Residual testing for Pooled mean group estimator

[Svetlana Kovaleva]

Hi all,

I would like to investigate the relationship between gasoline and crude oil prices.

For these purposes I am applying the user-written command xtpmg (available from SSC) that implements the pooled mean group (PMG) estimator. My dataset is balanced and has a dimension of T =108(months) and N=68(regions). I am wondering how I can run diagnostic checks of residuals after xtpmg command (e.g, test for autocorrelation, heteroscedasticity).





Svetlana

[Anat Tchetchik]

Dear all,
I am looking for an answer to Svetlana Kovaleva question regarding how to run diagnostic checks of residuals after -xtpmg- command. In Blackburne and Frank's (2007) article in The Stata Journal, their is no mentioning of any post estimation options

http://www.stata-journal.com/sjpdf.html?articlenum=st0125

any hint will be greatly appreciated

[Vincent Lee]

Dear Svetlana and Anat,

I'm also looking for the answer to the question on running the diagnostic checks of residual after xtpmg command. May I ask if any of you have found a solution? Could you kindly share it? Much appreciated.




Vincent

[Jesse Wursten]

In my own research I've applied the xtqptest command for serial correlation to the residuals generated by xtmg (similar to xtpmg, different long run assumption). It produced decent size/power results in monte carlo exercises.

[Alain Kabundi]

Does anyone know how to extract residuals from xtpmg (pmg) regression? Using "predict residuals" yield results which are different from the residuals I obtain by running the same model in Eviews.

[Manoel Calvo]

Dear Alain,

I obtain the same fitted values than in Eviews applying the option "full" and with the following code:

gen yhat = .
forval i = 1/38 {
predict temp if id==`i',eq(id_`i')
replace yhat = temp if id == `i'
drop temp
}

my data base have 38 individuals, so the loop is predicting the values with the PMG estimation of each individual (long-run parameters are the same but the short-run different). Important the "id" variable is the Panel variable (i).
Then generate residuals substracting from the dependente variable "yhat":

gen residuals = d.tasa - yhat

This residuals are exactly the same as in Eviews.

Hope i have been helpful.
Manoel.

[Alain Kabundi]

Thank you, Manoel!