Hedge Fund Replication: Linear and Non-Linear Techniques with Broader Factor Categorizations

Abstract

Certain literature has attempted to replicate hedge fund returns using “clones” (de- fined as passive hedge fund replicators) that utilize various factor models and weighting techniques. In this thesis, we study and extend the methodologies from seminal papers in the field and address their shortcomings. We investigate the linear models used by Hasanhodzic and Lo ([13]) and the nonlinear models used by Amenc et al. ([2]), specifically Markov Regime Switching and Kalman Filter methods. Then, we introduce the LASSO factor selection model used by Giamouridis and Paterlini ([10]) along with broader factor options. We use the new model and factor data in combination with the linear and nonlinear methods from previous literature. By applying 14 types of these hedge fund clones to 10 different investment strategies with 11 asset exposures, we find that these clones still underperform the corresponding funds based on both in-sample and out-of-sample tests. We thus demonstrate the need for a more comprehensive factor selection model and replication strategy for these clones to succeed in future research.