An in-sample evaluation of exchange rate models: In search of scapegoats

A modified dynamic model averaging framework, which allows for inferences regarding the shifting relevance and significance of explanatory variables, is employed to evaluate the in-sample performance of exchange rate models. This analysis is based on a set of 16,384 model specifications derived from 14 canonical and newly introduced explanatory variables. Our findings indicate: (a) frequent changes in the model specification that best describes an exchange rate, (b) the relevance of individual explanatory variables is not stable over time and varies across exchange rates, with these variables exhibiting differential and sometimes opposing effects, and displaying non-uniform strengths across different exchange rates and periods, (c) the combination of economic and/or financial variables that enhances the empirical evidence of purchasing power parity (PPP) is specific to each exchange rate. These results underscore the challenges associated with employing a single exchange rate model or the scapegoat hypothesis to describe all exchange rates across all time periods.