|
The mortality of funds of hedge funds Authors: Greg N. Gregoriou Source: The Journal of Wealth Management Date: Summer 2003 |
Survivorship bias is one of the potential biases that affect databases. Survivorship bias occurs if the database only contains information on "surviving" funds. Since this bias has a positive impact on returns, it is interesting to analyse the mortality of funds as part of performance measurement. In this paper, Gregoriou conducts a survival analysis that focuses on funds of hedge funds (henceforth FoHFs). The data used, provided by ZCM, covers the period from January 1990 to December 2001. It contains 344 live and 191 defunct funds.
The effects of several predictor variables on survival time are examined. These covariates are average monthly return, average millions managed, age, performance fees, management fees, leverage, redemption period and minimum purchase.
Endpoints occur if funds stop reporting to ZCM for three consecutive months. Censored funds are funds that are still alive at December 2001. FoHFs born from January 1990 on and dead before December 2001 are included in the survival analysis, as are censored funds, in order to avoid a downward bias of survival time.
Three types of methods can be distinguished: non-parametric methods, semi-parametric methods and parametric methods. The non-parametric methods employed by the author are the Kaplan-Meier estimator and the life table method. One semi-parametric method is used, namely the Cox proportional hazards model. An example of a parametric model is the accelerated failure time model.
A life table exhibits a median survival time of 7.45 years at the beginning of the period. Using Kaplan-Meier estimates of survival times, it appears that the greater the assets under management, the longer the mean survival time. Focusing on the minimum purchase, the results are less homogeneous. Considering cutoffs of $25,000, $50,000 and $100,000, the larger the minimum purchase, the higher the mean survival, while an inverse relationship is observed when considering cutoffs of $250,000, $500,000, $1,000,000 and $2,000,000.
The hazard function shows that the risk of failure varies according to the survival in years. There are peaks between 2 and 3 years, between 4 and 5 years, and between 7 and 8 years. A decreasing trend is observed from the first peak to 8 years, and after that the risk of failure increases.
Log-rank tests are conducted, on the basis of Kaplan-Meier estimates of survival times, for several covariates. Cutoffs are determined by the median. It appears that the funds that survive longer present the following characteristics: assets under management greater than $14.9 million, average monthly returns higher than 0.82%, performance fees higher than 20%, a minimum purchase higher than $250,000, low leverage and an annual redemption period. Management fees seem to have no impact on the survival times.
The results on assets under management are reinforced by the fact that the proportion of dead funds decreases when the assets under management increase. Using the Cox proportional hazards model and the accelerated failure time model with Weibull distribution leads to a conclusion on the significant impact of the amount of leverage, the level of minimum purchase and mean monthly returns on survival time. The higher these covariates, the longer the survival times.
