Empirical Models of the Exchange Rate: Separating the Wheat from the Chaff

Various popular exchange rate models (a standard monetary model, a portfolio balance model, and sticky-price models) are estimated and evaluated using U.S.-Canadian data for the 1970s. Nonnested hypothesis tests demonstrate that none are correctly specified. The data suggest: 1) the exchange rate persistence observed is not fully explained by any of the models; 2) the small Durbin-Watson statistics indicate longer lags are required; 3) the current account is a useful explanatory variable; 4) the portfolio balance model fits the data well, but has potentially serious problems measuring the stock of foreign assets.