KOSPI200 변동성 예측성과：주기적 모형추정과 다기간 예측

There have already been numerous empirical researches on the forecasting performance of various volatility forecast models. But the forecast evaluation scheme of the recent domestic studies on volatility forecast does not answer the real-life question on which model would produce the best forecasts of the future volatility. This paper performed the moving window estimation each day to calculate the out-of-sample forecast error. It also tried the Diebold-Mariano test to see whether a certain forecast model’s superior performance is statistically significant. We compared the KOSPI200 volatility forecasting performance of historical volatility, Riskmetrics EWMA, VKOSPI, GARCH and HAR-RV. Overall forecast performance of EGARCH model was best, but VKOSPI and EWMA were very close in forecast accuracy. It is not surprising that DM test on the equal forecasting accuracy of EGARCH, VKOSPI and EWMA was not rejected. Another special feature of this paper is the study on whether periodically estimated GARCH models would produce the reasonably accurate forecasts compared with daily estimated model. We found that GARCH models, although estimated every 3 month or 6 month, does not deteriorate in forecast accuracy. EGARCH model’s forecast accuracy became worse, when estimated periodically. But the deterioration of forecasting performance is not serious.