Forecasting exchange rate in a large Bayesian VAR model : the case of Taiwan

We study the out-of-sample forecasting performance of 32 exchange rates vis-a-vis the New Taiwan Dollar (NTD) in a 32-variable vector autoregression (VAR) model. The Bayesian approach is applied to the large-scale VAR model (LBVAR), and its (timevarying) forecasting performance is compared to the random-walk model in terms of both forecast accuracy and Giacomini-Rossi fluctuation tests. We find the random-walk model outperforms the LBVAR model in a short-run forecasting competition. Moreover, the dominance of a random-walk in the competition is stable over time. Accordingly, we do not find any benefit of incorporating a rich set of information in predicting the exchange rates vis-a-vis the NTD.