Forecasting exchange rates using asymmetric losses: A Bayesian approach

The forecasting of exchange rate returns has long been an issue in finance literature. The use of the best forecasting model is usually sensitive to the data frequency and the sample period used. Model evaluation is usually based on either minimizing error losses or maximizing profit strategies and other likelihood-based measures. Although, much work has been devoted to model evaluation based on maximizing profits strategies little to no work has been devoted to the issue of estimating a forecast model under the same principles. Here, we propose a Bayesian framework that estimates exchange rate models by considering measures such as directional accuracy and trading rules in a form of asymmetric loss functions. Estimation is implemented using Laplace-type estimators applied in cases where the likelihood function is not of a known form. We illustrate this method using simulated and real weekly exchange rate series. The results demonstrate that the use of profit maximizing strategies within estimation can significantly improve the forecasting ability of certain exchange rate models.