Model Uncertainty and Exchange Rate Forecasting

We propose a theoretical framework of exchange rate behavior where investors focus on a subset of economic fundamentals. We find that any adjustment in the set of predictors used by investors leads to changes in the relation between the exchange rate and fundamentals. We test the validity of this framework via a backward elimination rule which captures the current set of fundamentals that best predicts the exchange rate. Out-of-sample forecasting tests show that the backward elimination rule significantly beats the random walk for four out of five currencies in our sample. Further, the currency forecasts generate economically meaningful investment profits