Performance and reporting predictability of hedge funds

Abstract

This paper proposes a predictive approach to forecast future hedge fund performances and reporting stops to a commercial database within a subsequent year. We found that gradient boosting of decision trees is well suited to make a prognosis about the future development and reporting stops of hedge funds. The derived models are trained and evaluated using a panel of 5,592 individual hedge funds. We rank the impact of 22 variables that are computed out of hedge fund reporting (micro variables) and three different market environments (macro variables) on the predictability of hedge fund performance. In this way, we show the economic reasonability of the computed models and demonstrate the superiority of statistical learning algorithms.