I am estimating an Error Correction Model using the two step approach.

In the usual form the ECM second-step regression includes lagged first differences of independent variables $X_{1}, ..., X_{k}$ :

$\Delta Y_t = \alpha + \beta_1\Delta X_{1,t-1} + ... + \beta_k\Delta X_{k,t-1} + \lambda resid_{t-1} + \varepsilon_t$ $ (1) $

I have seen examples where contemporaneous first differences of $X_{1}, ..., X_{k}$ are included in the second-step equation. Under which assumptions is it possible to include $\Delta X_{1t}, ..., \Delta X_{kt}$ in equation $(1) $? Any textbook reference would be greatly appreciated. Thanks.


The problem with including contemporaneous terms $\Delta X_{i,t}$ on the right hand side of equation $(1)$ is that they may be correlated with the error terms $\varepsilon_t$. In other words, (loosely speaking) $\Delta X_{i,t}$ might be endogenous (precise definition of endogeneity may be a bit more complicated, but that does not hurt the intuition of the argument). In presence of correlation, the OLS estimators for the coefficients on $\Delta X_{i,t}$ will be inconsistent.

The condition under which it is fine to include $\Delta X_{i,t}$ on the right hand side of equation $(1)$ is that $\Delta X_{i,t}$ is uncorrelated with $\varepsilon_t$.

  • $\begingroup$ So, in line of principle, if Xt is the lag of a macroeconomic variable (i.e. the past quarter's GDP growth) this will likely be uncorrelated with future error terms (εt). Is it right? $\endgroup$
    – Math
    Jan 31 '17 at 9:06
  • $\begingroup$ @Math, That is an assumption we make in these models. $\endgroup$ Jan 31 '17 at 9:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.