Binary MIDAS-GAS Factor Volatility Model Based On Investor Sentiment And Its Application

Volatility has always been an important factor in measuring financial markets.High frequency intraday data reflects the advantages of measuring and predicting variance and covariance over daily data,and volatility and return are interrelated.This raises a natural question: how to use high frequency data to improve the structure and prediction of daily returns and volatility of financial assets? At the same time,behavioral finance believes that the decision-making behavior of investors will cause asset prices to be inconsistent with their intrinsic value,leading to low market efficiency.So far,a large number of documents have found that many characteristics of the stock market(such as price,yield and volatility)will be affected by investor sentiment(IS).In an environment of big data,people's demand for using different frequencies of data is increasing,and they are no longer limited to use the same frequency of data.In this case,a mixed data sampling(MIDAS)model is proposed.This model not only pays attention to low frequency data,but also pays more attention to the information in high-frequency data.Gorgi,Koopman and Li(2019)proposed a mixed data sampling generalized autoregressive scoring(MIDAS-GAS)factor model rely on GAS model.The dynamic model uses the real time information of high frequency variables to provide a flexible and easy-to-implement framework for predicting low frequency time series variables.Empirical research proves that the MIDASGAS factor model is superior to the competitive model.In summary,this article uses the realized volatility(RV)metric,combines the investor confidence index,sets up a score-driven volatility update project,and establishes the MIDASGAS factor volatility model to make full use of the high yield large amounts of data.Frequency data.Intraday information greatly improves the accuracy of model volatility forecasts.Taking the 5-minute high frequency data of the HS300 constituent stocks from July 13,2017 to August28,2020 as empirical data,analyze whether the MIDAS-GAS factor model with realized volatility and sentiment index has better prediction accuracy.The main content of this article includes five chapters:Chapter One briefly describes the background and significance of this study and its major content structure.The second chapter gives a more detailed introduction to the GAS,MIDAS and MIDASGAS factor models used in this article.Chapter Three stablishes a multivariate MIDAS-GAS factor model.Briefly describe the weighted maximum likelihood estimation method and proves its large sample nature.Finally,under the condition that the errors obey normal and t distribution respectively,the univariate and bivariate MIDAS-GAS factor model are subjected to Monte Carlo simulation to compare the prediction accuracy of the conditional variance of different models.Chapter Four,empirical analysis.Based on the HS300,this paper expounds the construction process of the realized volatility measurement and investor sentiment index,and discusses the impact of two factors on the volatility of financial assets.The SPA test is used to compare the prediction performance of the three volatility models: MIDAS-GAS-RV,MIDASGAS-IS and MIDAS-GAS-RV-IS.Chapter Five,summarizes the conclusions of this paper and prospect the future research and development trend.