Forecast Combinations Under Long Memory Structural Breaks

The continuous development of financial markets has brought great convenience to enterprises and individuals in managing funds and allocating assets,however,the impact of increased volatility in global financial markets brought about by financial globalization can't be underestimated.Financial time series mostly have the characteristics of long memory and structural break.A large number of studies show that in the prediction,ignoring the existence of break points,the prediction results will be imprecise.Based on chen Chen(2018),this paper makes a forecast combination of the time series with structural break and long memory properties,based on the detection statistics of long memory time series break points.The results of the Monte Carlo simulation show that the prediction results of the two combination predictions is better than that of the AR model and ARFIMA model based on single window prediction.In addition,In addition,the forecast combination has more advantage when the level of the long memory parameters and breaks are high.Realized volatility contains intraday trading information,and the calculation is simple,and is widely used in the measurement finance field.Therefore,in the empirical part,this paper applies the proposed forecast combination method to the prediction of realized volatility.According to the study of the volatility realized from October 24,2016 to October 22,2019,the break point detection method used in this paper detects 14 break points of the Shanghai Composite Index and the Shenzhen Index,which can better monitor the break information of the Shanghai Composite Index and the Shenzhen index.On this basis,the forecast combination method proposed in this paper has better performance in both the Shanghai Composite Index and the Shenzhen index,and its prediction effect is better than that of the AR model and ARFIMA model based on single window width modeling.And using the realized volatility obtained from the combination forecast,the resulting Va R value can better predict the risk of financial markets.