Proc Panel and Parks

[Lilly12]

Hi,

I have a balanced dataset of 1285 observations (257 firms, 5 years). I have been trying Proc Panel using Parks method.

The SAS code is as follow:

proc sort data=Warranty.Paneldataset;

by ASXCode Year;

run;

proc panel data=Warranty.Paneldataset outest=Warranty.Reg1_Parks;

ID ASXCode Year;

Reg1_Parks: model Leverage = WBtoTA Size AssetCollateralValue ROA ROAVolatility NonDebtTaxShield TobinsQ TotalAccruals HerfindahlIndex RDIntensity GICS1010 GICS1510 GICS2010 GICS2020 GICS2030 GICS2510 GICS2520 GICS2530 GICS2540 GICS2550 GICS3010 GICS3020 GICS3030 GICS3510 GICS3520 GICS4510 GICS4520 GICS4530/ Parks;

quit;

Here is what is I get in the log

"NOTE: Estimated autoregression parameter RHO=1.032 did not satisfy |RHO|<1.0 criteria for cross section ASXCode=WEB.  It will be set to a value within range.

ERROR: The number of observations is too small to permit estimation of the regression parameters in model statement REG1_PAR."

The NOTE is written 34 times for different firms.

Please note that all of the below are dummy variables 

GICS1010 GICS1510 GICS2010 GICS2020 GICS2030 GICS2510 GICS2520 GICS2530 GICS2540 GICS2550 GICS3010 GICS3020 GICS3030 GICS3510 GICS3520 GICS4510 GICS4520 GICS4530

DW and Godfrey tests confirm the presence of autocorrelation.

Any ideas why Parks is not working, how can I solve this problem.

Thanks

[ets_kps]

Hello Lilly,

I have talked to the developer of the PROC and it appears the problem can been seen by looking at http://support.sas.com/documentation/cdl/en/etsug/65545/HTML/default/viewer.htm#etsug_panel_details2...

You will notice that the degrees of freedom correction requires the length of the time series to exceed the number of parameters to be estimated. In essence, the Park's method is expecting a long time series and you only have 5 periods of data.

Also, the developer told me that a literature on this estimator is very scarce. There is a much more supportive literature on using the Arrellano-Bond estimator for  short T, big N dynamic panel models. An example appears here http://support.sas.com/documentation/cdl/en/etsug/65545/HTML/default/viewer.htm#etsug_panel_example0...

Please let know if I can be of additional help.-Ken

[Lilly12]

Hi Ken,

Thank you for your reply.

My model is not dynamic so I do not think using Arrellano-Bond estimator is wise. Is there any option model linear regression for panel data with AR(1) disturbance?

Thank you in advance

[ets_kps]

Thanks Lilly,

Any model that has a lagged value of Y as a dependent variable (such as with an AR specification) links the current value to the past and is therefore considered dynamic (as least as SAS is concerned).  The Park's method would tend to be a nice option, but of course, you don't have enough data to pull that off.

An alternative to modeling your AR(1) disturbance would be to use the HAC option in PANEL and simply correct your standard errors for Heteroscedasticity or Autocorrelation of unknown form. It sounds as if you might have tried this.  These options are new in SAS/ETS 12.1.  Ensure that you have 12.1 and it should work.

Thanks-Ken

[Lilly12]

Aha i see, I am working on SAS 9.3. Thank you Ken. I would need to do that manually then :smileycry:

[SteveDenham]

I was afraid this kind of confusion would result under the new versioning system.  SAS/ETS12.1 is SAS9.3.  Unless you are missing a service pack, everything Ken mentioned should be available.

Steve Denham

[Lilly12]

The HAC option on Proc Panel statement is not recognised!! is it not valid anymore?

[Lilly12]

I am not sure how to add the HAC option. Can I download a certain service pack from the SAS hot fixes?

[ets_kps]

Hi Lilly,

Here is the link to the 9.3 maintenance release with the 12.1 Analytic Products. http://support.sas.com/software/maintenance/

The link you are really interested in appears at the bottom of the page. It requires you to make a "request" for the update.

Best of luck.  With this install you will get quite a few new features in ETS (Econometrics) including new Bayesian tools and your HAC options that appear for panel data and autoregressive models.

Ken