Research On Stock Index Futures Volatility And Tail Risk

Under the shadow of abnormal fluctuations in 2015,investigating volatility and risk management of Chinese stock index futures is particularly important.Modeling and forecasting volatility is an effective way to investigate market volatility dynamics,and is also the basis of risk management.In recent years,the realized volatility research has made new progresses,it can be decomposed by two forms: continuous-jump variances,good-bad volatilities;Meanwhile,under the big data environment,online search volume has become a new entrance of financial research,and the technology breakthrough of deep learning enrichs our research methods.This paper takes the high-frequency volatility as the main line,using jumps,good-bad volatility and investor attention as the core variables,to investigate the volatility and tail-risk measurement of Chinese stock index futures.On the basis of the latest works of high-frequency volatility,using methods of quantitative modeling,data mining and deep learning,we firstly study the jump characteristics of stock index futures from the bull and bear markets.And,a series of HAR extended models and deep learning LSTM models are proposed to forecast realized volatility of stock index futures.Finally,the tail-risk measurement methodology are constructed under the RV-EVT framework.The main contents and innovative achievements are concluded as follows:(1)An improved jump detection method is proposed,the jump characteristics of Chinese stock index futures in bull and bear markets are explored.Considering the influence of overnight return,the more robust jump test statistic Z(lcrv-med)is constructed based on linear combination realized volatility(LCRV)and ADS jump test to separate the daily jumps and intraday jumps from realized volatility.The empirical analysis reveals that jumps in bull and bear markets are asymmetric,and daily jumps have intra-week effects while intraday jumps exhibt intraday effects.The intraday effect ACH model and the intraweek effect ACH model are established,which can effectively characterize the persistence of the jump durations and help to track the jump risk of stock index futures.This study provides a theoretical tool for the identifing and tracking of abnormal fluctuation of asset prices.(2)A series of HAR(Q)-type models are proposed to study volatility dynamics of stock index futures.Realized volatility,good-bad volatility,signed jumps are revised based on realized kernel estimator,and the modified statistic Z(rk-med)of ADS jump test is constructed to separate the jumps from realized volatility.Based on jumps,good-bad volatility and signed jumps,a series of extened HAR-type models are constructed.Considering the noise correction using Med RQ,HARQ-type models and HARQF-type models are proposed.Main conclusions of empirical study: Good volatility and bad volatility has asymmetric volatility impact that good(bad)volatility weaken(exacerbate)future realized volatility;Jumps and signed jumps has negative impact on future realized volatility;The decomposition of good-bad volatility outperforms that of continous-jump volatility in the framework of HAR-type modeling,and the Med RQ can improve the forecasting accuracy of HAR-type models.This study provides more robust estimators of high-frequency volatility,reveals the impact pattern of newly volatility estimators,which is of great value for investigation of quantitative trading strategy.(3)The novel HAR-type and LSTM-type models are proposed to explore the volatility forecasting value of online search volume.Taking the Baidu index,together with jump,good-bad volatility and signed jumps as the core variables,the novel HAR-type models and the LSTM-type models are established.The empirical analysis shows that the Baidu index significantly improves the prediction ability of the HAR-type models,and the rolling prediction evaluated by MCS test shows that the HAR-RV-SJ-BI model considering signed jumps is the best.The out-of-sample forecasting shows that the Baidu index can significantly improve the accuracy of the LSTM models,and the LSTM-RV-RS-BI model with good-bad volatility achieves the best performance.This study confirms the volatility predictive value of online search volume,provides an open framework of volatility prediction based on deep learning LSTM network,and fills in the blank of artificial inteligent research in volatility forecasting.(4)The tail-risk measure of RV-EVT framework is constructed based on high-frequency volatility and conditional extreme value theory(C-EVT).Embedding the HAR-type models in C-EVT model,the theoretical framework of RV-EVT is established to compute the Va R and ES metrics for tail-risk measure.Considering the jump,good-bad volatility and signed jumps,logarithmic HAR-type models are established,and its' prediction value is converted into conditional return volatility based on the connection function.Backtesting shows that evaluating Va R and ES metrics is feasible and effective under the framework of RV-EVT,and ES is more closer to the theoretical value than Va R,and evaluating Va R and ES based on HAR-RV-RS and HAR-RV-SJd models are more better.This study establishes a bridge between high-frequency volatility and risk management,and provides an effective scheme for the tail-risk measurement of stock index futures,which is of great significance to asset allocation and risk control.