Hedge fund replication with a genetic algorithm: breeding a usable mousetrap

Abstract

This study tests the performance of 14 hedge fund index clones created using parsimonious out-of-sample replication portfolios consisting solely of easily accessible assets. We employ a genetic algorithm to integrate two traditional hedge fund replication methods, the factor-based and pay-off distribution replication methods, and evaluate over 4500 commonly held stocks, bonds and mutual funds as replicating portfolio components. In-sample performance indicates that hedge funds have return series similar to portfolios of commonly held assets, and out-of-sample results provide evidence that the in-sample relationships can hold with infrequent rebalancing. This hedge fund replication attempt rates well relatively to prior efforts as 11 replicating portfolios have out-of-sample correlation values of at least 60%. Overall, these results show promise for using a genetic algorithm technique to replicate hedge fund returns.