Hedge Fund Performance: End of an Era?

Abstract

This article documents a decline in aggregate hedge fund performance over the past decade. We tested whether a set of prediction models can select subsets of individual funds that buck the trend and subsequently outperform. Two of the predictors reliably picked funds that lowered the volatility and raised the Sharpe ratio of a multi-asset-class portfolio relative to a stock/bond portfolio over the full 1997–2016 sample. Hedge fund allocations reduced volatility in two subperiods but failed to improve the Sharpe ratio from 2008 onward. We explore potential explanations for the erosion of hedge fund performance.

Declaration of Interests

Disclosure: The authors report no conflicts of interest.

Acknowledgments

The authors thank Antti Ilmanen, Olga Kolokolova, Peter Nyberg, Russ Wermers, and seminar participants at Boston College, Northeastern University, Texas A&M, the University of Arizona, Vanderbilt University, the 2017 Chicago Alternative Investments Research Conference, the Office of Financial Research, the 2018 Midwest Finance Association Meeting, and the 10th Annual Hedge Fund and Private Equity Research Conference.

Editor’s Note

Submitted 28 October 2020

Accepted 21 April 2021 by William N. Goetzmann

Notes

¹ The index earned 7.1% annually, versus 2.2% for the five hedge funds selected. See Burrows (2018).

² We also executed all analyses by using other allocations within the stock/bond portfolio and obtained qualitatively identical results. Similarly, as discussed later in the article, we demonstrated robustness to the allocation weight on hedge funds in the multi-asset-class portfolio.

³ Although the Fung and Hsieh (2004) seven-factor model has become the standard for estimating risk exposures and managerial performance, in most hedge funds, this model leaves more than half of the variation in returns unexplained. Therefore, establishing an appropriate benchmark is difficult.

⁴ Agarwal, Daniel, and Naik (2009) showed a positive relationship between the dollar value of a manager’s performance-based compensation and subsequent hedge fund returns.

⁵ Alternatively, the relationship between the ranking-period predictor and holding-period performance can be measured in a cross-sectional regression. See, for example, Goetzmann and Ibbotson (1994).

⁶ Investors who use the predictors to avoid the worst funds do, in effect, benefit, but they ultimately need to invest somewhere.

⁷ See J.P. Morgan (2018). In addition, evidence in Teo (2013) and Brown, Gregoriou, and Pascalau (2012) indicates that a portfolio of 15 funds achieves the majority of possible diversification benefits. Consistent with these results, we found, in unreported analysis, that funds in the 15-fund portfolios we created featured an average correlation of 0.27. Furthermore, we found qualitatively similar results when we increased the number of funds to 30.

⁸ Jackwerth and Slavutskaya (2016) also conducted a simulation exercise of the benefit of hypothetical allocations to hedge funds for a sample of UK pension funds. For each pension fund, they simulated a hedge fund allocation by selecting a single subset of hedge funds at random and measured the change in the pension fund’s performance. In contrast, we repeated the random selection many times to construct a distribution of outcomes, which permitted a rich characterization of the investor experience.

⁹ We show later in the article that our results are robust to increasing this AUM requirement.

¹⁰ Technical details are available from the authors.

¹¹ We used a 24-month rolling window in estimating our predictors for two main reasons. First, a longer window would potentially introduce both reverse survivorship bias (Linnainmaa 2013) and multiperiod sampling bias (Fung and Hsieh 2000). Second, prior literature documented that hedge funds’ risk exposures vary from month to month (Bollen and Whaley 2009) and even within months (Patton and Ramadorai 2013). Hence, a longer window would likely result in a larger estimation error for many of the predictors.

¹² Note that the alpha of the S&P 500 was not measured, because it is one of the Fung and Hsieh (2004) seven factors. Also note that we treated the three-month LIBOR as a risk-free asset, so we report only its average value.

¹³ An additional reason for measuring performance by quantifying the utility of a return series is to avoid potential factor model misspecification, as noted in a performance prediction context by Carhart (1997). Suppose, for example, that a factor model omits a relevant risk factor. To the extent that exposure to the omitted risk factor is serially correlated, the alpha will be as well.

¹⁴ As noted by Bali et al. (2013), a benefit of the MPPM over standard performance measures is that it incorporates nonnormalities in hedge fund return distributions.

¹⁵ In response to a referee question, we also considered accommodating binding capacity constraints by measuring the benefit of investing in top-quintile actively managed equity mutual funds, as ranked by the Carhart (1997) alpha, instead of hedge funds. We found the same qualitative results with actively managed mutual funds—namely, a modest benefit in the early period (although not as large as with hedge funds) but no benefit in the later period. This finding is consistent with results in Choi and Zhao (forthcoming). In the interest of brevity, we do not report the results, but they are available on request.

¹⁶ In unreported analyses, we considered several other combinations of predictors, including composite rankings from the top individual predictors in the first half of the sample, which were tested out of sample in the second half. Results were qualitatively unchanged.

¹⁷ To improve the estimates of risk premiums for stocks (the S&P 500) and bonds (the VBTIX), we used monthly returns starting from February 1972 to estimate them. The bond returns were spliced in such a way that starting from October 1995, we used the VBTIX, and before October 1995, we used its benchmark index (Bloomberg Barclays US Aggregate Bond Index).

¹⁸ We used a number of alternative performance criteria and obtained similar results.

¹⁹ See, for example, Jakab (2012) and Fletcher (2015) in the Wall Street Journal.

²⁰ Increased correlations between risky assets may explain why Sullivan (2021) found a decline in the standard error of alpha in an equity hedge fund index. He labeled this risk “active risk” because the returns of funds’ equity allocations were more fully explained by systematic risk.

²¹ We also conducted the analysis by using the spread between top and bottom quintiles and obtained qualitatively similar results.