Price Prediction And Quantitative Strategy Construction Of Stock Index Futures Based On KPCA-gcForest

When traditional prediction methods are difficult to solve the problems of large noise and nonlinearity in financial time series,Machine Learning(ML)has gradually been widely used in quantitative investment by virtue of its performance advantages in complex data classification and regression.At the same time,various feature processing methods can also help investors dig out more feature information in financial time series data,thereby reducing the difficulty of forecasting.With the development and improvement of China's futures market,stock index futures trading has become the focus of research in the field of quantitative investment.Researchers have gradually focused on exploring the application of multiple feature processing and machine learning methods in quantitative investment in stock index futures.This article uses the CSI 300 stock index futures as the research object.First,select the deep forest model(gcForest)at the forefront of machine learning,predict the price rise and fall of stock index futures based on historical market data,and compare them with the prediction results of the SVM and XGBoost models.Second,select technical indicator factors and historical market data to form 26-dimensional input features,and use Kernel Principal Component Analysis(KPCA)method to reduce and extract features to form a KPCA-gcForest combined model,which is used for prediction of changes of stock index futures prices.Finally,the combination model is selected to construct a quantitative timing strategy with CSI 300 stock index futures as the trading object,and the strategy is optimized based on the backtest results.The research results of this paper show that: First,the gcForest model performs well in comparison with the SVM and XGBoost model's predictive capabilities,with the highest accuracy,AUC and other indicators,which proves the effectiveness of the gcForest model in predicting the price rise and fall of stock index futures.Second,after KPCA feature dimensionality reduction,12-dimensional input features are extracted.After training the gcForest model,the KPCA-gcForest combined model is obtained,which has a better classification effect in predicting the price of stock index futures,and the accuracy is 62.18%,which exceeds the result of only using the gcForest model to predict.It proves that the KPCAgcForest combination model has better classification performance in stock index futures price prediction.Third,the CSI 300 stock index futures quantitative timing strategy based on the KPCA-gcForest combined model has achieved good risk-return performance in the backtest and has actual quantitative investment value,indicating that the model can be a quantitative investment in the financial market which can provide reference and help.