Long Horizon Multifactor Investing with Reinforcement Learning
56 Pages Posted: 12 Aug 2022 Last revised: 17 Sep 2022
Date Written: August 10, 2022
We provide a novel approach to multifactor investing for long term investors who take into consideration medium- to long-term volatility and liquidity risks, while minimizing long-term portfolio level rebalancing needs. Conditioning on multiple factor characteristics and macro-economic states, reinforcement learning (RL) approach allows explicitly imposing multiperiod dynamic volatility, illiquidity and turnover constraints for the overall portfolio level. We find that training the model under explicit long-horizon holding investment period considerations and low frequency rebalancing, which can only be accommodated via RL, dramatically changes perspective of long-term investors' portfolio performances vis-à-vis their short-term peers. Not only we find that a portfolio performance of long-term investor can remain competitive to those of short-term investors, we show that it can substantially outperform the latter, while further benefiting from lower turnover and trading costs.
Suggested Citation: Suggested Citation