 0.39 to 0.96. The data that is available to us for these analyses range from approximately four years to approximately 10 years. If the conclusions of Ambarish and Seigel (1996) or Roll (1992) are to be believed, this amount of data is likely to be insufficient to make a comment with any degree of confidence that there is skill in the currency overlay business.
In Table 4, we make the adjustment for the fact that the volatility of the actual portfolio is different from that of the benchmark (also called the M-2 measure) and provide risk-adjusted annualised outperformance.3 This measure is in units of performance and can be compared to the unadjusted annualised excess returns. We can conclude that all portfolios continue to outperform even on a risk-adjusted basis and in only three mandates was more risk taken relative to benchmark to generate alpha. Also, the ranking of alpha is changed as the risk-adjusted alpha is lower for client 3 and increased for client 10.4 This is consistent with accounts that have different hedge mandates. The fully hedged benchmark (with no ability to more than fully hedge) has the lowest currency risk and active managers must take on more risk to add alpha.5
In Table 5 we provide information for each client on how confident we are that performance has come from skill6 or alternatively how many years of data would we require to determine that we can be 94% confident, 84% confident or 67% confident that this outperformance is from skill.7 The confidence in the outperformance of currency overlay accounts is astounding – ranging from a low of 58.83% to a high of 99.84%. Further, eight of the accounts – or more than half – have a confidence rating above 90%, and all but one have a confidence rating above 80%. Even the two accounts with the lowest correlation (clients 1 and 3) have a confidence rating above 90%! This in some part can be attributed to systematic outperformance over a relatively long history. This table indicates that most of these clients can be extremely confident that the outperformance that they have achieved comes from skill.
The “poorest” performing mandate on this metric (client 4) has been a client for a relatively long period and has a portfolio that is highly correlated with the benchmark, but only outperformed by 4bps. However, this was a portfolio that was managed relative to a fully-hedged benchmark – which was the best performing benchmark of those provided in Table 2. This mandate was also undertaken over a period of relative dollar strength, during which it was difficult to add value as currency overlay managers could not more than fully hedge such accounts.
The third observation is that there is greater confidence in the accounts initiated in 1989–90, than those in 1992–93 period even though the annualised outperformance for the latter group is higher. Therefore, the longer history and potentially lower tracking error for the 1989–90 accounts have compensated for lower excess returns. The same, however, does not hold for a comparison between the 1992–93 initiated accounts and those from 1994–95. In this case, one can say with a very high degree of confidence that the 1994–95 accounts’ excess returns came from skill, ie, the higher annualised excess return for the latter group, with lower tracking error compensates for the shorter data history. In the context of currency markets, portfolios that were live during the 1994–mid 1995 period experienced the most significant deviation away from fundamental equilibrium levels, especially Japanese yen, and hence the 1992–93 annualised tracking errors appear to be significantly higher.
One can ask whether this analysis is flawed as it appears to be inconsistent with the earlier suggestion that one would need 175 years of data. The reason we get a divergence in these results from the conclusions of Ambarish and Seigel (1996) and Roll (1992) can be attributed to a number of factors. However, the most significant of these is that
q the benchmark and actual portfolio standard deviations in currency are substantially lower than in other asset classes;
q the portfolios are reasonably highly correlated with their benchmarks and hence the tracking error is not too high; and
q in many cases, the risk of the actual portfolio is lower than that the risk of the benchmark.
Esssentially, since currency overlay produces the same alpha as that of other asset classes, but does so with significantly lower risk in the benchmark or the actual portfolio, the payoff for a plan sponsor per unit of risk budget is extremely attractive. We were able to demonstrate that even on a risk-adjusted basis, the portfolios continue to outperform. Further, given the high degree of confidence in the results (eg, 80–99%), one can conclude that the diversified strategy of using fundamentals and technicals captures the nature of the currency markets.
Arun Muralidhar is head of currency research at JP Morgan Investment Management. This article does not represent the views of JP Morgan or any of its affiliates. All errors are those of the author. Thanks to Harriett Richmond, Paolo Pasquariello and the currency overlay group for valuable input and comments