FMOLS Equation objects

[cgreene]

For a COINTREG FM-OLS or CCR object or command, it would be very useful if the procedure created output matrices for the estimates of Omega-hat, Lambda-hat, and (for CCR) sigma-hat. I would find this useful for two reasons: (1) for FM-OLS this output can be used as a basis for constructing the single breakpoint F-statistic and also the unknown breakpoint Sup-F test statistic as in Hansen (J of Bus & Econ Stat, 1992). (2) When results are varying substantially by choice of sample, kernel or bandwidth, being able to look at whether one or all variance estimates are substantially changing would be useful.

In my documentation I don't see any mention of an ability to access these matrices, but Eviews must be creating them as part of the COINTREG calculations.

[EViews Glenn]

Sounds perfectly reasonable to me. I'll look into how hard it would be to export the calculations in a future version of EViews.

I will note that we've added the stand-alone computation of long-run variances and covariances to the series and group objects in EViews 7 so that you can do these calculations while you wait for us to implement your request :)

[EViews Glenn]

Looks like we were (sort of) ahead of you.

Existing cointreg equations have been saving the relevant information. We weren't sure if there would be general interest in getting access to these intermediate calculations, but I wanted to have them available in the event that we decided that we wanted to provide them. So adding them was easy enough since it didn't involve any backward compatibility issues.

The currently posted EViews 7 has added the following data members to equations:

@lrcov
@contempcov
@lambda12cov

which return the relevant parts of the calculation in FMOLS and CCR. The first is the two-sided long-run covariance for the system residuals. The second is the contemporaneous correlation of the residuals from the first-stage regression and the other residuals in the system. The latter is the one-sided long-run covariance for the covariance between the first-stage residuals. Note that we don't provide the entire covariance here as only this portion is relevant to the calculations. I believe that @contempcov is only available for CCR.

[ClintonGreene]

Thanks. This is useful for two reasons:

1) To be sure I understand the procedure, I am trying to reproduce the COINTREG results using my own Eviews program. No luck yet, my own fmols coef estimates differ by about 10%.

2) The intermediate results are needed to construct the stability test statistic in Hansen (1992, J of Bus and Econ Stat) "Tests for Instability with I(1) Processes". This is where I am really headed, but first want to play with (1) more.

Thanks again

[EViews Glenn]

There was a program we provided a while ago which went through the calculation of FMOLS. You may or may not find it useful. It's in the repository somewhere.