Research On Stock Index Forecasting Based On Regularized Sparse Model And Xgboost Algorithms

Because the stock market is affected by a variety of uncertainties,the stock index sequence is usually highly non-stationary and non-linear.It is a challenging problem to accurately predict the stock index sequence.In the early years,researchers mostly used single financial time series models such as ARIMA model,GARCH model or single machine learning model such as neural network model and support vector machine model to predict the stock index series.Although to some extent,they can effectively grasp its changes,the prediction accuracy still needs to be improved.In recent years,some researchers have tried to expand the relevant characteristics or construct a combination model by using the advantages and disadvantages of each model in the research of stock index prediction in order to better grasp its changing trend.In order to improve the prediction accuracy of stock index series,this paper makes efforts to construct more effective features and combination models.Firstly,from the perspective of technical analysis,this paper collects and collates the historical trading data of Shanghai Stock Index and Shenzhen Stock Index.Secondly,based on the historical transaction data,the daily transaction data indicators and common technical indicators are constructed.With the help of the multi-resolution analysis ability of wavelet analysis,the three-layer maximum overlapping discrete wavelet transform is applied to the daily transaction data indicators with DB4 wavelet as the wavelet basis function to obtain the wavelet decomposition sequence.Then,taking the three kinds of variables of daily trading data index,technical index and wavelet decomposition sequence as candidate characteristic variables,and the closing price of the next day as the target variable,the time series data set needed in this paper is constructed and preprocessed.Finally,Lasso,SCAD and MCP regularized sparse models are combined with Xgboost algorithm respectively to predict the two exponents,and the optimal prediction model is determined according to the prediction performance of the model on the test set.Among them,the parameters of Xgboost algorithm are optimized by grid search method,and the optimal prediction model is Lasso-Xgboost model.The results of empirical analysis show that: Firstly,the variables selected by the three regularized sparse models include three kinds of variables: daily trading data indicators,common technical indicators and wavelet decomposition sequence,which shows that the three kinds of variables we constructed contain a lot of information related to stock index changes.Secondly,the first ten variables based on Lasso-Xgboost model include technical index and wavelet decomposition sequence,but not daily trading data index.It shows that the technical index and wavelet decomposition sequence obtained by extracting the information contained in the daily trading data index can more effectively reflect the changes of stock index sequence.Thirdly,the combined model Lasso-Xgboost has good prediction performance on Shanghai stock index and Shenzhen stock index,and its prediction accuracy is better than SCAD-Xgboost model,MCP-Xgboost model,BP neural network model,SVR model,Xgboost model,Lasso-BP model and Lasso-SVR model.