Modeling and Forecasting Exchange-Rates With a Bayesian Time-Varying Coefficient Model

Abstract

This paper employs a multivariate Bayesian time-varying coefficients (TVC) approach to model and forecast exchange rate data. It is shown that, if used as a data-generating mechanism, a TVC model induces nonlinearities in the conditional moments and leptokurtosis in the unconditional distribution of the series. It is also shown that leptokurtic behavior disappears under time aggregation. As a forecasting device, a Bayesian TVC model improves over a random walk model. The improvements are robust to several changes in the forecasting environment.