변동성 모형 적합성과 표본외 VaR의 예측성과

This paper examines interrelationship between the fitness of volatility model and out-of-sample VaR forecasting performance. To find out the optimal volatility model that can explain in sample data property, the asymmetry between positive and negative returns on conditional variance of KOSPI200 index is investigated and the characteristics are incorporated into ARCH family model. And two-stage model selection procedure is employed for measuring the out-of-sample VaR forecasting performance. The first stage is the test of statistical accuracy and the second stage is filtering of the surviving models with using loss function. Major research results are as follows : The direct connections from the fitness of in sample volatility model to out-of-sample VaR forecasting performance are not taken effect accurately. But results show that the asymmetric GARCH models increased the suitability by reflecting the characteristics of the sample data and considering the impact of the asymmetric response on conditional variance increase VaR forecasting performance rather than GARCH model simply to describe the volatility clustering phenomenon, In addition, for the better prediction performance of 99% VaR estimation to predicts the extreme values more than 95% VaR estimates, the distribution of the error term is to be extended into the GED and Student’s t distribution and the suitability of the model is increased to explain the reality of the return distribution showing a thick tail.