Research on the Dynamic Hedging Trading Strategy for Portfolios Based on Deep Reinforcement Learning

For the current stage of the portfolio strategy for the overall market downside risk that is, systemic risk perception is not obvious, the investment process using only the portfolio strategy will be exposed to too much risk. Meanwhile, in order to flexibly respond to market changes, this paper establishes a dynamic hedging trading strategy, dynamically calculates the hedging ratio and investment weight through different adjustment cycles, and establishes an alternative pool of hedging assets to dynamically select suitable hedging assets through volatility. Using the Actor-Critic algorithm based on deep reinforcement learning; and using the Long Short-Term Memory network to predict the probability of the action; adopting the experience replay technology and the online learning method to randomly sample the past experience; The experimental results show that the dynamic hedging trading strategy on the testing set adjusted with a 30-day period improves the Sortino ratio by $692.94 \%$, reduces the downside risk by ${8 7 . 9 9 \%}$, and reduces the maximum retracement by $1.11 \%$ compared to the traditional fixed strategy. The experimental results show that the portfolio hedged trading strategy based on deep reinforcement learning algorithm can reduce the overall investment risk, ensure the stable growth of cumulative return, and determine a relatively good adjustment cycle.