Forecasting Volatility Using Competitive Time Series models --Evidence from Foreign Exchange Market

The paper compares the out-of-sample forecasting performance of monthly volatility using daily USD/GBP prices from the post-Bretton Woods era to the present day. Competitive time series models including RW, HM, MA, WMA, ES, EWMA, AR (I) MA, regression, and regression on dummy variables, ARCH, GARCH, TARCH, EGARCH, PARCH, CGARCH, and each of ARCH class models on dummies are employed. All models are run under both rolling and recursive windows for 1-step-ahead forecast. The forecast performance is evaluated by forecast accuracy and efficiency tests. ES is the best although the various model rankings are shown to be sensitive to the error statistics used to assess the accuracy of the forecast. ARMA (1, 1) under a recursive window is recommended as a result. It is concluded that non-ARCH class models are superior to ARCH class models. However, ARCH class models take predominance on overprediction that is more heavily penalized