| Deep reinforcement learning is a popular research field in recent years.It combines the advantages of deep learning algorithm and reinforcement learning model,and extracts features from the environment through deep learning,to enhance the decision-making ability of reinforcement learning.On the one hand,the trading process of financial markets can be approximately regarded as a Markov decision-making process.The complete trading process can be modeled by deep reinforcement learning,which can reduce unnecessary losses caused by subjective factors compared with artificial trading.On the other hand,all kinds of information in the financial market can enable deep learning algorithms to extract characteristic information to a certain extent,thus improving the judgment level of the reinforcement learning model.However,the frequency of such information is not uniform,so it is difficult to deal with it effectively and uniformly.Therefore,based on the above discussion,this paper studies cotton futures as an example,uses the deep learning algorithm of fusion frequency data processing to extract the characteristic information between data of different frequencies,and inputs it into the reinforcement learning model for learning,to obtain a complete set of trading strategy and form a visualization system.The main work of this paper has the following aspects:The GM-LSTM model is constructed by using the Long Short-Term Memory(long short-term Memory,LSTM)deep learning algorithm and the mixed frequency data fusion model(Generalized Auto Regressive Conditional Heteroskedasticity-Mixed-Data Sampling,GARCH-MIDAS)to process the main factor data affecting the cotton futures price obtained from the fundamental analysis method.Experiments show that the results of mixing frequency data fusion after adding the fundamental data of cotton futures are better,which can better reflect the regular characteristics of the price fluctuations of cotton futures.Input the fused conditional volatility into the deep learning algorithm LSTM.The results also show that the fitting effect of the model with conditional volatility is still good in the face of sharp price fluctuations.The Deep Q-network(DQN)in the deep reinforcement learning algorithm is used to design appropriate trading strategies to design appropriate trading strategies to reduce the losses caused by human factors,and the GM-LSTM model was integrated into the DQN-GM-LSTM model to improve the feature extraction ability of the reinforcement learning model.Experiments show that trading strategies constructed by DQN-GM-LSTM model can outperform benchmark strategies and perform better in bear market and volatile market conditions.Flask framework is adopted to develop cotton futures strategy system to realize data integration and visualization,to facilitate user operation. |
