The Research And Implementation Of Stock Index CTA Strategy Based On Machine Learning

Compared with the spot market,stock index futures have a good liquidity and flexible opening and closing system,which makes stock index futures play a role in reducing the systemic risk of portfolios in the asset allocation process.Since the SSE 50 stock index futures market has only been listed for 4 years,the historical transaction data is limited.At present,there are few studies on the financial futures market,and there is no reference to the mature stock index trading strategy.This article discusses some of the issues that can accurately predict the trend of stock index prices and get the spreads.Firstly,the ΔT time series sample sets with different time series are constructed as model inputs.Then,based on the optimized machine learning prediction model and the improved CTA trend strategy,a machine learning based CTA strategy model is constructed.Finally,this machine learning based CTA is built.The strategy model was validated.The main research contents and achievements of this paper include the following aspects:(1)A set of ΔT timing samples with different timing periods is constructed.Since the SSE 50 stock index futures only have 4 years of historical transaction data,in order to solve the shortage of the model input data,a ΔT time series sample set is constructed by changing the refined sample size of 1 minute,1 hour and 1 day.Using the strong correlation between stock index futures and stock index to expand the data breadth,the data set of 1 minute reached 460,000,effectively solving the problem of insufficient data of SSE 50 stock index futures,providing sufficient training for subsequent model training.Data foundation(2)Optimize the machine learning prediction model and improve the CTA trend strategy.The opening price,closing price,volume and other characteristics of the stock index index and the ΔT time series are characterized as N*T feature matrix as input of DNN,CNN and LSTM models,which optimizes the machine learning prediction model and improves the accuracy of the model;Some improvements have been made to the CTA trend strategy.The Bollinger Band Strategy and the Tang Qi'an Strategy with different parameters have improved the flexibility of the strategy and improved the short-term timeliness of the traditional CTA strategy.(3)Construct a CTA strategy model based on machine learning.Using the optimized machine learning prediction model and the improved CTA trend strategy,the LSTM-Donchian IH continuous contract 1 day strategy and the CNN-Donchian IH continuous contract 1 minute strategy are used to give the machine learning model prediction signals a certain weight and the Tang Qi'an strategy.Signal weighting,get new trading signals,and integrate into a new strategy model;(4)Verify the effectiveness of the machine learning based CTA strategy model.Through experiments,the accuracy of the two mixed models CNN-Donchian and LSTM-Donchian models is 86.75% and 75.49%,and the yields are 100.16% and 98.8%,which is significantly higher than other models,which proves that this paper is based on machines.The effectiveness of the CTA strategy model of learning.