Announcement

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

  • #16
    I do not know why Cess Jadhav tagged along on a thread for small N large T data, as his data is small T large N variety, and Carlo gave good suggestions regarding large N small T.

    Back to the point of the Original Poster, Professor Wooldridge suggested -xtscc-, and the only problem of -xtscc- is that it is OLS and does not take into account the likely correlation structure in the data, and therefore there are efficiency losses.

    I have written two post estimation commands which calculate (cluster) robust variance and standard errors after -sureg- and -xtgls-, and therefore resolve the problem pointed out by Professor Wooldridge that "The problem with -xtgls- and -xtregar- is that they do not allow for robust inference".

    The -sureg- and -xtgls- are very appropriate methods for the situation OP describes, and which of the two is more appropriate depends on whether one wants to allow only heterogeneous intercepts among the 6 firms, or heterogeneous all estimated coefficients (intercepts and slopes).

    My post estimation command that robustify variances post -sureg- and -xtgls- can be installed by typing from within Stata:

    Code:
     ssc install suregr
    and

    Code:
    ssc install xtglsr

    Comment


    • #17
      Hi, I am trying to do an Unit Root test for my unbalanced panel data. My Country is 78 and Time is 24 years. I don't understand how to detect when N is small and T is Lare, is there are threshold. I am getting confused on which tests to use.

      Comment


      • #18
        There are no strict rules, but your data seems to be more of the large N=78, small T=24 variety.

        What is large and what is small depends on the context, I tend to consider less than 30 small, and bigger than 30 large.

        Originally posted by Sabrin Rahman View Post
        Hi, I am trying to do an Unit Root test for my unbalanced panel data. My Country is 78 and Time is 24 years. I don't understand how to detect when N is small and T is Lare, is there are threshold. I am getting confused on which tests to use.

        Comment


        • #19
          Dear All,

          I am using the system GMM method based on David Roodman's (2009) paper. I have used the following command

          xtabond2 y L.y x1 x2 x3, gmmstyle(l.(x2. x3, x1) ,eq(diff)) twostep robust. However, the coefficient values are very small (.0001977, .0041519, -.0396245)

          may I know what could be the reasons for this?

          Jadhav Chakradhar

          Comment

          Working...
          X