Research On Realized Volatility Prediction Based On Text Big Data

In the financial market,the study of stock market volatility has important reference significance for investors and regulators.The study of volatility is actually the prediction study of financial time series.The traditional GARCH model is usually adopted for prediction,but the data used by this model are low-frequency data,which is difficult to capture more effective financial information.Currently,many scholars use HAR-RV model based on high-frequency data for comparison,and find that the prediction accuracy is improved compared with the GARCH model.With the gradual expansion of deep learning algorithm in the financial sector,more and more financial time series prediction problems have been proved to be able to use deep learning algorithm to better fit and predict.Since most of the objects predicted by the deep learning algorithm are stock prices.This paper attempts to apply the deep learning algorithm to the prediction research of stock volatility.Therefore,the neural network GRU model suitable time series is selected to conduct prediction research on the realized volatility of Chinese stock market.Previous studies have shown that news media has an impact on stock market volatility.Under this background,how to extract the impact of news on stock market volatility and apply it to the increasing prediction of volatility has become a new research direction.Based on text big data,this paper extracts news text information by constructing emotion dictionary and quantifying emotion indicator,then add it into the HAR-RV model and the GRU model based on high-frequency data to form its expansion model.In addition,the loss function method and the MCS test method as the evaluation system,this paper compares these models to determine whether the emerging deep learning algorithm in the stock market volatility has better prediction performance,and whether the performance of the model with news emotional indicator is better than that without.Finally,this paper takes CSI 300 index as the sample to conduct empiricalresearch.The research results show that the prediction results of HAR-RV model and GRU model based on high-frequency data are significantly higher than that of GARCH model based on low-frequency data.With high-frequency data,the GRU model has better prediction performance than the classical HAR-RV model,and adding emotional indicator of news media can indeed improve the prediction effect of the stock market volatility model.