| In the context of the increasingly clear characteristics of sector rotation in the A-share market,industry rotation strategies are increasingly favored by investors.However,most of the current industry rotation strategies in the industry are based on manual analysis and empirical judgment.They are often affected by subjective factors,emotional fluctuations and other factors,making it difficult to achieve the optimal combination.Building strategies based on deep reinforcement learning(DRL)can eliminate subjective influences,extract rules that are difficult to mine with ordinary methods,and achieve adaptive strategy adjustment.The strategy in this paper is based on the DRL model,and considering the particularity of time series data,the strategy implementation uses the DRL variant model DRLSTM model to better learn the long-term dependencies of time series data.The strategy implementation is based on the Fin RL framework,and the action subspace and risk control module are customized in the transaction environment class.Since the LSTM network is not implemented in Fin RL,the LSTM model part is implemented through Keras customization.In the backtesting implementation of the strategy,the data set is first divided into a test set,a verification set,and a transaction set to test the trading performance of the model based on different data sets.Then the original data is subjected to feature engineering and factor selection.The factor test method adopts single-factor IC test and single-factor grouping test to test the correlation effect and stratification benefit of factors on the target rate of return.Then,the final selected eight factors are input into the DRLSTM model,and parameters are tuned based on the customized LSTM network parameters to obtain stable model backtest results.The backtest results show that under the condition that the benchmark risk return is13.87%,the strategy can achieve an annualized excess return of 28.157% and an annualized volatility rate of 20.312%,with an excellent risk return level.In order to further prove the applicability of the model in the backtest,three additional sets of control strategies based on the same data set backtest were added,namely buy and hold strategy,factor regression strategy,and random forest strategy.The results show that the strategy performs well in terms of profitability,risk resistance,and stability.The shortcoming is that the ability to resist extreme risks is insufficient.To sum up,the strategy in this paper has achieved good results in capturing industry characteristics and obtaining risk and return.The practical experience in this paper can also provide a certain reference for future industry rotation and other stock investment strategies. |
