Hedge Fund Return Predictability in the Presence of Model Uncertainty and Implications for Wealth Allocation

This paper studies hedge fund return predictability in a multivariate setting. Our research design and analysis is motivated by the empirical observations that a specific forecasting model that is going to perform well is not known ex-ante and that modelling time varying return covariances/correlations improves our ability to construct optimal hedge fund portfolios. We employ a multivariate GARCH specification to model time-varying covariances/correlations of hedge fund returns and we develop a stochastic search algorithm to compute posterior model probabilities for alternative predictive specifications. Our empirical analysis indicates that introducing dynamic covariance/correlation modeling improves the out-of-sample performance of optimal hedge fund portfolios. Moreover, introducing predictive factors provides incremental additional portfolio outperformance