A problem with xtdcce2

[Rouzbeh Parsi]

Hi folks

I faced an issue using Xtdcce2, i'm trying to run a model like:

xtdcce2 y l.y x l.x z l.z W l.W , cr(y x z) cr_lags(3)

but the result changes if the order of variables is changed, for example if I run this model I find the results quit different:

xtdcce2 y l.y z l.z W l.W X l.x , cr(y x z) cr_lags(3)

JanDitzen can you please help me? is that normal? what is the problem here?

[JanDitzen]

This should not happen. Does xtdcce2 issue an warning message? Do you use any dummies or time invariant variables?

[Rouzbeh Parsi]

yeah, I got a few dummies and I receive a collinearity warning, but the issue persists even if I remove the dummies. Moreover, the outcome varies when I request the constant (reportc).

[JanDitzen]

It sounds to me like you have collinearity problems. xtdcce2 checks for collinearity and then removes from the right. For example say x2 and x3 are collinear, then using the command line xtdcce2 y x1 x2 x3 x4, nocross, x3 will be removed. Likewise, if you change the order and x3 is listed before x2, then x2 will be removed. Since results change when you add the constant, it sounds to me at least one of your variables is constant across the cross-section dimension.

I would carefully check if any variables are constant within each cross-section. It would also be helpful if you could provide a copy of the warning.