Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Engle-Granger 2-step ECM

    Hi everyone!

    I need your advice for Engle-Graner 2-step procedure in Stata. Suppose that the model uses 5 variables and among them there is one cointegrating relationship, then I can apply the ECM using the first differences, right? The command is : egranger var1 var2 var3 var4 var5, lags(1) ecm reg Then, the 1-st step regression shows me the long-run equilibrium and the short-run is from the 2nd step regress. I'm confused as far as the second-step, I have to check the p-values for significance as in the typical OLS or I have to compare t-statistics with Engle-Granger tables? I'm really confused. Does anybody have an idea? Thank you in advance for your response.

  • #2
    I have to check the p-values for significance as in the typical OLS or I have to compare t-statistics with Engle-Granger tables?
    I'll admit, I'm a little confused with what you're asking. I would suggest posting your output using CODE delimiters. I would also suggest you use Sebastian Kripfganz's arld (autoregressive distributed lag) command instead. An ARDL is essentially a reparameterized error-correction model and it has many advantages over the Engel Granger approach. You can find virtually all the information you'll need here, but there are also helpful threads on this topic here and here.

    Comment


    • #3
      Sorry, it's ardl not arld.

      Comment


      • #4
        Thank you very much for your response.I want to run this ECM for my thesis project, that's what my supervisor adviced me in order to investigate the shor-run relations among the variables. But, I'm not sure if I can do this because I have 5 variables and the Engle-Granger is applied only for a bivariate regression, right? Does Engle-Granger procedure fit in the case of multivariate regression? Or, I have to perform a VEC or ARDL as you suggested me above? Thank you in advance.

        Comment


        • #5
          I would recommend an ARDL over the Engle-Granger approach. I'm not positive, but it looks like you specify the number of lags in the egranger command, whereas you can fit the 'best' model in ardl by minimizing either AIC/SCB. The link above will tell you all you need to know, why, and how to implement the estimation properly in Stata. If this is for your thesis, I would strongly advise against searching for one specification. You can do egranger, ardl, and a VEC and then discuss the differences/drawbacks /assumptions in each of the approaches and report out on the results (all will give you a long run equation -- error correction term).

          Comment

          Working...
          X