This paper analyzes the exchange rate in a ''no-arbitrage'' or ''real business cycle'' equilibrium model and provides empirical evidence for this model vis-a-vis PPP. Our contribution is to show, based on a generalization of the equilibrium model of exchange rates, that (i) the test equation linking the exchange rate to fundamentals should allow for international heterogeneity in time preferences and risk attitudes, as well as noise-that is, the model should not be tested as an exact relation; (ii) empirical work should use levels of variables rather than first differences; and (iii) tests on the existence of long-run relations should be complemented by test on the signs of the coefficients. Our empirical work shows that, as a long-run relation, the generalized model outperforms PPP; moreover, under the generalized model, quarterly deviations from the estimated long-term relation have a half-life of only four to five quarters.