Create Risk Prediction Variable from 2 Different Heckman Correction Models

[JJ Singh]

Hi everyone,

I'm wondering if there's a way to create a risk prediction variable from the results of two different Heckman Correction models--one for Selection Bias and one for Self-Selection Bias. I think I've seen this done in academic literature before but the paper didn''t explain its methodology in detail. For example, let's say I have;

// For Self-Selection Bias (Variables for Heckman: lnsales, lnemp, fcf_w, mktshr, mss, hhi, lnbseg, lngseg, unrate):
heckman status c.preld c.durld c.postld i.cu c.jdg i.presid c.retexl c.retvolatl c.retskewl c.turnovrl c.bm_w, select(status = i.regul c.lnsales c.lnemp c.fcf_w c.mktshr c.mss c.hhi c.lnbseg c.lngseg c.unrate c.bm_w i.cu c.jdg i.presid c.retexl c.retvolatl c.retskewl c.turnovrl) nocons twostep first mills(imr) //vce(bootstrap)

/* (After running the first heckman bit of code just above, type "drop imr" in the command line before you run the second heckman bit of code just below here.) */
// For Selection Bias (Variables for Heckman: FPS variable, fcf_w):
heckman status c.preld c.durld c.postld i.cu c.jdg i.presid c.retexl c.retvolatl c.retskewl c.turnovrl c.bm_w c.fcf_w, select(status = i.fps c.bm_w i.cu c.jdg i.presid c.retexl c.retvolatl c.retskewl c.turnovrl c.fcf_w) nocons twostep first mills(imr) //vce(bootstrap)


(Note that variables preld, durld and postld needed to be removed from the 1st stages of the models because of the economic nature of those variables.)

The top model involves 9 "Heckman variables," while the bottom one only involves 2. One such variable is common to both models--fcf_w.
Any advice would be greatly appreciated, thank you!!
J.J.

[JJ Singh]

Maybe I should try to clarify my question. I'm quite inexperienced with Stata and econometrics in general, so maybe I worded the question in an obtuse way. Stata does have a command called predict that I've come across, but I'm not sure it would be appropriate in my case. Is it correct that if you use the predict command right after running a logit regression, you can create a predicted risk variable which you can then use in a subsequent logit regression? For example:

logit y x1 x2 x3 x4, vce(robust)
eststo m1
predict predrisk1
logit y predrisk1 x2 x3 x4, vce(robust
eststo m2

Here I believe I'm using a fitted predicted risk variable instead of x1 in the 2nd Stage. Is that correct? And, if correct, is there any application of this to my Heckman question?
Thanks very much!!
J.J.

[JJ Singh]

Maybe I should try to clarify my question. I'm quite inexperienced with Stata and econometrics in general, so maybe I worded the question in an obtuse way. Stata does have a command called predict that I've come across, but I'm not sure it would be appropriate in my case. Is it correct that if you use the predict command right after running a logit regression, you can create a predicted risk variable which you can then use in a subsequent logit regression? For example:

logit y x1 x2 x3 x4, vce(robust)
eststo m1
predict predrisk1
logit y predrisk1 x2 x3 x4, vce(robust)
eststo m2

Here I believe I'm using a fitted predicted risk variable instead of x1 in the 2nd Stage. Is that correct? And, if correct, is there any application of this to my Heckman question?
Thanks very much!!
J.J.