Risk Management of Hedge Funds Using Fuzzy Neural- and Genetic Algorithms

Abstract

The article investigates the use of adaptive learning algorithms in constructing dynamic portfolios replicating the return characteristics of a given hedge fund. The emphasis is on out of sample conditional predictive capabilites as necessary to serve as a valuable risk management tool, rather than simply explaining hedge fund behaviour over an in sample period. The algorithms learn dynamic trading rules and strategies along with which factors to base those on, within an integrated learning mechanism. It thus generalizes previous approaches by exploring a wide class of nonlinear and dynamic trading strategies to participate in explaining and predicting hedge fund behaviour. The conditional predictive capabilities of the algorithms can specifically be employed to quantify future fund behaviour. It will be useful in constructing quantitative risk measures for individual hedge funds. The article also provides some empirical data for out of sample behaviour of this method.

Keywords: Hedge funds, risk management, fuzzy neural nets, genetic algorithms

JEL Classification: C00,C13,C22,C30,C32,C39,C40,C45,C53,G00,G10