Introduction

Forecasting the nominal exchange rate path is one of the most challenging aspects of an inflation-targeting framework. According to Bank of Brazil estimates, the pass-through from nominal exchange rate movements to inflation is around 10% in each quarter. Therefore, an accurate forecast of the nominal value of the currency is very important for the efficiency of an inflation-targeting regime. If the evaluation of the future exchange rate path can be made more precise, it may reduce the variance in output and inflation.

Uncovered interest parity (UIP), which relates the expected nominal depreciation to the nominal interest rate differential, has been a popular model for exchange rate forecasting. But UIP has been questioned as an adequate tool to forecast future exchange rates because many empirical tests have found a negative correlation between exchange rate changes and the interest differential, in contradiction to what is predicted by UIP. This situation has led us to consider what can be gained and lost with other models for forecasting the exchange rate.

A simple alternative is to assume that the exchange rate follows a random walk and is not cointegrated with any exogenous variable for which we have data. The expected future exchange rate therefore should be equal to the current value. This first approach, although simple and transparent, does not preclude the risk of occasional large forecast errors in the exchange rate and hence inflation.