Different use of time dummies in xtdpd and xtabond2

[Rodrigo Donini]

Hi to all,

I have a doubt regarding the use of time dummies in xtdpd and xtabond2.

The examples in the xtdpd manual include the time dummies using the div() option, thus they are used as instruments for the differenced equation.

On the other hand, the examples on the 2009 Roodman paper (How to do xtabond2) include time dummies using the ivstyle( ,equation(level)), thus they are used as instruments for the level equation.

Therefore, in dynamic panels, should I use time dummies as instruments for the differenced equation or the level equation?

[Sebastian Kripfganz]

I recommend to specify the time dummy instruments for the level equation only.

Please see the following two discussions on this topic:

• System GMM - Time dummies
• XTSEQREG: new Stata command for sequential / two-stage (GMM) estimation of linear panel models

[Rodrigo Donini]

Mr. Kripfganz,

Thank you for your response. I have read both discussions and they are very informative. I understand why it is better to specify time dummies for the level equation. especially if the weighting matrix is set to h(2), even though I was also advised to use time dummies for both level and difference.

And what about the inclusion of a time trend along with the time dummies for the differenced equation? Which is what I think they do in the examples of the xtdpd Stata manual, when they use the year dummies and the variable "year" as instruments for the differenced equation. Is that a recommended specification?

[Sebastian Kripfganz]

I would treat a time trend exactly in the same way as the other time dummies, i.e. specifying the respective instruments for the level equation only. Conceptually, there is no difference between time dummies and a time trend.