panel group FMOLS

[atarca]

I could not manage to include neither an intercept nor a trend term into the RATS(8.0) codes so that I could have an estimate of the intercept (or a trend) coefficient in the panel group FMOLS regression results. Pedroni (2004, Econ. Theory, 20: 597-625) and Pedroni (2007, J. Applied Econometrics, 22: 429-451) report estimates for the slope parameters as well as the average mean and average trend values with their standard errors and significance probabilities. I would appreciate it if one could suggest the necessary changes on the panelfm.src and paneldols.src files given on the webpage of ESTIMA.

Regards,

atarca

[TomDoan]

The updated versions which handle those are now available: @PANELFM and @PANELDOLS

[atarca]

It helped me a lot. I greatly appreciate your efforts. Thank you very much Tom.
I also tried to obtain residuals computed by %%ibetas, but I could not succeeded. How can I change the row below to obtain what I want?

* Define the full-sample residuals
set %resids startl endl = %if(%valid(u),%eqnrvalue(0,t,%beta),%na)

Regards,
atarca

[TomDoan]

It helped me a lot. I greatly appreciate your efforts. Thank you very much Tom.
I also tried to obtain residuals computed by %%ibetas, but I could not succeeded. How can I change the row below to obtain what I want?

* Define the full-sample residuals
set %resids startl endl = %if(%valid(u),%eqnrvalue(0,t,%beta),%na)

Regards,
atarca

For the individual-specific residuals:

set %resids startl endl = %if(%valid(u),%eqnrvalue(0,t,%xcol(%%ibetas,%indiv(t))),%na)

[celineboulenger13]

Hi, for @panelfm how can we recover the intercept value and the R-squared, LM, RESET and HE parameters?

Thanks!

[TomDoan]

Hi, for @panelfm how can we recover the intercept value and the R-squared, LM, RESET and HE parameters?

Thanks!

For @PANELFM, there is a separate intercept for each individual (not one intercept), and they are treated as nuisance parameters. Why do you think those other statistics exist in any meaningful way? There is no R^2 because it's not a least squares procedure. LM statistic for what? RESET against what?

[celineboulenger13]

So is there any way to get one intercept? @panelfm is used to find the cointegrating relationship between variables and an intercept is needed for that. When looking at other papers using Panel FMOLS they all report an intercept value.

Thank you!

[TomDoan]

So is there any way to get one intercept?

Not with @PANELFM. It's for heterogeneous cointegration vectors, so not only do the intercepts vary, but so does the cointegrating vector. A single intercept would make no sense.

@panelfm is used to find the cointegrating relationship between variables and an intercept is needed for that.

That's not true. Cointegration requires that a linear combination be stationary. A stationary process can have a non-zero mean. The Johansen estimates for cointegrating vectors almost never include an intercept---those are only present for very specific forms of cointegration (with mean restricted to the cointegrating vector to prevent drift).

When looking at other papers using Panel FMOLS they all report an intercept value.

Such as? Even if you assume a common CV, I find it hard to believe that the intercepts wouldn't be heterogeneous.

[sanjeev]

Dear Tom,
I ran panel group FMOLS command on version 8 of RATS,it gave me the results.But I also want to get the results for a subset of my whole time period.Please let me know what command I should use for that.
Thanks,
Regards.

[TomDoan]

Dear Tom,
I ran panel group FMOLS command on version 8 of RATS,it gave me the results.But I also want to get the results for a subset of my whole time period.Please let me know what command I should use for that.
Thanks,
Regards.

Use the SMPL option to isolate the time periods that you want. See https://estima.com/forum/viewtopic.php?f=8&t=1467 for an example of @PANELFM using SMPL. He's using it to do different groupings of countries. See "Selecting Subsamples" for the type of SMPL option you need for subsamples based upon time.

[sanjeev]

Dear Tom,
I am estimating my cointegrating vector using grouped-mean panel fmols.My estimation period is 1990-2014 and I want to control for the crisis period(2007-09) using a dummy variable.Can you please let me know how I can do that?

Thanks.
Regards.

[TomDoan]

You can't. FMOLS isn't designed to handle things like that. Any change to the deterministics requires a whole different asymptotic calculation for the correction.

[sanjeev]

Dear Tom,
While estimating a model using group mean FMOLS,can you tell me what does the RATS assume about kernel and bandwidth options? Also do we have the option to change the default settings?


Please reply soon!

Thanks,
Regards.

[TomDoan]

The LAGS option controls that: https://estima.com/ratshelp/panelfmprocedure.html

[sanjeev]

Dear Tom,
I estimated my model using group-mean FMOLS technique on both Eviews version8 and RATS version 8.When I compared my results,I found out that most of the times they are same.But for one of the cross-sections,the coefficients of two variables are drastically different by the two softwares.Indeed the two are giving me opposite signs.Also I tried to change the default settings in RATS but all in vain.
Could you please help me out?

Please reply as soon as possible!

Thanks,

Regards.