Date:         Sun, 8 Oct 2006 22:22:05 -0700
Reply-To:     David L Cassell <davidlcassell@MSN.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David L Cassell <davidlcassell@MSN.COM>
Subject:      Re: Correction for heteroskedasticity and autocorrelation in a
              fixed effects model
In-Reply-To:  <1160083708.424865.323030@b28g2000cwb.googlegroups.com>
Content-Type: text/plain; format=flowed

gdfernan@SYR.EDU wrote:
>Hi All
>
>I need to run a fixed effects model on a panel data set which has
>heteroskedasticity and autocorrelation. I use the following code to run
>the fixed effects model {Code in curly parenthesis are my comments}
>
>proc glm data=c.datae;
>class cusip {variable identifying the firm} year {variable identifying
>the year};
>model con=cusip year size tenure/solution;
>run;
>
>Is there any way of correcting for heteroskedasticity and
>autocorrelation in SAS. I normally use the Newey-West correction with
>proc model using the kernel correction (Shown below)
>
>
>proc model data=Merge2;
>parms Int CoeFF_LnMVE CoeFF_ROA CoeFF_RET;
>Comp= Int + CoeFF_LnMVE*LnMVE + CoeFF_ROA*ROA + CoeFF_RET*RET;
>fit Comp / gmm kernel = (bart,1,0.383439 {Bandwith calculation});
>instruments lnMVE ROA RET ;
>run;
>
>I would be most grateful for some clarification.
>
>Thanks
>Guy Fernando
>PhD Candidate (Accounting)
>Syracuse University

Normally, the way to handle autocorrelation is to model it, and then
include that structure in the model.  If you have a time series, I
recommend that you analyze it and incorporate the information.
PROC MODEL will certainly allow you to specify an autocorrelation
structure.

As for heteroskedasticity, it depends.  HOW are your data heteroskedastic?
Is the variability changing over time, or is it changing across panels, or
is
it changing with the size of one or more of your regressors, or is it
changing with the vlaues of your dependent variable?  What does the
heteroskedasticity represent in your model?  Can it be caused by a couple
outliers?  Does it indicate that you have a mixture of panel behaviors?
There's a lot of room for analysis and functionality here.

Sorry I can't give you better advice from way over here...
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
Be seen and heard with Windows Live Messenger and Microsoft LifeCams
http://clk.atdmt.com/MSN/go/msnnkwme0020000001msn/direct/01/?href=http://www.microsoft.com/hardware/digitalcommunication/default.mspx?locale=en-us&source=hmtagline