The database used, provided by Financial Risk Management, covers the period from January 1992 to December 2000. It contains both surviving and defunct funds. Hedge funds are sorted into six strategies, depending on the investment process, the asset class and the geographical region: Directional Trading, Traditionally Directional, Stock Selection, Specialist Credit, Relative Value and Multi-Strategy.

Firstly, the authors consider the relative performance persistence of individual hedge funds during the period from January 1992 to December 2000. This period is split into equal length sub-periods of either 1, 3, 6 or 12 months. The performance is defined as the cumulative return computed over a sub-period. The median performance of the funds is used as a benchmark. A binomial test at a 5% level is carried out with the following null hypothesis: the fund performance is equally distributed on each side of the median. It appears that high and low returns have more persistence in the short term (1 month) than in the long term (12 months). Moreover, the Specialist Credit and Relative Value strategies dominate the others across the different time horizons in terms of the proportion of funds that perform above the median.

Secondly, the performance persistence of hedge fund portfolios is investigated. Over the whole period, formation and holding sub-periods of 1, 3, 6 , 12, 18 and 36 months are used. An interesting point is that 10 equally-weighted portfolios are created, where the top 5 portfolios contain the 100 best-performing funds only (20 funds per portfolio created), and the other 5 portfolios include the 100 worst-performing funds only. As opposed to a traditional "deciles" approach, this allows portfolio heterogeneity to be avoided over time, because of the growing number of funds entering the database. A two-sample t-statistic test is carried out at a 5% level in order to compare the average monthly return of the best 5 portfolios to the worst 5 portfolios. For a formation and a holding period equal to 1-1 month and 1-3 months, the best portfolios significantly outperform the others, but if the formation and holding periods increase there is no significant persistence. On the other hand, the best portfolios show a downward tendency as the length of the holding period increases, while the worst portfolios exhibit an upward tendency.

Finally, the performance persistence of hedge fund portfolios is tested on a risk-adjusted basis. The performance measurement is based on Ross's (1976) Arbitrage Pricing Theory. Even though the APT is not appropriate for capturing the non-linearity of hedge fund returns, factors which take the non-linearity of the returns into account are included in the model. The factors are determined via Principal Component Analysis. 8 factors, with model explanatory power of 79% to 90%, are selected. Due to the dynamic trading strategies of the hedge funds, the whole period is split into three sub-periods of equal length. In panel A, the formation period is from January 1992 to December 1994, and the holding period is from January 1995 to December 1997. In panel B, the formation period is from January 1995 to December 1997, and the holding period is from January 1998 to December 2000. The funds are ranked according to their abnormal performance in the holding period. For all the different strategies at an aggregated level, the abnormal performance of the 5 best portfolios decreases from the formation period to the holding period. In lower proportions, the opposite phenomenon is observed for the 5 worst portfolios. In panel A, the best portfolio (containing the best 20 funds during the formation period) remains the best in the holding period. That is not the case for panel B, where the best portfolio in the formation period is the worst portfolio in the holding period. This may indicate that the selection of persistently winning funds depends on the formation period. Considering the six strategies separately, the tendency of the abnormal performance from the formation period to the holding period differs according to the strategy. The Traditionally Directional and Stock Selection strategies show a reverse tendency: the top portfolios have a downward tendency, while the worst portfolios have an upward tendency. According to the authors, the Multi-Strategy strategy exhibits the strongest continuity of its risk-adjusted performance, but no statistical test is conducted in this part of the study. The results therefore have to be considered with care.