|
Conditional performance of hedge funds Authors: Hossein Kazemi and Thomas Schneeweis Source: CISDM, Working Paper Date: February 2003 |
This study provides an interesting contribution to the search for an efficient hedge fund performance measurement model. According to the authors, the distributions of hedge funds returns are neither normal nor identical through time. That is why the traditional performance evaluation models are not appropriate.
Kazemi and Schneeweis propose a conditional model of performance, based mainly on previous work by Chen and Knez (1996) and Cochrane (2001): the Stochastic Discount Factor model. The SDF model has a principal advantage in that it permits the fact that the relationship between hedge fund returns and primary asset classes is time varying to be taken into account. The major assumption behind the SDF approach is the absence of arbitrage in financial markets. Under this condition, the SDF is a positive random variable which adjusts future payoffs for the passage of time and uncertainty.
If there are no arbitrage opportunities, it is always possible to find a portfolio of available assets in order to mimic the behavior of the Stochastic Discount Factor. In this paper, the SDF is represented by a portfolio of primitive assets, where the weights are estimated so that at least the primitive assets themselves are correctly priced by the model.
The SDF model is applied to two different sets of data. The first data set consists of monthly returns on Hedge Fund Research (HFR) hedge fund indices covering January 1990-December 2001. The second data set (which consists of monthly returns covering January 1995-December 2001) was created by the authors through a combination of 5 hedge fund databases: HFR, CISDM, Altvest, Hedgefund.net and TASS. The aim is to avoid the problem of the manager's self-declared style not truly belonging to the particular investment style: here, the style classifications are verified, and after that the funds are assigned to 6 equally weighted portfolios (convertible arbitrage, hedged equity, distressed securities, event driven, merger arbitrage and equity market neutral). In the next step, at the beginning of each year, large and small fund series are created for each investment style, according to the size of assets under management. The "primitive" assets are represented by four Fama and French indices, based on stocks, high yield bonds, long-term bonds and Treasury bills.
The first results, which come from the first set of data, concern hedge fund indices. The study confirms that the return distributions are not normal for some strategies. The risk-adjusted returns obtained through Jensen's model are significantly positive and the estimated alphas are significant. These results are approximately the same whether a single-benchmark or a multi-benchmark framework is used. Measures based on the Stochastic Discount Factor approach show similar results for all strategies.
The other results, which use the second set of data, relate to hedge fund managers. The single factor provides similar alphas whichever benchmark is employed, and the size is not a determinant in terms of performance. Using a multi-factor model and an SDF approach, the estimated alphas are close to the alphas given by the single factor model, except for large hedged equity, small hedged equity and large convertible arbitrage funds, whose alphas are more significant using the SDF model.
Kazemi and Schneeweis therefore conclude that the set of primitive assets and conditioning variables that they use are not capable of capturing the type of trading strategies followed by most hedge fund strategies.
