The Multi-fractal Identification And Application Of High-frequency Realized Volatility In Chinese Financial Markets

The research of volatility in financial markets is always one of the essential researches in financial research.Being one of the most representative volatility,realized volatility is able to reflect the attitudes of investors.In the recent decades,with the great progress of computer science,financial markets also had improved a lot,which leads to the rapid increase in daily trading frequency of Chinese financial markets.Under this circumstance,the research of high-frequency realized volatility is going to be more and more valuale.Among the diversified research methods,phicical financial method had introduced multi-fractal into financial research.In this essay,multi-fractal analysis was firstly applied in the heterogeneity research of high-frequency volatility.Then,based on the selfsimilarity which is one of the multi-fractal features,the multi-scale multi-selfsimilarityinterval comparison method was introduced to describe and forecast the high-frequency realized volatility.Finally,based on the selfsimilarity,the new application of high-frequency realized volatility in financial markets was introduced for daily trading and risk control purposes.The main works of this essay can beconcludedas below.1.In the heterogeneity research of financial investors,the multi-fractal values were introduced to estimate the multi-dimentional vectors in order to identify the differences between investors based on the Euclidean distances between the vectors.In the traditional researches the investor features are depended on different environments.However,with the establishment of multi-level capital markets,diversified derivatives are now available to find the different features of investors even in the exactly same environments.In this research,the multi-fractal analysis of SSE50 index was firstly applied both in static and dynamic situations.In this step,it is found that the investors who invest SSE 50 index in different way such as,futures,stocks and ETF,represent very different features even in the exactly same time period.What's more,different from the traditional analysis which are mostly determining the nature,the multi-fractal analysis is able to provide values to show the results.Based on these multi-fracatl values,it is able to apply them in summarizing the investing features by the Euclidean distances between the vectors.After this,the false nearest points test was applied to show that the multi-fractal values are essential in this process.Finally,the empirical analysis was applied on typical underlying such as,individual stocks with highly turnover and barometer stock.The results illustrate that the investing features of individual stocks some times are similar to institutional investors while some times are similar to individual investors,it changed fast while its change level is strong as well.It demonstrated the complexity and variability of Chinese financial markets.2.In the research of description and forecasting of high-frequency realized volatility,the multi-scale multi-selfsimialrity comparison method was applied.Considering that the high-frequency realized volatitliy is the observation of investing features,the investors who affect these features should be identified first.Based on the statistics bulletin offered by Shanghai exchange,Shenzhen exchange and Chinese financial future exchange,these investors are identified as individual investors.The main trading strategy of individual investors is based on technical analyisis which is based on the assumption that the price trend will be periodically occurred,which provided the realize evidence of selfsimilarity of high-frequency realized volatility.Furthermore,the fact that the multi-fracatal feature implied the feature of selfsimilarity provided the theoridical evidence of selfsimilariry of high-frequency realized volatility.Based on these facts,multi-scale multi-selfsimialrity comparison method was established.This method firstly tests the selfsimialrity level under different time scales and then finds the suitableselfsimilarity spaces under each time scale.Based on these selfsimilarity spaces,applying the suitable forecasting method,the high-frequency realized volatility were predicted.The empirical analysis was based on SSE50 index futures and the comparison methods were selected as ARFIMA and LSTM.Compared with there two methods,the multi-scale multi-selfsimilar comparison method can not only provide accurate prediction but also provide strong explanation ability of data characteristics.Specially,in the prediction of future trend,the accuracy rate had reached70%.During the comparison,it is also found that the fixed models such as ARFIMA and LSTM is not able to maintain its accuracy in long term prediction,the reason is that the investing features are always changed quickly and strongly.Finally,the discussion of parameter selection of this method increased its explanation ability.For now,the high-frequency realized volatility was affected by all time scales,thus in the prodiction process,the short,medium and long term time scales should be considerd,and the short term affected more than the other two types of scale,therefore,more short time scales should be taken part in the prodictionprocesure.3.Based on the multi-fractal features of high-frequency realized volatility,it is able to determine the selfsimialr level under different time scales.In this essay,the selfsimilar vectors were estimated by capture the characteristics values of realized volatility.By calculating the Euclidean distances between vectors,the centrue vector under different time scales can be estimated and the selfsimilar level can be presented by the Euclidean distances of the vectors.It is proved by the empirical analysis that with the strongest selfsimilar level,the fluctuation during the same time scale will be similar.Therefore,based on the theselfsimilar space estimation,it is possible for investors to find the potential trading opportunities in the future by analyze the past selfsimilarity of high-frequency realized volatility.These results had been proved in different underlying such as,ETF,high turnover rate stocks and barometer stock.4.The research on high-frequency realized volatility can not only help in daily trading but also may be helpful in risk control processes.Nowadays,the trading frequency had already reached a very high level in Chinese financial markets.Thus the researches of daily return intervals of extreme events are very important.In this essay,based on the selfsimilarity feature of realized volatility,the stretchedexponential distribution was estimated to describe the return intervals of extreme fluctuation of high-frequency realized volatility.The parameters of this distribution can be calculated by the Hurst coefficient which can be estimated in the multi-fracatl identification process.Compared with traditional exponential distribution,the stretched exponential distribution can cover the tail part better.Furthermore,compared with kenerl density estimation,the stretched exponential distribution can provide a continuous function which is more convient in the future application.