Dynamic Portfolio Management Based on Pair Trading and Deep Reinforcement Learning

Abstract

Existing portfolio management methods have made great progress in diversifying non-systematic risks, but they have ignored systemic risks. In response to this issue, we proposed a dynamic, market-neutral, risk-diversified portfolio management model by combining the ideas of pair trading strategy, deep reinforcement learning with traditional portfolio management model. We conduct an experiment on the Chinese A-share market by selecting 32 pairs of stocks. The experiment results showed that the proposed pair-based deep portfolio model has superiority for dynamic portfolio management problem in trade-off investment returns and risks.