A Forecasting-model Comparison for Whole-day Realized Volatilities Including Overnight Variations

Abstract

One of the most important advantages of the realized volatility (RV) is that it is easy to implement and it provides better forecast for volatilities than other statistics based on daily data sets. But time is an important consideration when we calculate the RVs. Working time RV is easy to implement, but cannot represent whole day variations. For this reason, we use the time series models with the realized volatility with different time zone: first, RV for business time of stock trading, second, RV for business time and non-business time, i.e. overnight, of no stock trading. The properties of the RVs of the log returns of the Korea Composite Stock Price Index(hereinafter KOSPI), the Korean won / US dollar exchange rate(hereinafter KRW/USD), and the US Standard & Poors 500 index(hereinafter S&P 500 index) are analyzed. HAR, LHAR, THAR models are investigated which consider long memory, asymmetry, heteroskedasticity properties of the RVs. The performances of the RVs pseudo out-of-sample forecasts are compared. And predictions are improved by adding implied volatilities to the models.;주식 시장의 실현 변동성은 구현하기 쉬울 뿐만 아니라 예측의 효율성이 여러 연구를 통해 입증되었다. 그런데 장내 시간의 실현 변동성은 구현하기 쉬우나 하루 전체의 변동성을 보여주지는 못한다. 그리하여 본 논문에서는 Korea Compsite Stock Price Index(이하 KOSPI), KRW/USD 환율, Standard & Poor’s 500 Stock Index(이하 S&P 500 Index) 세 가지 지수에 대하여 장내 시간만 고려한 실현 변동성과 장내, 장외 시간을 모두 고려한 실현 변동성을 구현하고, 어떤 특징을 가지는지 살펴보았다. 실현 변동성의 장기 기억성과 비대칭성, 이분산성을 반영한 HAR, LHAR, THAR 모형을 구현하고 pseudo out-of-sample 예측을 해보았다. 또한 내재 변동성을 추가한 모델을 사용함으로써 실현 변동성의 예측력을 개선하였다.