CCE and xtdcce

[Ali Bahador]

Bests,
Ali

[JanDitzen]

Hi Ali,
I am assuming your question relates to the user written command xtdcce, respectively xtdcce2, which replaced xtdcce. Please make sure you use xtdcce2, as this is the latest version of the program and my answer is related to it.

Let me start how xtdcce2 treats the common effects. xtdcce2 without further options assumes that the common effects are unknown and in line with Pesaran (2006) and Chudik and Pesaran (2015) replaces them with cross sectional averages to obtain a consistent estimate of b_mg. To do so, xtdcce2 calculates the cross sectional averages as defined in crosssectional() and then partials them out. In Pesaran (2006) the observed common effects are partialled out as well (see the definition for M_w and H_w, equation (15), which includes D, the observed common effects). The reason is, that the main interest of the paper (and so xtdcce2) is a consistent estimation of b_mg.

Coming back to your question: My assumption is you are interested in estimating A_i. The option crosssectional() defines the variables to be partialled out, i.e. the cross sectional averages. What you could do is, you create (or already have) the variable Gamma_t (the time individual common effects), but you do not define it in crosssectional(). In a theoretical way, you would move matrix D from the matrix M_w to the matrix which contains the other observed dependent variables (usually X).
Your example would read:

xtdcce2 y gamma x , crosssectional(y x)

With this setting you would account for unobserved common effects (via the cross sectional averages) and for the observed common effects (via Gamma_t). However you need to thing about two points: 1. Do you have constant and if so, what kind of constant (hetero or homogeneous across cross sectional units) and 2. do you assume unobserved effects.

Hope this helps!
Best,
Jan