Re: st: within estimator and effecomic effects

[Christopher Baum <kit.baum@bc.edu>]

On Feb 1, 2012, at 2:33 AM, Erasmo wrote:

> To assess economic significance, people in finance (including myself)
> usually rely on the effect of 1 standard deviation increase in the
> independent variable (multiplied by the coefficient estimate). The
> standard deviation is usually the sample standard deviation. In
> principle, this seems incorrect when one is using the within estimator
> (such as the estimator from xtreg). In this case, it would seem more
> natural to rely on the within-unit standard deviation.
> 
> But I am wondering whether this is the correct way of looking at things.
> 
> I would appreciate if you could share any thoughts on this.

I don't see what the problem is. If you presume the model is y = X b + D g + epsilon, with D the matrix of units' dummy variables, you are proposing
to consider a 1 sigma change in X times the appropriate b. Although when you apply the within transformation, the sigma of the within-transformed variable will be smaller than that of the original data, the model is unchanged, and you can still sensibly consider a one-sigma change in the original X times the respective coefficient. If you estimated the model without the within transformation and with dummies, the question would not arise. So it should not arise here either, and you should use the overall sigma of X, not the sigma of within-transformed X.

Kit

Research Papers in Economics (RePEc)
http://repec.org




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/