Rooting Out Biases in Hedge-Fund Data

Daniel B. Eagan

In a recent article, I discussed the conclusions about hedge fund’s historical returns and risk we reached after rooting out biases in the data available. Here’s how we sought to eliminate those biases.

We started with the Lipper TASS hedge-fund database, which had 1,419 hedge funds reporting results at the end of 2011, and 3,502 funds that had stopped reporting. For funds that are currently reporting, we calculated compound average returns of 9.8%. Then we adjusted the results to correct for two important biases: survivorship bias and backfill bias, as the display below shows. There are also two biases we can’t correct for: small sample sizes and unreported final-period results.

Hedge Funds Have Offered Equity-Like Returns, Despite Biases in Data

Survivorship Bias. Many indices exclude the historical returns of funds that were once in their database but are no longer reporting, frequently because of poor performance. (To be fair, some funds stop reporting because very good performance has led to so much growth in assets under management that they have stopped marketing to new investors.) The Lipper TASS database maintains a separate database of so-called graveyard funds. We added back the returns of the funds which had stopped reporting, so that performance data wouldn’t be inflated by excluding the funds that didn’t survive. This reduced the index returns by 1.4 percentage points.

Backfill Bias. Many indices include returns that were reported retroactively (backfilled), in order to provide a more comprehensive series of historical returns. But many hedge-fund managers report performance for new funds only after they have a few quarters or years of success under their belts; they don’t report the results of new funds with poor results. To correct for this bias, we included fund returns only from the point the managers began reporting to the database, which cut the index’s results by another 1.1 percentage points, to 7.3%.

Sample Size. The Lipper TASS database has over 20 years of data, but in the early 1990s there were very few funds in the universe. When we corrected for backfill, there were even fewer. We therefore began our performance history in 1996, the point at which we felt we had a reliable sample of funds. After making these adjustments to the data, we found very few funds with long track records. Even today, more than 39% of the hedge funds reporting have track records of less than three years; another 23% have records of more than three years but less than five years. Only 11% of the funds could report 10 years of results, and only 0.7% have track records for the full 16 years of our adjusted database.

Unreported Final-Period Results. While hundreds of new hedge funds have been launched in some years, funds with poor results tend to stop reporting. On average, in their last 12-month periods, funds that stopped reporting performed seven percentage points worse than the index average. Their actual results may be even lower than we can measure: many funds don’t report the final periods of performance before closing.

The views expressed herein do not constitute research, investment advice or trade recommendations and do not necessarily represent the views of all AllianceBernstein portfolio-management teams.

Daniel B. Eagan is Head of the Wealth Management Group at Bernstein Wealth Management, a unit of AllianceBernstein.

One comment

  1. Hi Daniel~

    Interesting calculations. It would also be interesting to look at standard deviation and kurtosis. It turns out that hedge fund standard deviation is much lower than large cap US stocks but kurtosis is much higher meaning that when risk happens it is extreme. I have some data from Marc Odo, Director of Applied Research at Zephyr Associates, in my blog. He has done a lot of work in this area that you may find it useful. Here is the link:

Leave a Reply

Your email address will not be published. Required fields are marked *

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>