Comparing And Ranking Parametric, Nonparametric And Semi-parametric VaR Models

In this era, opportunities are accompanied by risks and financing environment is more and more complicated, thus as a finance institute in Chinese capital market, how to control and manage its investment risk is the key point to survive in the market.Nowadays, the most popular tool to measure the risk of assets is Value-at-Risk, however, there are so many existing models and methods to estimate VaR and each one has its own advantages and disadvantages, therefore this article aims at finding out the best models for measuring the VaR in the context of Chinese stock market. The article selects the daily return data of stocks which had or have options from 2000 to 2007, applies parametric, nonparametric and semi-parametric models, including five kinds of GARCH models with Normal and Student-t distribution assumptions, three kinds of improved Historical Simulation methods, Monte Carlo method, Extreme Value Theory. Filtered Extreme Value Theory and Conditional Autoregressive Value-at-Risk, makes one-step ahead prediction. Finally, under three kinds of lengths of predicted periods and two kinds of critical levels, uses Kupiec test and Quantile Loss test to compare and rank those models and methods from two different aspects, for giving some references of which model or method is the best under some given conditions.With a series of empirical studies, the article comes to the conclusion that GARCH models and Conditional Autoregressive Value-at-Risk usually perform well, and they are also applicable when lack of data because of not being sensitive to the length of data, on the contrary. Historical Simulation methods and Monte Carlo method do not work well, except filtered Historical Simulation and filtered Extreme Value Theory which perform well under Quantile Loss test.