panel group FMOLS

[TomDoan]

1. Is there any compelling reason to believe that these series are cointegrated?
2. Is there any compelling reason to believe that these series are cointegrated, with rank one?

If they're not cointegrated, then you are doing a panel version of spurious regressions. If they're cointegrated, but of rank higher than one, then FMOLS (or DOLS, or almost anything else) won't work because there are then an infinite number of different possible cointegrating vectors. I'm not sure I've ever seen a serious piece of econometric work that claimed a single cointegrating vector among 7 I(1) variables.

[sanjeev]

Hi,
Actually I have been trying to put and read countries' names also in my file.But RATS doesn't
accept the names.It only allows Id's for countries.Is it because I am using RATS 8.1 version?
Please reply.

Thanks.
Regards.

[TomDoan]

No. You can't read a series of strings into RATS. That, of course, has nothing to do with the numerical issues.

[sanjeev]

Dear Tom,
Please tell me,what is the null hypothesis in the parameter heterogeneity test?

Thanks.

[TomDoan]

That they (the parameters being tested) are homogeneous (i.e. equal).

[sanjeev]

Dear Tom,
Using group-mean FMOLS technique, I am estimating a model in which I include an interaction term of two of the explanatory variables.Is there any way that I can check the joint significance of the interaction term with the coefficient of one of the variables included separately in the model? Secondly suppose we estimate two models for two different industries.Is there any way that we can test the significance of the difference in the coefficient of a variable in the two models so as to make a comparison between the two models?

Thanks.
Please reply!

[TomDoan]

Dear Tom,
Using group-mean FMOLS technique, I am estimating a model in which I include an interaction term of two of the explanatory variables.Is there any way that I can check the joint significance of the interaction term with the coefficient of one of the variables included separately in the model?

There are no explanatory variables---other than (possibly) time dummies, everything is an endogenous variable. I'm not sure what you're trying to do, but @PANELFM is clearly the wrong tool to do it.

Secondly suppose we estimate two models for two different industries.Is there any way that we can test the significance of the difference in the coefficient of a variable in the two models so as to make a comparison between the two models?

If you're willing to assume the estimates from the two industries are independent, then you have %BETA and %XX from each, and a Wald test comparing coefficients in the two sets of estimates can be done using the fact that the covariance matrix of the difference between the BETA vectors is the sum of the covariance matrices.

[sanjeev]

Dear Tom,
Sorry to frame the question in a wrong way.I am trying to estimate the coefficients of the RHS variables using Group-Mean FMOLS afetr having got a strong evidence of the existence of a cointegrating vector between the variables in the model(using Pedroni's panel cointegration test). Suppose my model is Y= alphaX1 + Beta1X2 + Beta2X1*X3. Is there any provision to test the significance of the coefficient(Beta1 + Beta2X2) in RATS?

Dear Tom,
Using group-mean FMOLS technique, I am estimating a model in which I include an interaction term of two of the explanatory variables.Is there any way that I can check the joint significance of the interaction term with the coefficient of one of the variables included separately in the model?

There are no explanatory variables---other than (possibly) time dummies, everything is an endogenous variable. I'm not sure what you're trying to do, but @PANELFM is clearly the wrong tool to do it.

Secondly suppose we estimate two models for two different industries.Is there any way that we can test the significance of the difference in the coefficient of a variable in the two models so as to make a comparison between the two models?

If you're willing to assume the estimates from the two industries are independent, then you have %BETA and %XX from each, and a Wald test comparing coefficients in the two sets of estimates can be done using the fact that the covariance matrix of the difference between the BETA vectors is the sum of the covariance matrices.

[sanjeev]

Dear Tom,
Could you please tell me how to get the program for it?

Secondly suppose we estimate two models for two different industries.Is there any way that we can test the significance of the difference in the coefficient of a variable in the two models so as to make a comparison between the two models?

If you're willing to assume the estimates from the two industries are independent, then you have %BETA and %XX from each, and a Wald test comparing coefficients in the two sets of estimates can be done using the fact that the covariance matrix of the difference between the BETA vectors is the sum of the covariance matrices.

[TomDoan]

Dear Tom,
Sorry to frame the question in a wrong way.I am trying to estimate the coefficients of the RHS variables using Group-Mean FMOLS afetr having got a strong evidence of the existence of a cointegrating vector between the variables in the model(using Pedroni's panel cointegration test). Suppose my model is Y= alphaX1 + Beta1X2 + Beta2X1*X3. Is there any provision to test the significance of the coefficient(Beta1 + Beta2X2) in RATS?

First of all, Beta2 X2 doesn't appear anywhere in that. Did you mean Beta2 X3?? Second, Beta2 X2 or Beta2 X3 isn't a "coefficient" since part of it is a series. I don't think this is a matter of you "framing" the question wrong. You seem to be treating FMOLS like it's a Y on X's regression, and it's not. You can't mix interaction terms of I(1) variables in.

[sanjeev]

Hi Tom,
Could you please let me know how important is it include common time dummies into the model(having checked for the panel cointegration) while estimating it using panel group FMOLS? Does inclusion of time dummies in the model make it robust?
Please reply!
Thanks.

[TomDoan]

From Pedroni's JAE paper:

For all of the results reported in table I, we also provide results for the case in which the data has been demeaned with respect to cross sectional dimension for each time period, which is comparable to the inclusion of time dummies. These serve to capture certain forms of cross sectional dependency, and the effect of common disturbances that impact all members of the panel. Consequently, when the data has been demeaned in this way, we can also interpret the results as relative to any common stochastic trend that drives the different countries of the sample. It is important to account for the possibility of such common effects when computing the panel statistics, and in table I, the null of no cointegration is rejected in all cases for the panel once we demean relative to the time effects.

[sanjeev]

Dear Tom,
Can we have country fixed effects in panel cointegration and group-mean FMOLS estimation?

Thanks.

[TomDoan]

Dear Tom,
Can we have country fixed effects in panel cointegration and group-mean FMOLS estimation?

This goes way beyond fixed effects. Not only is there a separate intercept ("fixed effects") but there is a separately estimated cointegrating vector for each country.

[sanjeev]

Board index ‹ Econometrics Issues ‹ Panel Data
Change font size
User Control Panel (0 new messages) • View your postsFAQLogout [ sanjeev ]

Dear Tom,
Thanks a lot for your reply!
Could you let me know the code to make a table of descriptive statistics in a panel data?

Please reply soon!

Regards.