Arnold Zellner and Franz Palm (1974) show that comparing the actual with the implied stochastic process es generating the endogenous variables in a system of dynamic structu ral equations provides important information about the system's corre ct specification. The authors apply their methodology to structural e xchange-rate models. They find that the log of the bilateral exchange rate is generally well approximated by a random-walk model. Thus, th e stochastic processes generating the exogenous variables should also be random-walk models, which in not borne out by our empirical resul ts. They suggest a reconciliation of their results based on a decompo sition technique developed by Stephen Beveridge and Charles R. Nelson (1981).