Has anyone gone through some papers using Vector Error Correction Models in causality applications with more than one cointegration vectors, say two. I guess there will be more than one ECM terms. How to assess the endogeneity of the left hand variables if t-stats on different ECM coefficients yield different (conflicting) results. For left hand side variables to be endogenous should we have both ECM coefficients negative and significant?
Javed Iqbal
