Panel Cointegration Test

[allister]

Hi Tom
I have a question about the use of panelfm and pancoint.src, written by Pedroni.
I have a rather simple question that I need some insight regarding testing for cointegration in a panel setting using the Pedroni Test. I have read alot of papers and still have a hard time deciding when to reject the null hypothesis on no conitegration.


After the estimation (with panelfm) you can obtain the results of Pedroni tests with pancoint.

However, I was wondering about which are the appropriate critical values to be taken into account, in 2 different respects, I have five regressors including the dependent variable and a intercept.

1. small sample. I have a panel with N=13, T=26. How do I take into account this fact?

2. number of regressors. I have a model with five regressors. Do I have to use Table II in Pedroni 1999 (Critical values for cointegration tests in heterogeneus panels with multiple regressors)?


Thanks a lot for the help that you could give me!!!!


In addition is there anyway to code the test Kao and Westerlund test in Rats?

[TomDoan]

Pedroni(1999)'s table 2 doesn't give critical values---it gives means and variances of the test statistics under the null. You don't need to use that because those are incorporated into the standardized test statistics reported by @PANCOINT. As with many panel data statistics, those have double-asymptotic distributions (large T followed by large N) and are asymptotically N(0,1) under those circumstances. Your N is probably less a problem for sample size than your T. You might want to check with Prof. Pedroni as to whether the asymptotic distribution is reasonable with your size data set.

[sanjeev]

Dear Tom,
Please tell me how do we decide whether we have to reject or not each of the seven statistics under Pedroni's panel cointegration test? So specifically I am asking for the critical values or p-values in the output.

Please reply soon!

Thanks.
Regards.

[TomDoan]

From the output:

All reported values are distributed N(0,1)

under null of unit root or no cointegration

[sanjeev]

Dear Tom,
Thanks for your reply! But what about the sign of the statistic?

Regards.

[TomDoan]

Dear Tom,
Thanks for your reply! But what about the sign of the statistic?

Regards.

You would reject non-cointegration in the left tail, that is, big negative numbers would cause you to reject non-cointegration in favor of cointegration.

[sanjeev]

Dear Tom,
Please tell me do we have the options to choose lag length for the ADF regressions according to different criteria like AIC,SBC,HQIC Also what is RATS taking as default criterion?


Thanks.
Regards.

[TomDoan]

Yes. And that's described in the documentation: https://estima.com/ratshelp/index.html? ... edure.html. The default is fixed lags.

[sanjeev]

Dear Tom,
Is it possible to conduct panel cointegration tests with structural break?

Thanks.
Regards.

[TomDoan]

There is a literature on that (Westerlund and Carrion-i-Silvestre are two people who have done work on that). So far as I know, none of that has been written in RATS.

[sanjeev]

Hi,
Can we retrieve p-values also along with the Pedroni's statistics?

Thanks.

[TomDoan]

No. Those are N(0,1). Do you really need p-values for something as standard as that?

[sanjeev]

Hi,
Is there any way to select the deterministic term while testing for the existence of a cointegrating vector and then estimating the model using group-mean FMOLS?

[TomDoan]

You mean other than the documented DET options in both?

[sanjeev]

You mean other than the documented DET options in both?

No, I mean on what basis do we decide which DET should we include in our estimations,constant,trend or nothing?
Please reply.

Thanks.