Volatility Forecasting Models for The Won-Dollar Exchange Rate

This paper attempts to compare the forecasting performance of various volatility models for weekly won-dollar exchange rates in terms of R² and news impact curve with diagnostic tests. We consider both linear GARCH model and nonlinear models such as EGARCH, GJR, GTARCH, and LSTARCH. Empirical results for the R² comparison suggest that the EGARCH model has an edge in predicting a large value of variances while GTARCH and LSTARCH models outperform alternative models in predicting a small value of variances. According to the news impact curve analysis, all models except the GARCH demonstrate asymmetric movements of exchange rate volatility around the threshold. That is, the conditional volatility increases more in case of a positive exchange rate shock (depreciation) than a negative exchange rate shock (appreciation). Although it is difficult to find definite grounds for choosing between the various models, we conclude from a series of sign bias tests that the EGARCH model is the best in predicting the volatility of weekly exchange rates among alternative models considered.