장기기억 속성을 이용한 주가 변동성예측에 관한 연구

This paper investigates the long-memory property in estimating and forecasting Korean stock market return volatility. Volatility is a central role in derivative pricing, portfolio allocation, risk management, and performance evaluation of funds. In consequence, there has been much research on estimating and forecasting return volatility. Little has been studied about forecasting return volatility, however, by exploiting the long-persistent property in Korean stock market. In this paper, we estimate and forecast return volatility by employing the long-memory property. For this purpose, we use the Fractionally Integrated GARCH and Fractionally Integrated EGARCH models. The estimation results and forecasting performance of the long-memory volatility models are compared with those obtained from the short-memory volatility model such as GARCH model. Many studies suggest that the conditional volatility of stock returns follows a long-memory process; shock dissipates at a slow hyperbolic rate. This type of persistence cannot be appropriately modeled by standard GARCH type models. In this aspect, the long-memory volatility models are needed to explain the high-persistent volatility. Baillie et al. (1996) and Bollerslev and Mikkelsen (1996) suggest the FIGARCH and FIEGARCH models which introduced the high-persistent property in the standard GARCH and EGARCH models. Therefore, we used these long-memory volatility models in forecasting Korean stock market return volatility. We identified some key findings from the results. First, the auto- correlations for the absolute and squared returns decline at very slow rate which suggest that there is a long-memory property in Korean stock return volatility. Second, it is difficult to say that there is a high-persistent property in the level of stock returns. However, the return volatility follows a long-memory process. The estimated values of long-memory parameters, d of FIGARCH (1, d, 0) and d of FIEGARCH (1, d, 0) models, are 0.356 and 0.584, respectively, and are statistically significant at the 1% significance level. Third, we conducted out-of-sample one-step-ahead and ten-stepahead forecasts using the FIGARCH (1, d, 0) and FIEGARCH (1, d, 0) models and compared the volatility forecasts of both fractionally integrated volatility models with those of the GARCH (1, 1) model as a benchmark. We found that the long-memory volatility models produce superior out-of-sample forecasts in terms of root mean squared error (RMSE), mean absolute error (MAE), and of Mincer-Zarnowitz regression. In addition, relative forecasting performance of the ten day ahead forecasts is better than that of the one day ahead forecasts. These findings suggest that the long-range volatility models are useful tools in forecasting the volatility of asset returns as well as pricing derivatives and hedging risks. The results of this study also will facilitate to induce variance swaps and variance futures in the Korean financial market.