Analysing nonlinear systematic risk exposures in hedge funds

Using a nonparametric statistical methodology this thesis analyses nonlinear risk exposures in portfolios and individual hedge funds. At the portfolio level an out-of-sample evidence of nonlinearities is documented in most of the styles; however, nonlinear features are found to be more pronounced in arbitrage related hedge fund styles, rather than in directional styles. A nonparametric approach based on a Generalized Additive Model (GAM) captures nonlinearities better than the widely accepted seven-factor Fung and Hsieh (2004b) model and outperforms linear multi-factor models in out-of-sample tests. At the fund level, one-fifth of funds exhibit significant nonlinearities detected using GAMs. In addition, individual funds with nonlinear risk exposures have on average lower raw and risk-adjusted returns and higher left tail risk than funds with only linear risk exposures. Thus, nonlinearities do not signal skill among fund managers. Finally, linear and nonparametric models are employed to replicate broad hedge fund benchmarks as well as investable hedge fund indices. It is found that the nonparametric model better tracks hedge fund benchmarks than the linear model, confirming the importance of nonlinearities.