st: RE: RE: RE: correlation, panel data

[Nick Cox <n.j.cox@durham.ac.uk>]

Put -capture corr- rather than -corr-. 

This rather underlines that if some of your panels are very short, then those correlations will be based on very few values, with consequent problems. For example, a correlation based on two distinct y and two distinct x points will necessarily be 1 or -1. A correlation based on no or one point cannot be calculated. 

Nick 
n.j.cox@durham.ac.uk 

Biljana Dlab

I tried to adjust the code, and am getting good results for the first two companies, but then it stops.

This is how I did it:

. tsset  gvkey_n  fyear
       panel variable:  gvkey_n (unbalanced)
        time variable:  fyear, 1971 to 2001, but with gaps
                delta:  1 unit

. levelsof gvkey_n, local(levels)

. gen corr=.
(7609 missing values generated)

. qui foreach l of local levels {
. corr cf r_d if gvkey_n==`l'
. replace corr=r(rho) if gvkey_n==`l'
. }
no observations
r(2000);

Nick Cox

One possibility without installing anything is to modify this recipe:

sysuse auto
levelsof rep78, local(levels)
gen corr = .
qui foreach l of local levels {
	corr mpg weight if rep78 == `l'
	replace corr = r(rho) if rep78 == `l'
}


*
*   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/