Research On Structural Breaks Of HAR-RV Model Based On Adaptive Group Lasso

In the financial market,the volatility of financial asset prices is central to asset allocation,derivative pricing and risk management.To a great extent,the volatility of a country's stock market can often reflect the fluctuation of the country's financial market.Therefore,more and more scholars are conducting research on the volatility of the stock market.Financial time series are subject to structural breaks because of various factors such as financial crisis,major changes in market sentiments.For the realized volatility series with structural breaks,if only one model is used to estimate the entire volatility series,there will be serious model setting errors,and the true volatility of the Chinese stock market will not be well grasped,and scholars may come to wrong conclusions,which have a worse impact on the decision-making of investors and management departments.In this case,we should first accurately estimate the number and location of structural breaks,and then the local model can be used to explain the existence of structural breaks.Then we divide the time into several adjacent time series segments according to the location of structural breaks,and fit a model in each time series segment,then each local model has the same structure but different parameters.Therefore,whether the number and location of structural breaks is correctly detected will play an important role in model fitting and estimation.The paper first proposes a HAR-RV model with structural breaks to study the volatility of the Chinese stock market.Then,this paper rewrites the model into a highdimensional linear regression model,and proposes to use the adaptive group Lasso to detect and estimate structural breaks.Compared with other Lasso methods,adaptive group Lasso method has the three properties of parameter estimation consistency,model selection consistency,and asymptotic normality.In addition,unlike traditional changepoint detection methods that first detect the number and location of structural breaks and then estimate structural breaks,adaptive group Lasso method can obtain the number,location and estimation of structural breaks at the same time.This paper conducts empirical research on the high-frequency data of Shanghai composite index,adopts the method proposed in this paper,the traditional CUSUM and MOSUM structural breaks test methods to detect and estimate the change points of the HAR-RV model respectively,and fit HAR-RV model based on the estimated structural breaks.The empirical results show that the structural breaks detection accuracy obtained by the adaptive group Lasso used in this paper is better than the structural breaks detection accuracy obtained by the traditional structural breaks detection method.The in-sample forecasting results of the HAR-RV model show that the segmented HAR-RV model based on the structural breaks estimation result of the adaptive group Lasso has the best in-sample forecasting effect,which is better than the segmented HAR based on the CUSUM and MOSUM structural breaks estimation results-RV model,and HAR-RV model that does not consider structural breaks.Next,this paper compares the effects of single-model forecast and forecast combination based on HAR-RV model under the premise of considering the structural breaks information.The forecast combination uses four kinds of weights,namely the weight based on the reverse CUSUM statistics(ROC)(including ROCW1 and ROCW2),the equal weight(EW),and the location weight(LW).The empirical results show that the out-of-sample forecasting effect of the forecast combination is better than that of the single-model forecasting.Considering the structural breaks information can further improve the effect of the forecast combination.Among them,the combined forecasting effect based on the structural breaks estimation of adaptive group Lasso is the best.