Study On The Impact Of Outliers On MS-GARCH Model Family Estimation And Forecasting

Oil financial security is a key concern for the country at present,and crude oil return volatility is one of the important topics for analyzing oil financial security.In the risk analysis of crude oil returns,outliers and structural transformation are two important factors that affect volatility modeling.At the same time,outliers also affect the estimation and forecasting of volatility structural transformation models,and it is of high theoretical and practical significance to incorporate outliers into the analytical framework of volatility structural transformation models.The implicit assumption that there are no outliers in the time series will be difficult to hold in the volatility analysis of asset return series as the impact of intervention events on the series has been increasing in recent years.The volatility series is very sensitive to intervention events,and the existence of outliers makes the set model unable to maximize the characteristics of the data generation process,and the study of outliers is of great significance to the analysis of volatility series.There are two main methods for outlier handling:outlier detection and correction;and robust estimation.The first method requires model-dependent to detect outliers,however,the presence of outliers can bias the parameter estimation of the model-dependent,which in turn affects the quality of outlier detection.The second outlier treatment method can obtain robust estimates of the model,but it cannot perform intervention event analysis and loses the opportunity to perform economic analysis of the outliers.Based on the application pain points of the two outlier processing methods,scholars introduced the idea of robust estimation into the outlier detection process to form a robust outlier detection technique.However,this technique also has application pain points:first,it can only detect one type of outliers,namely Additive Outliers(AO),and cannot continue to identify three types of outliers,namely Innovative Outliers(IO),Transient change(TC),and Level Shift(LS).Second,the models-dependent are currently limited to ARMA models,a family of symmetric GARCH models,and other volatility models that can be converted to symmetric GARCH models.Therefore,this robust outlier detection technique cannot be applied to the family of Markov regime switching GARCH(MS-GARCH)models that can characterize serial structural shifts.On the other hand,the presence of outliers affects the description of the structural switching features of sequences,but there are few articles that incorporate outliers into the structural switching analysis framework.In order to incorporate the outliers into the structural switching analysis framework and identify all types of outliers as much as possible,the paper innovatively proposes a"two-step"approach.In the first step,the series are pre-processed with both outlier processing methods to remove the influence of outliers on the modeling.The advantages of such treatment are two:first,the effects of the two outlier treatment methods in structural transformation modeling can be observed simultaneously in the same scenario to prepare for the generalized application of structural transformation modeling with outlier treatment.Secondly,four types of outliers can be identified.Specifically,the paper sets up a control group and two experimental groups.The control group model does not consider the effect of outliers and only performs structural transformation modeling.In Experimental Group Ⅰ,outliers are processed using the time series outlier detection technique to detect the intervention event and analyze the economic impact of the intervention event.After that,the outlier detection and corrected data are brought into the structural transformation modeling framework.In Experimental Group Ⅱ,the series are pre-processed using the BIP-ARMA(τ)model,and the data after processing are brought into the structural transformation modeling framework.With Experimental Group Ⅰ,the specific impact of intervention events in the economy and the improvement of outlier detection techniques on structural transformation modeling estimation and forecasting are examined.With Experimental Group Ⅱ,the role of robust estimation techniques on the improvement of structural transformation modeling estimation and forecasting is examined.The comparative analysis of the two experimental groups allows for the examination of the similarities and differences between the two outlier handling methods on improving the effectiveness of structural transformation modeling.If the characteristics of the two outlier handling methods in improving model estimation and prediction can be found,the outlier handling method that can improve the structural transformation model estimation and prediction to a greater extent can be selected for a specific market.It is found that the models built under the "two-step" approach have significantly improved estimation and forecasting.However,in the dimension of estimation effectiveness,the two outlier treatments differ in the extent to which they are useful in different markets:Experiment 1 outperforms Experiment 2 in Brent and Oman,and Experiment 2 outperforms Experiment 1 in WTI and SC.There is no clear criterion for which outlier treatment is specifically appropriate for a particular market.In the dimension of forecasting effectiveness,the two outlier treatment methods also perform differently in front of the six forecasting effectiveness evaluation indicators.On the three indicators SE1,QLIKE and AE1,the correlation model of Experimental Group Ⅰ performs better;on the indicator SE2,the correlation model of Experimental Group Ⅱhas better forecasting ability;on the indicator R2LOG,the family of MS-GARCH models with direct modeling is able to make better empirical forecasts;on the indicator AE2,the forecasting effect of the correlation models of Experimental Groups Ⅰ and Ⅱ The difference is not significant.Based on China’s oil financial security issue,we study the impact of outliers on the estimation and prediction of the MS-GARCH model family.GARCH model family;third,the outliers and structural transformation are incorporated into the analysis framework of China’s oil finance risk at the same time.In the empirical application,outlier treatment is very important for the robustness and rigor of model conclusions.In the selection of outlier processing methods,risk analysts can refer to the actual situation and select the estimation methods in a targeted manner according to the penalizing characteristics of the six predictive evaluation indicators.