Announcement

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

  • Cross-sectional Dependence in Random-effects Model

    Dear there,,,
    I have two questions, appreciate help.
    First Question:
    Since Regression with Driscoll-Kraay standard errors works only for both pooled and fixed-effects models.
    What I shall do to correct the cross-sectional dependence problem in Random-Effects Model, when Hausman test show that random is preferred than fixed.

    Second one:
    Since the cross-sectional dependence is the correlation of two independent signals with the time shift.
    What is the implication of the significant test for the simple regression i.e, with a sole Predictor variable. [Pesaran's test of cross sectional independence = 7.037, Pr = 0.0000]

    Thanks in advance

  • #2
    You can try using cross-sectional dependence robust estimators, such as the CCE and IFE estimators. You can also chose to ignore the hausman test, RE is more efficient, but FE is always consistent which is generally more important (efficiency is nice, consistency is essential).

    Comment


    • #3
      Thanks Jesse for helpful response,
      For the Common Correlated Effects (CCE) I have found some things about on the web, and if I'm right the code would be xtdcce.Yet, for the Interactive Fixed Effects (IFE) few is exist about and the code is not clear.
      However, your suggestion about utilizing the Fixed-effects by the way and regardless the Hausman test seems interesting but how can I theoretically support this option.

      Comment


      • #4
        There's a regife command on ssc that does IFE estimation I think.

        Your beta coefficients will be correct with fixed effects even if Hausman says you can use random effects. Your standard errors might be a bit larger than in random effects, but depending on the field, subject and results, this might not be an issue.

        Comment


        • #5
          Thanks x2 Jesse for help. I just have another question:
          Is there any implication of the significant test for the simple regression i.e, with a sole Predictor variable. [Pesaran's test of cross sectional independence = 7.037, Pr = 0.0000]

          Comment


          • #6
            Originally posted by Soliman Rakha View Post
            Thanks x2 Jesse for help. I just have another question:
            Is there any implication of the significant test for the simple regression i.e, with a sole Predictor variable. [Pesaran's test of cross sectional independence = 7.037, Pr = 0.0000]
            I'm afraid I don't understand your question?

            Comment


            • #7
              Sorry for unclear, I meant performing the cross-sectional dependence test after simple regression i.e, with a sole Predictor variable.
              The model was: xtreg NTD DEFM, fe
              Then: xtcsd, pesaran
              The outcome was: Pesaran's test of cross sectional independence = 7.037, Pr = 0.0000
              Is there any implication of the significant test???

              Comment


              • #8
                It means the residual from your regression is not cross-sectionally independent. Whether that's an issue largely depends on what's common in your literature.

                Comment


                • #9
                  So that was right to perform cross-sectional dependence test after simple regression (only one independent variable), because the cross-sectional dependence is the correlation of two independent signals with the time shift. I just have to be sure if that was right or not???

                  Comment


                  • #10
                    Cross sectional dependence means the various groups in your panel do not move independently over time. Say you have employment in 50 US states, then the CD test will check whether employment moves independently in these states, or whether they are correlated. In most cases, you don't look at the variables themselves, but the residuals after a regression. Then, you check whether the residuals are correlated across states, which could imply there are common trends you are not filtering out. Omitting these trends might in turn bias your results.

                    Comment


                    • #11
                      That's really helpful, thanks xx so much

                      Comment

                      Working...
                      X