Biased Estimation in a Simple Extension of a Standard Error Correction Model

This paper considers an expectations augmented version of the Engle and Granger (1987) error correction model and shows that standard inference about the adjustment coefficients can be severely biased. This has implications for long–run causality and impulse–response analysis in particular. However, a sometimes simple remedy exists which only requires some additional regressions. The results are illustrated using U.S., German and Swiss data.