[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: Inverse Mills
I have the following model.
Y1 = b0 + a1Y1 + a3X3 + e1 (Y1 is binary)
Y2 = b1 + a2Y2 + a4X4 + e2 (Y2 is observed but its truncated at 0)
I want to estimate both equations simultaneously using probit for Y1 and OLS
for Y2. But because Y2 is truncated, it would not be consistent to use OLS.
I could use a Tobit for Y2 but it is very difficult to program. Morever I
have reason to believe that a two-stage estimator using the Cragg model is
more suitable for the data set that I have.
Hence I plan to do a Cragg Model where I do a probit on Y2* (where Y2*=0 if
Y2=0, Y2*=1 if Y2>0)
Y2* = b1 + a4X4 +e1 ->Save the inverse mills (Y2* is binary)
Then delete all the observations where Y2 is equal to 0 and estimate the
Y1 = b0 + a1Y1 + a3X3 + e1 + inverse mills
Y2 = b1 + a2Y2 + a4X4 +e2 + inverse mills
The inverse mills is included to control for selectivity bias.
Thanks for reading this far. Thus my questions are
1. How do I calculate the inverse mills for the initial probit on Y2*. Is
the following correct?
After the probit for Y2*
predict fitted_name, xbs
gen invmill = normden(fitted_name)/norm(fitted_name)
2. Can anyone see anything fundamentally wrong with the model I am
I have seen similar models estimated using a double OLS in a few finance
papers. (See Barton, J., 2001, Does the use of financial derivatives affect
earnings management decisions? Accounting Review 76, 1-26 and Pincus, M. and
S. Rajgopal, 2002, The interaction of accounting policy choice and hedging:
Evidence from oil and gas firms, Accounting Review 77, 127-160)
Thanks again, I really appreciate any comments.
* For searches and help try: