A Study Of Risk Measurement Of Gold Futures Market Based On Extreme Value Theory

With the rapid development and increasing complexity of financial market, the vulnerability in financial system makes the outbreak of financial crisis more frequent and harmful. As a result, risk management of the financial system becomes much more important to researchers, investors and regulators. On January9,2008, gold futures contracts were formally listed on the Shanghai Futures Exchange, making a landmark development. The launch of gold futures, which have commodity attributes, monetary attributes and financial attributes, plays the crucial role of pathfinder.However, there are few risk magnanimity researches of the gold futures at home and abroad, and the model constructed is single, so four models are constructed in this paper. Firstly, the VaR model based on variance and covariance method and the GARCH-VaR model are constructed based on traditional theories. Secondly, in order to compensate the deficiency caused by the neglect to extreme events of traditional VaR model, it is recommended to introduce extreme value theory to solve the problem of underestimation of tail risk. We use kurtosis method to select threshold, make GPD distribution fitting to super-threshold samples, and build the VaR model based on extreme value theory. Furthermore, by combining the feature of large remaining part and volatility clustering, we establish GARCH-EVT-VaR model to improve the accuracy of risk magnanimity model. On the basis of theoretical research, we studied the Shanghai gold futures data from January9,2008to December31,2013, and use the method of failure rate of back testing and Kupiec likelihood ratio to compare and analyze the four kinds of models. The results show that: At higher level of confidence, extreme value theory can describe the distribution of tails well and improve the accuracy and effectiveness of the model. The estimated VaR of GARCH-EVT-VaR model are more accurate than the EVT-VaR model. And at lower level of confidence, the traditional VaR model is more effective for long position investors, while the VaR model based on extreme value theory is more accurate for short position investors.