Volatility Analysis And Prediction Of Financial Time Series Based On GARCH Models And Machine Learning Algorithm

Volatility,as an important attribute of financial derivatives,can be effectively applied to the pricing of financial derivatives,allocation of financial assets and risk management,so the analysis of volatility has always been one of the hot issues in financial research.In this paper,we mainly study the direction of volatility prediction.Considering the advantages of Extreme Gradient Boosting machine learning algorithm,we combined it with GARCH class model to build GARCH-XGBOOST class model.At the same time,three kinds of stock index data are selected for comparative analysis of the model to test the prediction effect of the model in the financial market.The main research contents of this paper include:1.In terms of theoretical analysis,this paper mainly studies low-frequency time series.On the one hand,the machine learning algorithm and GARCH class model are combined to analyze time series data,and the GARCH class model and XGBOOST algorithm axe innovatively combined for stage prediction,and the GARCH-XGBOOST class model is established.At the same time,XGBOOST adopts grid search method to optimize the hyperparameters.The reason for choosing the XGBOOST algorithm is that XGBOOST algorithm uses the CART regression tree as the basic learner and uses a more accurate approximation algorithm,which can improve the model speed and prediction accuracy.The threshold value is introduced in the segmentation of regression tree,which limits the complexity of the tree.Meanwhile,the loss function can be customized to normalize the model structure and effectively prevent over-fitting.On the other hand,considering that the residual of the GARCH model is subject to the student t distribution and the generalized error distribution(GED),the GARCH-GED-XGBOOST model and GARCH-t-XGBOOST model are respectively combined with the XGBOOST algorithm.Thus,the characteristics of high peak and thick tail of data can be better described.In addition,in order to describe the data asymmetry,the GJR model combined with the XGBOOST algorithm is considered to propose the GJR-XGBOOST class model.2.In terms of empirical analysis,this paper selects three stock index data of CSI 300 Index,SSE Composite Index and SSE SME Composite,and uses three loss functions of MSE,MAE and QLIKE to test the prediction accuracy of the model.The prediction accuracy of GARCH-XGBOOST class model,GARCH-SVR class model,G JR-SVR class model and GJR-XGBOOST class model were compared and analyzed.The experimental results show that GARCH-XGBOOST model has better prediction effect than GARCHSVR model.In most cases,GJR-t-XGBOOST model and GJR-GED-XGBOOST model have greater generalization ability for data and improve the accuracy of volatility prediction.This paper selects three stock index data for empirical analysis to expand the practical effect of the model.In addition,the square of the deviation between the return of stock index and its mean value is selected as the volatility measure to test the accuracy of the staged prediction method,so that the method has extensibility.