Research On Stock Market Volatility Prediction And Risk Measurement Based On AT-BNLSTM-GARCH Model

Volatility in the financial market is the core of asset pricing and risk measurement.Accurate prediction of volatility has always been one of the hot topics studied by the financial academia and the industry,and has great theoretical and practical significance.In recent years,artificial intelligence technology has developed rapidly.With the maturity of artificial intelligence theory,machine learning and deep learning models have been gradually applied to the financial field,especially widely used in financial time series forecasting.This article closely follows cutting-edge theories and technologies,and uses deep learning models combined with traditional measurement GARCH family models to build a new model to predict volatility.In this paper,the LSTM model is improved through Batch Normalization method,and then the Attention mechanism is introduced in it,in order to assign different weights to the input feature vectors at different times,thus constructing the AT-BNLSTM model.Then the realized volatility calculated from the high-frequency data and the Baidu index reflecting the macroeconomic terms are used as the input features of the model,and then the traditional GARCH family models are combined to construct seven different AT-BNLSTM-GARCH models.Use different GARCH model parameters at different time points to build dynamic AT-BNLSTM-DGARCH models.Various models adopt rolling forecast method.This article selects the daily volatility of the Shanghai 50 Index's return from January 5,2015 to September 30,2020 as the research object.The empirical results show that the prediction performance of the two deep learning models,LSTM model and AT-BNLSTM model constructed in this paper,is much better than that of the traditional GARCH family model.The six loss functions of the seven AT-BNLSTM-GARCH models on the test set are all smaller than the AT-BNLSTM model,and the p-values of the MCS test of the loss function are all greater than the AT-BNLSTM model,indicating that adding the parameters of the GARCH family model to the deep learning model can indeed improve the prediction performance of the deep learning model.The six loss functions of the dynamic AT-BNLSTM-DGARCH model on the test set are all smaller than other models,and the p-values of the MCS test of the loss function are all larger than those of other models,indicating that the dynamic AT-BNLSTM-DGARCH model has the best performance in predicting the volatility of the Shanghai Stock Exchange 50 Index.After that,the volatility predicted by the dynamic AT-BNLSTM-DGARCH model was used to calculate the Va R measurement risk through the quantile regression method and the failure rate test was performed.The results showed that the calculated Va R passed the Kupiec test at the 95% confidence level,indicating that the model accurately describes the value of risk.Therefore,the hybrid model constructed in this paper provides a new method for accurately predicting volatility and measuring risk.