Study On Accuracy Of VaR Models Based On ACF

more popular in both academic field and financial market.Under this background, this paper focuses on how to describe investment risk with VaR by evaluating the degree of accuracy is the point that this paper concentrates on. More specifically, four VaR models have been discussed: GARCH-normal distribution, GARCH-t distribution, EVT-POT and EVT based on definition.Firstly, about the content, this paper extends its work from the following two views. On one hand, this paper builds a relationship between fat-tailedness of artificial stock's return and accuracy of VaR models. On the other, this paper links the cluster of volatility of the artificial stock with accuracy of VaR models.Secondly, about the method, this paper originates its work from Agent-based Computational Finance (ACF). Specifically, this paper gets the artificial stock on a platform named Netlogo. Then it makes statistical analysis with Matlab.Finally, about the conclusion, this paper ends its work in the following results. For one thing, with the tail of return's distribution becoming thicker and thicker, EVT-POT model and EVT model based on definition maintain their accuracy. For another, when the volatility runs in a region where the volatility is relatively low, GARCH-t distribution model is accurate while GARCH-normal distribution is inaccurate. When the volatility runs in a region where the volatility is relatively high, GARCH-t distribution model loses its accuracy while GARCH-normal distribution remains inaccurate. This means in such a cluster of high levels of volatility, it is not wise to estimate risk with GARCH-t distribution model. For the last thing, when a cluster of low levels of volatility is the condition, among EVT based on definition, EVT-POT and GARCH-t distribution models, the last one always tends to under-estimate risks while first one always tends to over-estimate risks.