Announcement

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

  • ARDL ECM bounds test

    Hi,

    I have selected optimal lags for an ARDL with two variables, plus impulse dummies (time series). The impulse dummies is not included in the LR relation. The ARDL is well-specified with respect to serialcorrelation, ARCH, normality, heteroskedasticity and RESET23. When I formulate it to an unrestricted ECM and the associated restricted ECM, these models do not pass all the mentioned diagnostics tests.

    I´m working with I(0) variables and the bounds test indicates the existence of a LR relationship in levels.

    Does anyone know if I could have done something wrong or if it is supposed to be like this?
    Or am I supposed to estimate ECMs with different lags and/or dummies than included in the ARDL?

    I´m so confused! Hope anyone understand my question
    Last edited by Mette Marit Johansen; 22 Oct 2019, 07:34.

  • #2
    The ECM is just a reformulation of the ARDL model. The ardl command actually does not reestimate the ECM but obtains its coefficients directly from the ARDL model estimates. The error terms from the two model specifications are identical. Without seeing the actual commands you have used and the test output you have obtained, there is unfortunately not much I can say about it.

    More on ARDL estimation in Stata:
    Last edited by Sebastian Kripfganz; 22 Oct 2019, 11:46. Reason: Links to further information added
    https://twitter.com/Kripfganz

    Comment


    • #3
      Thank you for answering, Sebastian Kripfganz!

      If I include step dummies and impose Long-run proportionality between my independent and dependent variabel in my ECM, should I include both the coefficient on the step dummies as well as the coefficient on the (y-x) term as part of the constant (in the static long run term)?

      Comment


      • #4
        Sorry, I am not sure I understand your question. What do you mean by imposing long-run proportionality?
        https://twitter.com/Kripfganz

        Comment


        • #5
          By long-run proportionality I mean that both the coefficients on y and x are the same, (or both are equal to one) which means that the LR relation can be written as LR=(y-x- constant) instead of LR=(y-bx - constant).

          If a assume that both of the coeffcients on x and y are equal to one, and I have structural breaks, which I also want to model in the long-run: do I interpret the coefficients (appearing in the static LR-relation )on the (y-x) term and on the dummies as part of the constant?

          Comment


          • #6
            I am afraid I can still not follow your argument. What is the (y-x) term? And how do you impose the long-run proportionality in the ARDL context? (It may be helpful for my understanding to see some actual Stata output.) Whatever these coefficients are, I believe they measure the effect of some variables that vary over time. I thus cannot see how they could become part of the constant term, similarly for any dummy variables.
            https://twitter.com/Kripfganz

            Comment

            Working...
            X