Systematic VaR Model Based On Multi-Resolution Analysis And Extreme Value Theory

Due to dynamic evolution over time of financial asset price, the distribution of earnings from financial asset tends to have "fat tails". Based on the heterogeneity of micro market, MRA-EVT model is established in this paper. Using the multi-resolution analysis of wavelet analysis, the model is able to capture time-varying characteristics of asset price volatility, so that series of return rate are decomposed into different time domain orthogonal components. For each component series, a certain ARMA-GARCH model is built. In addition, by linear superposition of these models, coupled with extreme value theory (EVT), fat tail can be modeled, and Value at Risk (VaR) could be predicted. At last, the model proposed in this paper is applied to predict VaR of HS 300 index. What’s more, based on Kupiec’s POF-test, the prediction performances are compared between the proposed model and traditional ones, such as ARMA-GARCH model, unconditional EVT model, MRA model. The empirical results show that the MRA-EVT model does improve the VaR prediction performance, which is conducive to financial practitioners to have a more conscious understanding of index market, and further benefit the risk management such as stock index futures in financial markets.