VAR model with population weights

[Tina Haussen]

Dear all,

I am about to re-estimate a VAR model from a published paper with 23 countries over 30 years with variables X and Y. The VAR model in the paper is "population-weighted". So, I initially thought the STATA command could look like

xtset country year
pvar X Y [weight=population], lags(3)

Regarding this, I have two questions:

1. As far as I understand the description of the STATA command pvar, weighting is not allowed. Does anyone know a possibility how to incorporate a weighting by population size in my estimation?
2. I asked the author of the original paper how he did it. He kindly answered that he has not used STATA and that he has implemented the weighting procedure manually (utilizing that weighted least squares is equivalent to ordinary least squares on the weighted variables). Then he has "transformed the model by multiplying ALL model variables (i.e., also the constant term and the year dummies) by the square root of the weighting variable, i.e. the square root of population."

--> OK, I can run simple weighted OLS equations instead of using the pvar (or var) command. But then a new question arises, namely how then to estimate the impulse response function (which is essential in VAR estimations)? If I would use pvar or var for estimating, I could then use pvarirf or similar to estimate the impulse response function. But if I apply simple OLS models (in order to include the weighting) then how to calculate the IRF?

I really hope that someone has a solution for my problem. Thanks in advance,
Tina

[Tina Haussen]

Dear All,

I was actually hoping to get some good clues here. Could someone at least tell me whether my questions are too trivial because there is a solution that I could have found myself (but haven't found)? Or is there actually no solution for my problems (weighting at var and pvar or irf after OLS)?

Thanks and best regards,
Tina