Forecast Evaluation Of Recent Exchange Rate Models

This thesis uses Bayesian methods to forecast exchange rates and compares the results to existing models such as OLS and the random walk. We focus on commodity currencies where mean reversion is thought to be more plausible. To estimate the Bayesian models, two different techniques are applied. In Dynamic Model Averaging (DMA), we use an analytical approach using Kalman filters for the variation in time as well as the change in posterior model probabilities. In Bayesian Model Averaging (BMA), we employ the traditional numerical method of Markov Chain Monte Carlo Model Composition (MC3) to simulate the posterior model probabilities. Assessment of the prediction performance is done by means of Diebold-Mariano tests. The study shows that the methods used yield good forecasting results when compared to traditional methods. In particular, the dynamic methods of model averaging or model switching prove to perform best.