Commercial hedge fund data are likely to overestimate the true performance of the universe of hedge funds. The authors suggest that the upward bias in the return data is the result of databases missing the worst-performing funds because of the voluntary and selective nature of reporting. Some of it is also the result of a delisting bias—not chronicling the poor performance of funds following delisting—because managers have discretion over timing and removal of a fund from a database.
Extending the current research on selection bias inherent in commercial hedge fund databases, the authors investigate the direction and size of the bias. Using a previously unexplored dataset of funds of hedge funds (FOFs) that have not reported or stopped reporting to commercial databases, they observe a considerable positive selection bias. Funds that report their performance to commercial databases (database funds) significantly outperform, on average, non-reporting funds (non-database funds).
How Is This Research Useful to Practitioners?
One of the problems with current hedge fund literature is the mainstream reliance on commercial databases to evaluate performance. But hedge fund managers can choose not to report to databases, which creates a selection bias. The result is that researchers using self-reported data do not observe the full spectrum of returns.
Drawing on holdings data of FOFs, the authors construct a distinctive set of returns for hedge funds that do not report to a database. They find by using database data that estimates of managerial skill are not representative of the universe of hedge funds and that the estimates have an economically significant positive bias. They report that databases are missing some of the worst funds. Furthermore, they find that the omission from the database of successful funds does not sufficiently offset the poor performance of the worst database funds to overturn the central tenet of their findings. They conclude that if returns of database funds are excluded, the average excess return of hedge funds does not differ strikingly from zero.
With a deeper understanding of the extent that performance data in commercial databases are prone to be skewed by selection biases, investors and policymakers can learn to better use such data to inform capital allocation or comprehend systematic risk. The findings also appear to challenge the notion that a hedge fund delivers truly superior returns for its investors.
How Did the Authors Conduct This Research?
The authors use regulatory filings from FOFs that registered with the U.S. SEC. Since 2004, registered FOFs have been required to disclose their underlying hedge fund holdings on a quarterly basis. Using these data, they create a sample of 10,126 quarterly returns from 1,445 distinct hedge funds for 2004–2009.
By tying these funds to the amalgamation of five major commercial hedge fund databases—Lipper TASS, Hedge Fund Research, BarclayHedge, Morningstar (which incorporates CISDM and MSCI databases), and EurekaHedge—they observe that about one-half of the sample’s returns are not disclosed to any of the five databases. By using returns taken from investors (from the SEC filings) rather than from the hedge funds themselves (from the commercial databases), the authors’ findings present a rare opportunity to compare directly the performance of database funds and non-database funds by using data largely free from selection biases.
By computing risk-adjusted returns using different factor pricing models, they find the alpha of database funds ranges from a statistically significant 72 bps a quarter to 120 bps a quarter. This finding is comparable in magnitude to the 3–5% annual alpha found in prior research of database funds. In contrast, the alpha of non-database funds drops to a statistically insignificant 5 bps a quarter.
To uncover evidence of a delisting bias, the authors identify funds in their sample that have delisted from the commercial databases but continue to operate subsequently for some time; 85% of delisted funds have at least two quarters of returns subsequent to the delisting date. On average, they find returns of delisted funds are −16 bps a quarter, 184 bps a quarter lower than the returns of funds that continue to report to databases. This evidence is indicative of a delisting bias because poor returns of delisted funds are missing from the commercial databases.
The authors should be credited with creating an innovative methodology of using the underlying return data from registered FOFs. This research seems to offer added arguments for industry critics who say hedge funds do not proffer market-beating returns. Because of certain limitations in the dataset used, such as limited date ranges and the unknown size and composition of hedge funds that are both missing from a database and not held by a registered FOF, it is difficult to draw definitive conclusions to resolve this controversy. The findings could arguably be a critique more of the reliance on commercial databases to study hedge fund performance rather than of the performance of the hedge fund industry itself.