09-09-2014 01:24 AM
Hi, I want to do a panel regression/pooled OLS regression in sas. My data looks like
permno date return size
10000 1 0.5 1
10000 3 0.2 0.1
10001 1 1 10
Please note: This is just a demonstration. There are many more firms(permnos) and for each firms there are many dates observations.
Originally date is actually monthly date like 03/31/2000. I manually changed the date variable to the numbers start from the beginning of my data. 1 means the first month in my sample. Permno is a unique ID for a firm. It's not necessarily continuous such as 10000,10001,10002... Also, the date is not necessarily continuous neither.
My questions are:
(1) Do I have to create a new firm count variable as 1, 2, 3, ....
(2) discontinuity in permno and date is a problem or not?
(3) How to do a panel regression? I assume to use proc glm to do it.
(4) As a pooled regression with time trend control. Do I need to create a time dummy variable?
(5) How to control for correlation between different firms as well as between different dates? Do I need to apply Whites or Newey West? If yes, can I do it in the regression with some options?
09-09-2014 01:14 PM
Do you have SAS/ETS licensed? If so, then your best approach to all of this will be PROC PANEL. If not, then PROC MIXED may be much superior to GLM, as it can handle the autoregressive error structure that is likely to be the case.