The Structural Break Test Of Realized Volatility

Since the establishment of China's stock market,"great bull market" and "great bear market" often occur alternately.Timely grasping the trend of market fluctuation can help investors to arbitrage from it,which makes volatility always a hot topic in the field of financial research.Historical volatility is the initial exploration of volatility by people.Its disadvantage is that it only uses the volatility information of daytime financial asset returns,and ignores the impact of intraday financial asset returns on volatility.With the development of computer technology,the generation of high-frequency trading and the emergence of high-frequency data,realized volatility(RV)has become a new direction of volatility research in recent years.However,realized volatility has the properties of long-term memory and structural break,which make traditional volatility modeling and estimation methods invalid.Therefore,this paper proposes a new statistics to test the structural break points of long memory time series,and derives the statistical properties of the statistics.These statistical properties can help us to monitor the position of volatility break points in real time.More importantly,people can model and predict volatility more accurately based on the data after the break points.These results have important practical significance for investors in Chinese stock market.Based on the statistics for structural break detection of short memory time series proposed by Lütkpohl(1988),this paper proposes t statistics for structural break detection of long memory time series.Monte Carlo simulation is used to detect break points,and the detection results are compared with the detection results of classical Cumulative Sum(CUSUM)statistics.It is concluded that when the parameter of long memory is less than or equal to 0.2,the detection effect of this paper's statistics is slightly inferior to that of traditional CUSUM statistics,but the difference is small.However,when the long memory parameter is greater than 0.2,the statistics in this paper can more accurately identify the structural break points.In the empirical analysis,the logarithmic realized volatility of Shenzhen component index and Shanghai composite index from October 2015 to September 2018 are detected with the quality of long memory,and on this basis,the algorithm proposed in this paper is used to detect the structural break points.The test results show that the test results based on recursive prediction are more stable.Both Shenzhen component index and Shanghai composite index are detected with 8 break points,and those 8 break time points have a good coincidence with external events in time.Six of them are completely consistent on the detection results of Shenzhen component index and Shanghai composite index.The detection results of the other two break points all show that the Shanghai composite index responds to the volatility changes earlier than the Shenzhen component index.