Binary Variable

[Fredrick Pavlecki]

Hello guys,

I've been recently reading a paper by Berk, Jonathan B. and van Binsbergen, Jules, (2016) "Assessing asset pricing using revealed preference". In their paper they run the following regression:

sign(F) = β0 + β1 sign(ALPHA) + p

, where sign(F) and sign(ALPHA) take on values in {−1, 1}.

I cannot understand how to run such a regression in Stata. I was wondering if any of you know how to do so?

Thank you

[Clyde Schechter]

Well, one possibility is to just do a linear regression with sign(F) as the outcome variable. It's a little bit weird to do that, but there's nothing in principle wrong with it. The error distribution likely will not be homoscedastic, but using robust standard errors would get around that.

Alternatively, you can think of this as a dichotomous variable and recode it to {1, 0} instead of {1, -1}. Then do a logistic or probit regression on the recoded variable, understanding that you are fitting a model of the probability of sign F being 1.