- xtcdf - now on SSC: Pesaran (2004, 2013) CD-test for cross sectional dependence

[Jesse Wursten]

Dear all

xtcdf is now available on ssc, with thanks as always going out to Kit Baum. It performs the Pesaran (2004) CD-test for cross sectional dependence, which can be used to test whether your variables or residuals are correlated between groups in a panel setting. E.g. does employment in the different US states follow similar trends or do they move independently? Being aware of such correlations might have an impact on your estimates (see the cross sectional dependence literature), but it's also just good to know these things about your data. The program also reports the mean correlation coefficient.

In the end, the CD test is based on a transformation of the sum of all pairwise correlations. The code is heavily based on -pwcorrf- (also on ssc), which calculates all these pairwise correlations in a more convenient and usually faster way than the official pwcorr command.

You can specify as many variables as you want (it loops internally).

Example usage (output at bottom of post)

sysuse xtline1.dta, clear
xtcdf calories
pwcorrf calories, reshape

This is not the only command out there to calculate cd-tests.
xtcsd was the first, but can only be used as a postestimation command
xtcd is very slow in large datasets and reports the wrong average number of joint observations
xtcd2 does not allow for multiple variables and assumes zero-mean variables (residuals)

Moreover, xtcdf is built to handle panels where some groups do not share (at least 3) observations and hence lead to meaningless correlations. At the time of writing, the other commands behave erratically in this context. On the other hand, xtcd2 can produce kernel densities and a histogram of the pairwise correlations, which can be very useful.

References
Pesaran, M. Hashem. 2004. “General Diagnostic Tests for Cross Section Dependence in Panels.” CESifo Group Munich CESifo Working Paper Series 1229

[Robert Akunga]

Hi Jesse, what if 'xtcd' and 'xtcd2, noestimation' gives you contradicting results?

[Jesse Wursten]

Can you provide the output from the three commands?

[Sorphasith jumong]

@Jesse Wursten can you explain the result ? is it acceptable to test one by one variable (each variable ) separately ? Many thanks

[Jesse Wursten]

Yes no problem to test variables separately

[Dr. Iqra Yaseen]

hi Jesse Wursten thanks for this command.
this commad says that

time variable not set, use tsset varname ...
r(111);

but then when i try to set time variable using

xtset panvar year, yearly

i get the error of "repeated time values within panel". I know the problem for this error lies with tha panvar. then when i try to set a panvar that uniquely identifies the data set using

egen panvar = group (year exporter importer sector_code

, the dataset gets xtset perfectly with the results

panel variable:  panvar (weakly balanced)
        time variable:  year, 2008 to 2022
                delta:  1 year

but the problem arises using xtcdf command as the error "in-sufficient observations" pops up. again the problem lies with panvar as this panvar has frequency (1) less than the required frequency (3). so in all cases i am unable to use the xtcdf command. any solution.

[Jesse Wursten]

I think you should exclude "year" from the panel var

[Dr. Iqra Yaseen]

Jesse Wursten thank you so much. that is exactly what I did, and it worked.