Risk Measurement Analysis On China Stock Market

The riskmetrics method-VaR is certainly to be the most efficient leading metric tool in financial field in 21 century as is indicated by many overseas researchers. So the introduction of VaR is of great significance. Finding risk metric methods fitting financial market of our country would help risk administrators to know the characteristics of risk in China Securities Markets, analyze reasons for market risk and enact corresponding supervision regulations. Under this background. this dissertation firstly measures the market risk in China stock market with different value-at-risk computation methods, then runs a test on the empirical analysis results and finally compares the applicability of each method. This dissertation contains the following parts:The first part states where the topic derives from, its significance and innovations. The meaning of research in this article is mainly embodied in the following aspects:Firstly, it explicitly introduces 3 major kinds of VaR methods and its corresponding models to help others to get a better understanding and thus engage in risk management research; secondly, by finding risk metric methods fitting financial market of our country through empirical analysis, this article would provide risk administrators and investors with criteria on how to analyze, hedge and control risks; Secondly, we discuss the applicability of each method on risk metrics in China Securities Market and propose reliable risk metrics methods or models. Finally, by comparing and analyzing the results, we find out the advantages and disadvantages of each method; The methods mentioned above could further provide references to banks, futures,options and enterprises about risk management, and prevent economic losses caused by various market risks.The second part focuses on the current researches by analyzing several problems in this field:Firstly, from the methodology perspective, the VaR value of different methods varies and the conclusions sometimes can be quite different, which make their applicability questionable. Recently. with the development of extreme value theory and quantile regression theory.people begin to introduce semi-parameter method. However this method needs to be further explored and perfected. Secondly, in the research of risk metrics, the fat-tail and non-symmetric characteristics of yield series have always been the key factors affecting the result of VaR. In addition, extreme situations happen frequently and its risks always come with great losses. Despite the drawbacks mentioned above, however, parameter-methods haven't take those problems into account when computing VaR. Although, many non-parameter methods and parameter methods can solve this problem, they require a large number of history data or complex computations. Finally. compared with overseas continuously developing VaR theory, our research in domestic are relatively lag behind. The domestic applications are mainly on parameter methods, especially GARCH family models, and the researches and applications on non-parameter methods and semi-parameter methods are relatively less, let alone quantile regression theory.The third part states the evolving process of risk metrics and introduces metric methods of VaR, including theories of parameter, non-parameter and semi-parameter methods. The parameter method is mostly about the applications on risk metrics of GARCH family models. This article computes VaR separately under normal distribution, student distribution and GED distribution assumptions. Meanwhile, this part shows two methods for VaR test evaluation that is the failure frequency test method and quadratic loss function method. Non-parameter method includes historical simulation method and Monte Carlo simulation method. Semi-parameter method embodies in risk metrics based on extreme theory and quantile theory, which is also the key point in this dissertation.The fourth part is the empirical research. This article selects daily log returns of SSEC data to run empirical analysis using different methods, runs test on the results and compares their difference. By processing the sample data, we find out that SSEC log daily yield series are fat-detailed and non-symmetric. In the parameter method part, this article chooses GARCH family models to dynamically metric VaR, respectively under normal distribution, student distribution and GED distribution assumptions. The results of parameter estimation show that: under different distribution, each model can describe the volatility of SSEC yield very well. Compared with normal distribution and student distribution. GED distribution can better represent the fat-detailed feature of yield series. As for GARCH family models. TGARCH models'overall dynamical VaR metric results is of the best.In the empirical analysis of historical simulation methods:this article adopts general historical simulation method. weighted historical simulation method and filtered historical simulation method. To enhance the accuracy, this article cuts traditional simulation region into several sections. The results show that, by adopting different history regions, the results of VaR metrics vary considerably. The possible reasons might due to the implied assumption that history tend to reproduce. That's to say, big region means more volatilities. Researches find out that in the three methods mentioned above, general historical simulation method tend to underestimate, filtered historical simulation method overestimate and only weighted historical simulation method is relatively reasonable. This article only adopts the historical simulation method, the author suggests that we need to improve the existing historical simulation method or adopt Monte Carlo simulation method.In the semi-parameter parts, this article mostly adopts extreme theory and quantile theory based risk metrics method. The extreme theory adopts a popular POT model. This method, which gets the marginal distribution of returns series through tail estimation, is a non-conditional method so it can metric VaR value and ES value quite well because that residual distribution is usually non-symmetric skewed distribution we modeling the tail of this distribution. This dissertation gets a good simulation by adopting GPD and extreme theory is relatively mature currently.As for choosing the threshold value, this article first fixes on the range of it and then figures out the optimum threshold value by circulating sentences under the criteria generated by MADE series. Meanwhile, this article filters returns series with ARMA-GARCH model and constructs POT model on the basis of generated standard deviation series and metric dynamic VaR and ES in turn.In the quantile regression part, this article first constructs a new framework which computes VaR on the basis of conditional quantile regression model. This method directly models and analyses conditional quantile at any levels. Its advantage is that it doesn't require s specific distribution assumption and distribution parameters, which makes it applicable to fat-tailed data. This article constructs conditional quantile regression model which based on 5th order lag yield explanatory variable and empirically analyses SSEC log daily yield by non-parameter dynamic smoothing estimate method.We can analyze the VaR evolving mechanism of portfolio by efficiently using CAViaR analysis framework, which is the unique characteristic of this method. By exploring the potential VaR evolving mechanism, risk management level and efficiency could be remarkably enhanced.Based on the data from Shanghai Stock Exchange and Shenzhen Stock Exchange since raising limit mechanism was introduced. CAViaR analysis framework is constructed. It contains 4 kinds of models:SAV, AS. IG and Adaptive and stress emphasis on research of non-symmetric slope model. The empirical analysis results show that AS model is the best one to Shanghai stock index and Shenzhen stock index because this model best represents the market quantile movement mechanism, in other words, market risk evolving mechanism. The analysis on hypothesis test shows that AS model is the most reasonable and its quantile objective function value is least.The concluding-remarks part proposes the shortcoming and prospect of this research. Facing the increasingly complex financial system, reinforcing the market risk metrics is more and more important. So. further explorations are needed. According to research results of this article, each of the existing methods has their own advantage, however. neither of them alone could conduce to a satisfactory result.