Research On Commodity Futures Trading Strategy Based On Wavelet Decomposition And Random Forest Algorithm

With the deepening development and reform of China's financial market,more and more financial innovations and investment methods are applied to the market.In the face of complicated financial market,the rapid development of computers brings new opportunities for financial investment.In this paper,it is expected that the stochastic forest model in machine learning and the wavelet decomposition theory with very good noise reduction effect can be used to create a profitable trading strategy in the futures market.This article through to the futures market is active only 31 futures varieties of 23 by wavelet noise reduction processing data,and then put them as the classification of the stochastic model attribute,and then to the random forest model parameters by setting the training set and cross validation set tuning,interval setting in the 2016-01-01-2016-01-01 tested four rolling subinterval,AUC values and correct selection as the basis,will be elected to the appropriate parameters?Based on the AUC value and accuracy,the parameters of the random forest model were selected as follows: the number of trees l=300,the optimal number of attributes s=5,the minimum number of samples n=20,and the minimum number of samples m=20 for the repartition of internal nodes.as a final parameter model to predict the futures price rises or not,on the basis of these preparations to build a one-way trading strategies,The unprocessed data were set as the unified control group.The final trading strategy based on wavelet decomposition and random forest model achieved an improvement of 5% to 15% over the control group.The average positive accuracy of the cross validation set of the optimized random forest model was 58.8%,and the AUC was 0.627.The average out-of-sample accuracy was 58.0% and the average AUC was 0.611.In the strategy back test stage,the average accuracy of the random forest model futures trading strategy is over 95% in sample,over 50% out of sample,20% in sample and 9% out of sample.The average sharpe ratio was 73.72% in the sample and 1.68% out of the sample.It can be seen that the combination of random forest model and wavelet decomposition in machine learning to process data and predict the future can indeed provide strong support for the construction of futures trading strategies,and provide some evidence for the application of machine learning in trading strategies.