The Research Of Realized Volatility Based On SSE 50ETF

In modern financial markets, volatility plays an important role in financial derivatives trading, pricing and risk management. In this paper, commonly-used realized volatility models are extensively evaluated and compared on the basis of high-frequency dataof the SSE 50ETF。The main contents could be summarized as follows:1. The construction, influencing factors and characteristics of realized volatility are firstly reviewed; andthe realized volatility jumps theory and corresponding identification methods, structural change theory and test methods are illustrated in detail. Furthermore, the HAR class model, a typical forecasting model of realized volatility, is derived.2.Statistical analysis of the SSE 50ETF Realized volatility sequence is carried out. It is found thatthe SSE 50ETF Realized volatility sequence is a stationary sequence, and has the features of the tail, long memory, and right after the peak bias. In addition,it’s a non-normal distribution in the standard situation, but its logarithmic sequence is quite close to normal distribution. Then the realized volatility is divided into continuous part and jumping part by the establishment of the Statistics Z,. The statistical characteristics and the distribution of the continuous part are similar to those of the realized volatility, with unobvious long memory feature in the jumping part. Considering the structural break effect, the BP method is employed to detect that structural break points of the realizedvolatility sequence. Herein, the total sample range can be divided into several sub-intervalsto be studied separately.3. Based the features of the realized volatility of the SSE 50 ETF, the HAR class model, containingthe HAR-RV model, HAR-RV-J model and HAR-RV-CJ model, in the total sample range is establishedto fit out the changing regularity of SSE 50ETF. Results show that the HAR class model has higher degree-of-fitting onthelogarithmic form of the realized volatilitythan the realized volatility itself, and the goodness of fit of the HAR-RV-CJ model is the highest. Thereby the HAR-RV-CJ model is applied to fit thechanging regularity in each sub-interval. It is demonstrated that the HAR-RV-CJ model has the higher degree-of-fitting for the times series which is close to the normal distribution and has the long memory feature.