On Sun, 22 Feb 2004, Vincenzo Verardi wrote:
> 1) run the regression lny=xb
> the I would type the command
> predict lnyhat, xb
> 2) I would then exponentiate my result generating a new variable
> gen ypred=exp(lnyhat)
> I would then have the predicted y.
This is not correct, because the desired quantity is E(y), not
exp(E(logyhat)). Because
logy = logyhat + err
one can exponentiate both sides to get:
exp(logy) = exp(logyhat + err)
and can then take expectations:
E(y) = exp(Elogy)) = E(exp(logyhat + err))
It is not, however, valid to break up the right side as
E(y) = E(exp(logyhat)) + E(exp(err))
because exp is not a linear function. One needs to integrate
the expected value explicitly. I have a memory, probably
incorrect, that E(y) = exp(logyhat) + 1/2*sigma-squared,
where sigma-squared is the variance of the error term, so
that E(y) is a little higher than exp(logyhat). But that
memory dates back to graduate school and isn't relaible
at this time....
Steve Schmidt
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/