The Statistical Performance of Econometric and Implied Forecasting Models

Abstract

This study examines the statistical performance of various econometric and implied correlation models applied to over-the-counter currency options. In evaluating the predictive accuracy of the forecast models, statistical performance measures, mainly in the form of the root mean squared forecast errors (RMSFE), are employed. Interestingly, we find that a simple exponential weighting scheme for variances produce the best results. Such models are very simple to estimate and forecast, thus potentially rendering considerably more complex and cumbersome models, such as the multivariate generalized autoregressive conditional heteroscedasticity (GARCH), hardly worth the additional effort.