Research On Investment Strategy Of Stock Index Futures Based On XGBoost

With the rapid development of artificial intelligence and machine learning algorithms,quantitative investment has gradually grown up in the financial field and become the focus of attention in the capital market.On the other hand,stock index futures,as an indispensable part of the financial market,plays an important role in the transfer and redistribution of stock market risks.However,the development time of Chinese stock index futures market is relatively late,and relevant literature on quantitative timing research on it is limited.At the same time,in the selection of feature vector indexes,most scholars take basic market indexes and technical indexes as input vectors,and existing literature finds that ETF return indexes have the function of price discovery for stock index futures.On this basis,the effect of ETF return indexes as prediction characteristics is studied and analyzed.This paper takes Shanghai 50 stock index futures as the research object,uses the XGBoost integrated algorithm which can better deal with nonlinear and non-stationary complex time series data,and takes June 16,2015 to December 31,2022 as the research interval,which is divided into training set,test set and back test set.Through EMD decomposition,data standardization and PCA dimension reduction,the initial sample data was transformed into input vectors for model learning.In this paper,the problem of predicting the price of stock index futures is transformed into a classification task of predicting the direction of price rise and fall,and the performance of the classifier is studied and analyzed under the condition that the input vector matrix is the basic market characteristics,technical index characteristics and ETF return index respectively.On this basis,the quantitative timing strategy is constructed respectively.Through the empirical study,the following three conclusions can be obtained:(1)when the input feature is ETF return index,the classifier constructed by XGBoost algorithm has a certain forecasting ability,although the accuracy is lower than that of the input vector based market indicators and technical indicators,but the accuracy is 50% higher than that of random classifier;(2)Compared with the basic market indicators,the technical indicators show better predictive ability.The reason for this difference is that the technical indicators contain more information;(3)For complex time series data,the use of EMD decomposition can better capture the hidden change rules of different frequencies and effectively enhance the prediction accuracy of the model.The quantitative investment strategy is constructed based on basic market indicators,technical indicators and ETF income indicators,and the backtest results without adding ETF characteristics are compared.It is found that both of the two strategies can obtain no less than 150% annual returns,and the maximum retracement is controlled within 3%.There is a limited difference in the excess returns obtained under the unit risk.Among them,the trading strategy constructed by combining all the characteristics performs slightly better in the annualized rate of return available.