An Evaluation of the Exchange Rate Forecasting Performance of the New Keynesian Model

This paper evaluates the dynamic out-of-sample nominal exchange rate forecasting performance of the canonical New Keynesian model of a small open economy. A novel Bayesian procedure for jointly estimating the hyperparameters and trend components of a state space representation of an approximate linear panel unobserved components representation of this New Keynesian model, conditional on prior information concerning the values of hyperparameters and trend components, is developed and applied for this purpose. In agreement with the existing empirical literature, the paper finds that nominal exchange rate movements are difficult to forecast, with a random walk generally dominating the canonical New Keynesian model of a small open economy in terms of predictive accuracy at all horizons. Nevertheless, the paper finds empirical support for the common practice in the theoretical open economy macroeconomics literature of imposing deterministic equality restrictions on deep structural parameters across economies, both in-sample and out-of-sample.