The RATS Software Forum

Dear Tom,
I am sorry for bothering you so much.I know how the cointergation and error correction models work in time series very well.I also know the meaning of an error correction model.However,I am trying to estimate a system of error correction model in case of panel data.The problem I am facing is the following: In the equation for each variable(its lag as the dependent variable),we have lags of all the variables(including the dependent variable) and an error correction term.My question is how do we introduce that error correction term in the equation,the way it has been done in the following paper:
http://www.sciencedirect.com/science/ar ... 8308001503


Thanks.
Regards.

They're a series of residuals which will be created by whatever instruction/procedure you use to estimate the cointegrating relation. You just put the lag of them into the regression (which is what they describe in the paper). Have you contacted the authors? They might very well have done that with RATS (given the reliance on Pedroni's work).

If you would ask better questions, you would get better answers. Your tendency is to ask very short, very vague questions.

Dear Tom,
Thanks a ton for your reply!
Could you please refer me the original papers that discuss parameter heterogeneity tests and granger causality in panel context?

Thanks.
Regards.

I would suggest you contact Prof. Pedroni. However, what parameter heterogeneity do you mean? There's a whole chapter in Baltagi's Panel Data book on tests of that, but that's probably in a different context. Note that most of them would be described as "obvious", as is the causality test in heterogeneous panels.

Dear Tom,
I am referring to the heterogeneity tests and panel causality tests described in the "Panel data e-course" offered by Estima.

sanjeev wrote:Dear Tom,
I am referring to the heterogeneity tests and panel causality tests described in the "Panel data e-course" offered by Estima.

The heterogeneity test is taken from a replication for Pedroni(2007). The causality test is a case of an obvious test.

TomDoan wrote:

sanjeev wrote:Dear Tom,
I am referring to the heterogeneity tests and panel causality tests described in the "Panel data e-course" offered by Estima.

The heterogeneity test is taken from a replication for Pedroni(2007). The causality test is a case of an obvious test.

The heterogeneity test done in Pedroni(2007) is an F-test while what is described in RATS e-course is a chi-square test.Could you tell me the rationale of converting F-test into chi-square? I need this because I need to mention the details of the test in my work.
Please reply.


Thanks.

Hi,
Is there a way to get the results of group-mean FMOLS(all individual cross-sections results also) in the format of a table in a word document? And also,can we get the values only upto 3 decimal places?

Thanks.

sanjeev wrote:

TomDoan wrote:

sanjeev wrote:Dear Tom,
I am referring to the heterogeneity tests and panel causality tests described in the "Panel data e-course" offered by Estima.

The heterogeneity test is taken from a replication for Pedroni(2007). The causality test is a case of an obvious test.

The heterogeneity test done in Pedroni(2007) is an F-test while what is described in RATS e-course is a chi-square test.Could you tell me the rationale of converting F-test into chi-square? I need this because I need to mention the details of the test in my work.
Please reply.

From the course document:

The following does a heterogeneity test for the slope coefficients using the estimated
individual coefficient vectors and covariance matrices using the restriction
matrix that each individual’s coefficient is equal to the average of all.
This is different from the F in the paper which uses sums of squared residuals,
though both give very clear rejections.

sanjeev wrote:Hi,
Is there a way to get the results of group-mean FMOLS(all individual cross-sections results also) in the format of a table in a word document? And also,can we get the values only upto 3 decimal places?

Thanks.

Reload the report (Window-Restore Report), adjust it however you want and export it:

https://estima.com/ratshelp/reportwindow.html

sanjeev wrote:

TomDoan wrote:They're a series of residuals which will be created by whatever instruction/procedure you use to estimate the cointegrating relation. You just put the lag of them into the regression (which is what they describe in the paper). Have you contacted the authors? They might very well have done that with RATS (given the reliance on Pedroni's work).

If you would ask better questions, you would get better answers. Your tendency is to ask very short, very vague questions.

Can you tell me,how do we retrieve these residuals?

Sorry to bother you for that but I have figured that out.
Can you please tell me,if I should estimate the VECM using pregress command? Should I use fixed effects or random effects option? Also, while conducting the granger causality test,which regression(fixed effects or random effects) are we doing?

I have been able to estimate the VECM for the entire panel as a whole using Fixed effects regression.Can we also estimate for each individual in the panel?

I tried the following program and got zeros against all the coefficients,t-stats and std. errors:

PREGRESS(EFFECTS=BOTH,METHOD=FIXED,smpl=%indiv(t)==1) dprod

# constant ecm dprod{1 to p} dk{1 to p} dtech{1 to p} dcomp{1 to p} dgov{1 to p} defi{1 to p} dfin{1 to p}

In my case,T=35 and N=7.

Could you please help?

That would make sense. You're doing a "panel" regression with only one individual, with time period dummies, which means that you have one dummy for each observation, thus a perfect fit on the dummies alone. You don't want to use PREGRESS---these are heterogeneous regressions, you do one LINREG per individual.

Hi,
I have following two questions to ask:
1. While conducting joint granger causality tests, if we get a large negative value for the statistic, what could be the probable reasons?
2. Under what conditions do the joint statistic and the sum of the individual statistics not match?
3. What would be the corresponding p-value of the sum of the individual t-statistics?
Please reply soon!
Thank you.