implement Anderson-Hsiao estimator using xtivreg, fd

[Andrew Zelstra]

Dear everyone,
Hello. I am a new guy here and need some help.
After consulting several web sites and some other posts, I am unsure about the syntax for the Anderson-Hsiao estimator for my panel data set using xtivreg, fd (recommended by Baum: http://fmwww.bc.edu/EC-C/S2013/823/E...n05.slides.pdf).
I would like to predict the expected marketing expenditure using two period lags and past marketing and return on asset(ROA) as set out in equation (1) and then predict expected marketing as per equation (2) shown in the attachment page 9. Source of the equations: Currim, I. S., Lim, J., & Zhang, Y. (2017). Effect of analysts’ earnings pressure on marketing spending and stock market performance. Journal of the Academy of Marketing Science, 1-22.)

So far I have used this Stata code based on other posts:

xtivreg mktg (l.mktg l.roa = l2.mktg l2.roa ), fd
predict unexp_mktg

Question: Does my Stata code reflect the equations?

[Phil Bromiley]

You're more likely to get a very helpful answer if you follow the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output and sample data using dataex. Also, don't post pictures or files - we won't or can't read them.

Your code does estimate the model and does do the predicted value. According to https://www.stata.com/statalist/arch.../msg00280.html it seems like the A-H estimator. If you care about the parameters and standard errors, you might want to use clustered standard errors.

[Sebastian Kripfganz]

There are two variants of the Anderson-Hsiao estimator: (i) differenced model and differenced instruments; (ii) differenced model and level instruments.

With xtivreg, fd you are implementing variant (i) with the differenced instruments. If that is what you want, then you are fine. Otherwise, you might want to use David Roodman's xtabond2 command or my xtseqreg or xtdpdgmm command.

Edit: The implementations with the ivreg command proposed in Phil's link also work.

[Andrew Zelstra]

Thank you for responding to my questions. It is good to know the Stata code works and I will add clustered errors.
I wasn't sure based on the article extract if I needed to do a loop for each firm so your confirmation helps me resolve this doubt.
I will remember the etiquette as well.
Thank you.