Announcement

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

  • Cross-sectional dependence test

    When I perform the cross-sectional dependence test on each of my variables separately, I get the same statistic and the same p-value for all variables.
    Is there a problem ? If so, at what level, for example?
    Thanks in advance for your answers.

  • #2
    When I perform the cross-sectional dependence test on each of my variables separately, I get the same statistic and the same p-value for all variables.
    Is there a problem ? If so, at what level, for example?
    Thanks in advance for your answers.

    Comment


    • #3
      Hi Wendel,
      I think it would be helpful if you provide us with the following details: 1) which program do you use, 2) The output of the test, 3) some info about the data.
      Cheers,
      Jan

      Comment


      • #4
        Hi JanDitzen,

        I am using panel data with N=80 and T= 19, after using the command " xtcsd, pesaran ". I got the error 2001," Error: The panel is highly unbalanced.
        Not enough common observations across the panel to perform Pesaran's test.
        insufficient observations
        r(2001);"

        so to check the cross-sectional independence, can I use xtcd2 ?
        and while using xtcd2, I am getting


        Pesaran (2015) test for weak cross-sectional dependence.
        Residuals calculated using predict, e from xtreg.
        (321 missing values generated)
        Unbalanced panel detected, test adjusted.

        H0: errors are weakly cross-sectional dependent.
        CD = 0.000
        p-value = 1.000

        please suggest is it correct to use xtcd2 instead of xtcsd, pesaran ?

        Best Regards
        Neha

        Comment


        • #5
          xtcsd, pesaran and xtcd2 are testing the same hypothesis with the same test statistic. Therefore the results should be equivalent. Looking at the output from both programs tells me that there is something going on with your dataset. Is it possible that your panel is unbalanced in such a way that there are not many observations across the same time periods? In this case the estimation of the correlations might be impossible, leading to the error message displayed by xtcsd.


          Comment


          • #6
            Hi JanDitzen,
            you are right that there are not many observations across the same time periods for the dependent variable only. But I still have a doubt that as xtcd2 is giving results for cross-sectional independence and I am getting "1" so can we get " 1" as a p-value in the results, Or do I have some issue in the dataset?
            Thanks a lot for your responses.

            Best Regards
            Neha

            Comment


            • #7
              I would say this is an issue with the dataset. xtcd2 is "stupid" with regard to the number of common observations across units (something I should actually improve...) and ignores it. Given that your test statistic is zero (is it exactly zero?), it seems impossible to calculate any correlations.

              Comment


              • #8
                yes, the value of the CD is exactly zero.
                my dependent variable is intra-FDI. I am using a log-log model and I assume the missing value is equal to zero, if I am taking log(1+ FDI) then it is showing results for the xtcsd, pesaran.results are as follows:-

                Pesaran's test of cross-sectional independence = 56.198, Pr = 0.0000 Average absolute value of the off-diagonal elements = 0.436

                Earlier, the place of missing values were kept blank(when it was giving error(2001).

                so my doubt is when I am assuming zero in the place of missing value, they are showing completely different results for xtcd2( p-value is "1") and xtcsd(p-value is zero) by just changing the missing value. please suggest whether I should fill in the missing value or not?

                Comment


                • #9
                  I think the log-log model specification is the source of the problem. If FDI is zero, then you take the log(1) which is zero. If intra-FDI (i.e. FDI between two cross-sections) is zero for the entire time period, then it is impossible to determine any dependence between the two units.

                  Can you please post the regression results from xtreg? This might be helpful to understand this further.

                  Comment


                  • #10
                    for taking a log of FDI where the value is zero, we have added a number to the full series except at the place of the missing value. In the place of zero observation, we have taken a fixed value to take the log.
                    The results from random effect results are attached here. I have uploaded the pdf as I was unable to paste the text due to the large number of words.
                    Please check the file and give your feedback.
                    Attached Files

                    Comment


                    • #11
                      What do you mean with "added a number to the full series" and "a fixed value to take the log"? I know there is some discussion what a missing value implies in trade datasets, but my uneducated prior would be that you can set it to zero. However then taking the log is making things difficult.

                      If you look at the results of your xtreg output, you see that at least one cross-section has only one time period. Thus for those it will be very hard to determine any cross-section dependence.

                      Comment


                      • #12
                        In my dependent variable, some of the values were negative and zero, and to take log was difficult in that case so to take log, I have added a number to the whole series to make it positive and get the value in the place of zeros.
                        after adding the number, I took the log then I checked the cross-section.

                        As one of the cross-sections is having only one value so should I drop those cross-sections which have very few values.

                        Thanks for the reply.

                        Comment


                        • #13
                          after deleting the cross-sections which are having very few observations.
                          I am getting results from command xtcsd pesaran abs.

                          is it right to do this to quote the results or apply some other method to check cross-sectional independence?

                          Comment

                          Working...
                          X