Research On Stock Index Futures Forecasting Based On PCA-GRU Model Of Adabelief Algorithm

The development of stock index futures is highly valued by the governments of all countries in the new period,which can not only make the two-way circulation but also realize the mutual promotion of domestic and international economy.As an important financial derivatives,stock index futures’ high yield characteristics are also favored by the majority of investors.In order to help the smooth development of government work and reduce the risk of investors suffering huge losses,it is particularly important to study the price change rule of stock index futures and reasonably predict its future price.Reasonable and effective prediction of stock index futures,for financial regulators,can help them make economic decisions,manage the system risk of the stock market,help enterprises make financing plans and finally promote the economic opening to the outside world;For investors,it can help them choose investment products more rationally so as to obtain higher expected returns.Therefore,it is of great significance to predict the price of stock index futures.Traditional single econometric models such as ARIMA and GARCH cannot capture the nonlinear characteristics of financial time series well.Because the stock index futures data has the characteristics of non-stationary and loud noise,which makes the prediction accuracy is not high even if the classical SVR model and BP algorithm are used to study it.So this thesis will focus on how to build a neural network model with higher accuracy to predict the financial time series.The specific approach is combining the wavelet analysis and principal component analysis with the gate recurrent unit neural network,we also add the emotional index factor of investors to the basic indicators and technical indicators.Finally the model that we has built is respectively used manual tuning parameter and the improved PSO algorithm,which aims to further improve the prediction performance of the model.This thesis makes an empirical study and uses three indicators to evaluate the predicted results of the model,which based on the daily price data of the main continuous contracts of China’s three major stock index futures.Firstly,the PCA-GRU model under Adam optimization algorithm is used to preliminarily investigate the CSI 300 futures in China,which wants to find a group of optimal parameters of PCA-GRU model under Adam optimizer.Then,we put the same group of optimal parameters into the PCA-GRU model under Adabelief optimization algorithm to continue training and testing the data of CSI 300 stock index futures The results shows that the PCA-GRU model based on Adabelief optimization algorithm achieves better prediction accuracy under the same dataset,the same model and the same parameters.Compared with the model,which based on Adam,the MAE of the model based on Adabelief decreases by 0.5%,the RMSE decreases by 0.5%,and the R2 improves by 1.1%.Secondly,I innovatively select four indicators which closely related to investor sentiment and uses principal component analysis to construct investor sentiment indicators.The we add these indicators into the model based on Adabelief.The comparison shows that under the same optimizer,the same model and the same parameters,MAE decreases by 0.4%,RMSE decreases by 0.3%and R2 increases by 0.8%after the addition of investor sentiment index.Finally,we explore a set of optimal parameters of PCA-GRU model under Adabelief.It is verified again that Adabelief algorithm is superior to Adam algorithm.In other words,under different optimizers,different parameters,the same model and the same dataset,the MAE of PCA-GRU model under Adabelief decreases 1.5%,RMSE decreases 1.4%,and R2 improves 3.5%.However,we also carried out corresponding research on SSE 50 and CSI 500 futures in order to avoid the contingency of the model results.Three kinds of stock index futures are used to forecast in order to analyze the application effect of the model,which bases on comparing the output data of the model with the actual data.In view of the high time cost of manual parameter tuning,particle swarm optimization algorithm and improved particle swarm optimization algorithm are used to realize the automatic optimization process of the model.The empirical results show that:(1)Adabelief algorithm has better convergence and generalization ability than Adam algorithm.(2)The prediction ability of the model is improved after introducing the investor sentiment index constructed by PC A dimensionality reduction method.(3)The GRU model optimized by the improved PSO algorithm has good robustness in terms of time and prediction accuracy.