10-19-2012 06:13 PM
I am estimating a cost model. I want to take a log of cost to ensure positive predicted values. Also, heteroscedasticity is an issue in this case. I did a lot of reading, and finally decided to do GLM using GENMOD with gamma distribution. I have two questions
1) The articles I found says, GLM takes care of the heteroscedasticity and hence the standard errors should be robust, right? Or do I have to specify some options to get robust standard errors?
2) The other option in my case would be to do OLS with robust standard errors, and apply a proper smearing factor to get the cost values. I am using this article as guidance, http://www2.sas.com/proceedings/forum2008/370-2008.pdf . However, I am confused how this would work for out of sample prediction? What would be the concept of variance in this case? For out of sample predictions, Proc score produces negative of the predicted values as residual, which are not really residuals? So, I am not sure what would be the smearing factor for out of sample predictions.
Any help will be highly appreciated.
Thanks in advance,