Right way of sorting before asreg (reproducability or random), Are out of sample FM slopes reproducable?

[Julien Maas]

Every time I run the following code I get different results (altough similar) for the out of sample predictive model. And a different number of observations. I do not get the same issue with the in-sample FM slopes.

For instance,

xtset permno date

asreg yearly_return ROA, newey(17) window(date -120 -1)

My first guess was that this was due to random sorting or something similar.

For in sample slopes I do not get the same issue.

Any intuition behind this?

[Julien Maas]

Note.

I think I found the solution.


Since I am doing cross sectional regressions I can simply sort based on the date and permno, to always start with the same sorting and get consistent results.

sort date permno 
 asreg yearly_return ROA, newey(17) window(date -120 -1)