Enhancing hedge fund index tracking factor model via deep reinforcement learning

Abstract

Hedge funds are getting attention as an alternative asset with their attractive performances including absolute return. However, some obstacles still exist regarding liquidity, accessibility and drawdown risk. In this article, we present an advanced framework for index tracking factor model based on integrated factor set and dynamic regularized regression. Rule-based parameter tuning is suggested and tracking error decreased meaningfully. Next, we bring reinforcement learning to factor investment in order to manage drawdown risk of hedge fund index. Deep Q-Learning is utilized to manage maximum drawdown of factor model along with new and delicate definition of reward. In addition, we suggest a modified algorithm by inserting “select-to-update” stage, which makes agent to converge more efficiently. Our DQN based factor model definitely enhanced the return and risk-adjusted return of fund.