Evaluation of Monthly Portfolio Value-at-Risk Forecasting Models |
Jae Jin Park |
월별 포트폴리오 VaR 측정 성과 |
박재진 |
한국은행 |
|
Abstract |
This paper is aimed to find out the statistical model to properly measure the monthly portfolio Value-at-Risk. For the purpose of this, I utilized four kinds of models, such as the Historical Simulation method, EWMA model, GARCH (1, 1) model and Dynamic Conditional Correlation-GARCH model. The portfolio was composed of five assets including the US Dow Jones Industrial Index and four kinds of the US treasuries, such as treasuries with maturities of one year, three years, five years and ten years. All five assets had the same weight (1/5) in the portfolio. The portfolio VaRs were predicted by the above four models under the 99% and 95% confidence levels, and evaluated not only by the Christofferson (1998)’s statistical method but also by other various analyses related to VaR features such as the mean relative bias, the oot mean squared relative bias, absolute mean of tail event, and the predictability of the portfolio VaRs. The above evaluation revealed that the VaR measured using the DCC-GARCH model under the 99% confidence level could be adopted as the proper monthly VaR measure, but the other VaRs measured using other models could not be the proper one under the both confidence levels. |
Key Words:
시장리스크 측정,역사적 시뮬레이션,EWMA,GARCH,DCC,VaR Forecasting,Historical Simulation,Dynamic Conditional |
|