Greetings, I'm facing some doubts regarding the solution of Driscoll-Kraay to account for the standard errors of the estimators with the phenomena of cross-section dependence.
From what I've read in the article of Hoechle of the Stata Journal (2007) when cross-sectional dependence is occurring the best approach is to estimate the regression model with the Driscoll-Kraay robust standard errors. I have a Panel with large T and small N, according to that article it can be implemented.
Here I ask, by using these Driscoll-Kraay errors in the regression, the cross-sectional dependence has been accounted? as Hoeckle states in his exercise with financial data. However, one should re-test for cross-sectional dependence after using the regression with the Driscoll-Kraay standard errors? or it's non-sense since the estimation has already taken into consideration and if it was before it's going to be after.
Assume I used a Distributed-LAG pooled OLS and get this using the CD test given by xtcd2 result.
So in this context I reject null and accept the errors of the pooled have a strong cross sectional dependence.
Now I perform the regression with Driscoll-Kraay errors like:
And the results of the xtcd2 test show this:
I got the exact same results of strong cross-sectional dependence, however, the robust standard errors with the DK are way different from the OLS.
My ultimate question is that If I should worry that the Pesaran test after regression with robust errors of driscoll-Kraay is a concern? or it doesn't matter since the estimators are already taking this existence and increasing the values to give valid statistics regarding each parameter.
From what I've read in the article of Hoechle of the Stata Journal (2007) when cross-sectional dependence is occurring the best approach is to estimate the regression model with the Driscoll-Kraay robust standard errors. I have a Panel with large T and small N, according to that article it can be implemented.
Here I ask, by using these Driscoll-Kraay errors in the regression, the cross-sectional dependence has been accounted? as Hoeckle states in his exercise with financial data. However, one should re-test for cross-sectional dependence after using the regression with the Driscoll-Kraay standard errors? or it's non-sense since the estimation has already taken into consideration and if it was before it's going to be after.
Assume I used a Distributed-LAG pooled OLS and get this using the CD test given by xtcd2 result.
Code:
Pesaran (2015) test for weak cross-sectional dependence. Residuals calculated using predict, residuals. (96 missing values generated) H0: errors are weakly cross-sectional dependent. CD = 44.557 p-value = 0.000
Now I perform the regression with Driscoll-Kraay errors like:
Code:
xtscc Change_Index L.Change_Index L2.Change_Index L3.Change_Index L4.Change_Index L5.Change_Index L6.Change_Index L7.Change_Index L8.Change_Index L9.Change_Index L10.Change_Index L11.Change_Index L12.Change_Index L13.Change_Index L14.Change_Index L15.Change_Index L16.Change_Index L.Ex_Rate_Change L2.Ex_Rate_Change L3.Ex_Rate_Change L4.Ex_Rate_Change L5.Ex_Rate_Change L6.Ex_Rate_Change L7.Ex_Rate_Change L8.Ex_Rate_Change L9.Ex_Rate_Change L10.Ex_Rate_Change L11.Ex_Rate_Change L12.Ex_Rate_Change L13.Ex_Rate_Change L14.Ex_Rate_Change L15.Ex_Rate_Change L16.Ex_Rate_Change L.GOLD_Change L2.GOLD_Change L3.GOLD_Change L4.GOLD_Change L5.GOLD_Change L6.GOLD_Change L7.GOLD_Change L8.GOLD_Change L9.GOLD_Change L10.GOLD_Change L11.GOLD_Change L12.GOLD_Change L13.GOLD_Change L14.GOLD_Change L15.GOLD_Change L16.GOLD_Change L.PLATINUM_Change L2.PLATINUM_Change L3.PLATINUM_Change L4.PLATINUM_Change L5.PLATINUM_Change L6.PLATINUM_Change L7.PLATINUM_Change L8.PLATINUM_Change L9.PLATINUM_Change L10.PLATINUM_Change L11.PLATINUM_Change L12.PLATINUM_Change L13.PLATINUM_Change L14.PLATINUM_Change L15.PLATINUM_Change L16.PLATINUM_Change L.SILVER_Change L2.SILVER_Change L3.SILVER_Change L4.SILVER_Change L5.SILVER_Change L6.SILVER_Change L7.SILVER_Change L8.SILVER_Change L9.SILVER_Change L10.SILVER_Change L11.SILVER_Change L12.SILVER_Change L13.SILVER_Change L14.SILVER_Change L15.SILVER_Change L16.SILVER_Change L.WTI_Change L2.WTI_Change L3.WTI_Change L4.WTI_Change L5.WTI_Change L6.WTI_Change L7.WTI_Change L8.WTI_Change L9.WTI_Change L10.WTI_Change L11.WTI_Change L12.WTI_Change L13.WTI_Change L14.WTI_Change L15.WTI_Change L16.WTI_Change L.BRENT_Change L2.BRENT_Change L3.BRENT_Change L4.BRENT_Change L5.BRENT_Change L6.BRENT_Change L7.BRENT_Change L8.BRENT_Change L9.BRENT_Change L10.BRENT_Change L11.BRENT_Change L12.BRENT_Change L13.BRENT_Change L14.BRENT_Change L15.BRENT_Change L16.BRENT_Change country_1 country_2 country_3 country_4 country_6 Cont_Change, ase
Code:
xtcd2 Pesaran (2015) test for weak cross-sectional dependence. Residuals calculated using predict, residuals. (96 missing values generated) H0: errors are weakly cross-sectional dependent. CD = 44.557 p-value = 0.000
My ultimate question is that If I should worry that the Pesaran test after regression with robust errors of driscoll-Kraay is a concern? or it doesn't matter since the estimators are already taking this existence and increasing the values to give valid statistics regarding each parameter.
Comment