Announcement

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

  • system gmm(or/and difference gmm) with xtabond2 for economic convergence model

    Hello!
    I am trying to estimate a conditional beta convergence model with gmm (sys and diff gmm). Below is my estimated equation.
    Click image for larger version

Name:	Capture.JPG
Views:	1
Size:	12.8 KB
ID:	1469480



    lnYit=gdp in level, yit=lnY(it)-lnY(it-1) thus basically yit is gdp growth.
    I found many convergence literatures do not use the yit-1 (the lagged dependent variable), but just the term lnYit-1. But one of the key papers closely related to my work includes the yit-1 as dependent variable to check the persistency of yit. I have questions at this point.

    1. If the yit-1 is included, doesn't this mean that we should allow AR(2)? Because, given that yit=lnY(it)-lnY(it-1) and y(it-1)=lnY(it-1) - lnY(it-2) , we obviously see the 2nd order correlation. However the key paper only allows until AR(1). Or is my understanding totally wrong?

    2. In other convergence literatures that do not use y(it-1), they check the finite sample bias by the coefficient beta(for lnYit-1) which should lie between OLS and Within Group estimator. But my key paper check this with rho(for yit-1). The reason to check the persistency is to ensure to minimize the presence of weak instruments that can occur in DIFF-GMM, right? Then, does it mean that I also have to check the range of all coefficients of endogenous/predetermined regressors? If I am going to put Xit-1 in gmmstyle of xtabond2 command, meaning that I am using instruments for Xit-1, Should I also compare its range with OLS and Within Group estimator?

    Thank you very much! I'd appreciate any comments.
    Last edited by Hyejin Lee; 08 Nov 2018, 08:46.

  • #2

    3. Can I say the additional moment condition for sys-gmm is valid when overidentifying restrictions (hansen test) get rejected, but Difference-Hansen is not?
    Whichever I change the inclusion of exogenous variables(I tried putting variables one by one either to gmmstyle or ivstyle), the overidentifying restrcition is always rejected.. I am aware of the bug with xtabond2 with time dummy, which was discussed in other threads in this forum. And, there is low chance of misspecification because the model has a strong theoretical ground. That is why I leave the rejection of overidentifying restrictions out of discussion.

    I understood my question #1, can anyone help me for the questions #2 & #3? Your comments will be tremendously helpful. Thank you!
    Last edited by Hyejin Lee; 14 Nov 2018, 04:08.

    Comment

    Working...
    X