Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
A "horse race" is performed among competing models of bilateral exchange rates between the United States and Canada, the United Kingdom, Germany, and Japan. The authors compute posterior odds that any of the competing models is true, using standard Bayes' rule formulae. The distinguishing feature of this horse race is that the posterior odds incorporate the information contained in the covariances among the forecast errors of competing models. It is found, in contrast to Meese and Rogoff (1983), that the random walk model is dominated by one of the Dornbusch overshooting model, the Flexible Price Monetarist Model or the Hooper-Morton Model for all bilateral rates for almost all of the sample. The authors are able to replicate the random walk dominance results by forcing the posterior odds to ignore covariance information, so that in a sense their results "encompass" those of earlier studies. Copyright 1991 by MIT Press.