A Research On Stock Trend-following Strategy Based On Reinforcement Learning

The trend-following strategy has been used in financial markets for a long time,but it still has the shortcomings of low win rate and large drawdown.Many researchers have also improved this strategy from different directions,but there are few studies on its improvement using intelligent algorithms.Therefore,this paper introduces more successful reinforcement learning algorithms in recent years to increase the win rate,reduce strategy drawdown of the trend-following strategy,and thus improve the overall performance.In this paper,we adopt the double moving average strategy as the input sample,train the reinforcement learning algorithm with the net value as the target,and at the same time evaluate the general trend tracking strategy,the improved trend tracking strategy and the strategy based on reinforcement learning proposed in this article by the expected return as the indicator.In view of some practical problems in the application of the algorithm in the field of investment strategy,this paper adopts methods such as discount reward normalization,discount reward reset and introduction of penalty items,which improves the training process and training results.Finally,in this paper,by limiting the amount of positions,the common,improved trend tracking strategy and the trend tracking strategy based on reinforcement learning were backtested in 110 market value and industry weighted sampling samples.The results show that the net value and expected return of the strategy proposed in this paper are better than the common trend-following strategy,which shows that the strategy has a certain practical application value.