Announcement

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

  • Dynamic panel data

    Dear users,

    I have a panel dataset composed of 28 countries (EU-28) for 20 years (1995-2015). I have government expenditure as dependent variable and several country-specific political characteristics as independent variables. I am wondering what is the best model I should use. I expect my dependent variable depends on its own past values, so as I would like to include lagged values of y in the regression. I think at dynamic panel models such as system-gmm, but I know that n>>t. Alternatively, I think at gls. Also, what about country and year fixed efffects?

    Thanks!

  • #2
    You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions.

    A general question like "what model should I use" is hard to answer.

    Comment


    • #3
      Thank you Phil for your suggestion.

      I will try to be more specific. I have a dataset that is quite large in the time dimension (with 20 years) and somewhat small in N=28 countries. Give the number of years ,I may have issues of non-stationarity and I should control for auto-correlation. I think that a regression only including time and id fixed effects does not solve my problem. Theoretically, I guess I should use a dynamic panel model but as far as I know the number of observations I have is too small to get consistent results. I am not such familiar with dynamic panel models.

      I may lag my dependent variable in the following way: reg cgexp L.cgexp controls i.year, cluster(country) but I am not sure to get rid of the above issues.
      In another post, I saw people suggesting XTGLS. Is it a way out of the problem of small N?

      thanks for any suggestion!

      Comment


      • #4
        Hey,

        try to log your variables and use first differences, this may get you rid of the stationarity issue.
        Further, for a dynamic macro panel take a look at the paper by Judson, Owen (1999), you might consider using the LSDVC estimator by Kievit (xtlsdvc) or the GMM Arrelano-Bond estimators (xtabond, xtabond2).

        Best regards
        Tobi

        Comment

        Working...
        X