Hidden Survivorship in Hedge Fund Returns

Abstract

This study identifies a previously unreported bias in the TASS database. Owing to a merger with the Tremont database, 60 percent of the funds added to the TASS database between April 1999 and November 2001 are likely to be survivors (i.e., funds that were selected only from funds that were live as of 31 March 1999). The resulting survivorship bias is substantial, averaging more than 5 percent a year. What would normally be termed the backfill period actually represents hidden survivorship. A sorting algorithm to exclude these fund histories is proposed.

In this article, the authors identify a previously unreported bias in the Tremont Advisory Shareholder Services (TASS) database. In March 1999, Tremont Capital Management purchased the original TASS database. Before the purchase, Tremont had compiled its own informal database of hedge funds. That database was not absorbed directly into the TASS database. Instead, Tremont invited its hedge fund managers to join the new TASS database. Not all managers joined the database immediately; this process continued gradually until November 2001, when the last of the Tremont funds were added. As a result, a large number of Tremont funds were added to the TASS database between 1 April 1999 and 30 November 2001.

Only funds that were still operating as of 31 March 1999 or later—that is, survivors—accepted Tremont’s invitation to report to the TASS database. Some 60 percent of the funds added to the TASS database between April 1999 and November 2001 are likely to be survivors. Existing methods of dealing with survivorship bias (e.g., including dead funds in the sample) are insufficient to correct for the Tremont survivor problem in the TASS database because at the time of the TASS/Tremont merger, there were no Tremont dead funds to include. The authors find that the size of the survivorship bias arising from the TASS/Tremont merger—measured as the difference between the performance of the survived funds and the rest of the sample—is more than 5 percent, on average, for 1994–2001.

Because of the database construction process, the survived Tremont funds are likely to exhibit long backfill periods. But the long backfill periods are not a result of hedge fund managers’ purposefully backfilling the data; rather, they are a spurious effect owing to the design of the database. Therefore, what would usually be called backfill bias really represents survivorship bias. To eliminate the effect of this survivorship bias in empirical analysis, the authors propose a sorting algorithm to exclude these fund histories. Their results should be useful to researchers who increasingly use the TASS database to evaluate the performance characteristics of hedge funds.