Volatility Forecasting Of Financial Assets Based On Long Memory And Structural Breaks

Modeling and forecasting of financial volatility have always been hot topics in academic research.And they also play crucial roles in financial markets.One of the most important features of the financial volatility sequence is that the auto-covariance coeffcient is slowly decayed.The main idea of this feature in the academic field is to interpret the data generation process as long memory or structural breaks.In term of different data generation structures,the performance of different forecasting strategies are different.In this thesis,we propose distinct forecasting methods respectively following two situations where either long memory or structural breaks exists and the case where long memory and structural breaks coexist.For the former case,we use a strategy based on discriminant test results to select the corresponding forecasting method.For the latter one,this paper proposes a data-driven two-step forecasting strategy based on cross-validation.Firstly,we use exponential weighted moving average approach to build the model;secondly,we capture the possible long memory feature in the residuals sequence via long memory model.The finite sample performences of these forecasting procedures are demonstrated to be better than the classical methods by a large number of simulations.Finally,we analyze the sequence of the log of weighted realized bipower variation of intraday high-frequency data of the Index of 300 stocks in Shanghai and Shenzhen market The real data example illustrates further that the two-step forecasting procedure proposed in this paper performs better than the traditional method when the long memory and the structural breaks coexist.Considering the simulation and the empirical conclusion,we find that:1)In the situation where either long memory or structural breaks exists,the work of discriminating the data structure based on the discriminant test improve the prediction.2)In the situation where long memory and structural breaks coexist,the two-step forecasting based on cross-validation is significantly better than the traditional forecasting.