Cointegration, error correction and exchange rate forecasting

The finding that error correction models do not forecast better than the corresponding first difference models has been explained predominantly in terms of errors in the estimation of the adjustment mechanism. We argue against the proposition that the restrictions implied by economic theory must be imposed to enhance forecasting power because the underlying theory may not be necessarily valid. We put forward the proposition that an ECM and the corresponding first difference model have similar dynamic structures, which means that the relative performance of the two models is an empirical issue. The results do support this proposition and have some practical implications.