[Gretl-users] DPANEL questions concerning accessors, a collapse switch and arguments

Riccardo (Jack) Lucchetti r.lucchetti at univpm.it
Wed Jul 3 05:34:03 EDT 2013


>> Have you already thought about incorporating a 'collapse' switch?
>
> Which would do what, exactly?

The 'collapse' switch is, as far as I know, a clever invention by the 
Stata-oriented guys (I believe Roodman himself). In a nutshell, it's a way 
to reduce the number of over-identifying restrictions without sacrificing 
the main idea, so to speak.

For example, imagine you have a DIF-GMM estimator with T=4: you could use 
3 orthogonality conditions

a. \Delta e_{3,i} y_{1,i}
b. \Delta e_{4,i} y_{1,i}
c. \Delta e_{4,i} y_{2,i}

or you could put together a. and c. as

\Delta e_{t,i} y_{t-2,i}

and go down to 2. This means, in practice, that certain matrices get 
'compressed' by squeezing out zeros.

>From a statistical point of view, this amounts to making some extra 
assumptions on the variance of disturbances, compared to ordinary GMM. 
>From a practical point of view, this is often innocuous, although it leads 
to different results in practice, where the differences should be small 
for very large N if the model is well specified. From the regression 
monkey's point of view, this is a great invention, because it multiplies 
the number of legitimate ways in which you can run the same model, thus 
giving you a few more chances to have the little stars in the output 
appear exactly where you want them.

-------------------------------------------------------
   Riccardo (Jack) Lucchetti
   Dipartimento di Scienze Economiche e Sociali (DiSES)

   Università Politecnica delle Marche
   (formerly known as Università di Ancona)

   r.lucchetti at univpm.it
   http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------


More information about the Gretl-users mailing list