Comparing beta coefficients of two different Stochastic Frontier models.

[Aurora Iannello]

Hi everyone!

I have two hedonic models explaining the annual rental price of two samples of apartments. The two models use the same set of independent variables, but differ in that the dependent variable i(the annual rent) is linear in the first model (Y=R) and logarithmic in the second model (Y= ln(R)).
I need to keep this two distinct functional forms because I need to perform a "cost" Stochastic Frontier Analysis on the first model and a "production" SFA on the second one, and keeping this two different forms gives me the appropriate distributions of the error term (right-skewed in the first case and left-skewed in the second).

I have just estimated the SFA coefficients of the two models (separately) using the command sfcross.

sfcross R X_n, cost d(t)
sfcross lnR X_n, d(t)

Now I want to compare the two sets of beta-coefficients obtained from the two estimations in order to understand if they are similar, but I have no idea of how I could do that. Can anybody help?
Thanks in advance.

[Clyde Schechter]

Now I want to compare the two sets of beta-coefficients obtained from the two estimations in order to understand if they are similar, but I have no idea of how I could do that.

In what sense do you think they might be "similar?" I can't think of any. These are different models, and there is every possibility that the coefficients will not resemble each other in any way--they can even have opposite signs. So what kind of similarity are you hoping to see?

[Aurora Iannello]

Thanks for your answer. I'll try to explain better what I'm trying to do; I'm obviously a beginner, so please tell me honestly if it makes any sense to you or if you can think of better ways to do what I'd like to.

The two hedonic regression models that I am considering refer respectively to uncontrolled and rent-controlled houses. I'd like to check if the differences in price that I observe in the two samples depend only on the inefficiency created by the price control. This is why I want to perform the SFA (cost-type for the uncontrolled houses, whose price is above the frontier/efficient price and production-type for the controlled house, whose price is instead below the frontier): I want to detect the random and non-random inefficiencies, and then I'd like to check if I measure similar price effects of my independent variables (which are the houses' characteristics) on rental prices.

Of course I realize that the different form of the dep variable is one of the problems (as I said, one is ln(R) and the other is R); can I still compare the two set of coefficients somehow?
Do you think it to be so impossible that I obtain (apart from the ln form) similar beta coefficients? My samples are quite large, more than 2000 units each.

Does it make any sense to you?
Thanks for your help.

[Clyde Schechter]

If you were using the same outcome variable in both analyses, it would be simple to quantify the differences between the coefficients of your predictors in the two subsets. You could combine them into a single analysis and use interaction terms.

But the fact that you are using two different outcome variables, as far as I can see, makes the comparison impossible.

If another Forum member thinks I am missing something here, please do chime in.