FM-OLS and t-statistics

[ryans]

Hi,

I've read on here that t-stats reported in Fully Modified OLS regression are valid under the appropriate assumptions. Can anyone tell me what these assumptions are, and point me in the direction of a reference for these assumptions?

Thanks

Ryan

[EViews Glenn]

They're not easy to explain in a few sentences (well perhaps others might be able to do so, but I can't)...

The references in the manual are the best place to start. For FMOLS, Hansen's (1992) Journal of Econometrics article (Hansen, Bruce E. (1992a). "Efficient Estimation and Testing of Cointegrating Vectors in the Presence of Deterministic Trends," Journal of Econometrics, 53, 87-121) is probably the clearest. Similar conditions hold for DOLS and CCR. Hamilton's textbook (Hamilton, James D. (1994). Time Series Analysis, Princeton: Princeton University Press) has conditions for the DOLS and FMOLS form of the estimator (Chapter 19. Cointegration), but I think it's kind of hard to see the required conditions in his discussion. If I recall correctly, the Park (1992) CCR article (Park, Joon Y. (1992). "Canonical Cointegrating Regressions," Econometrica, 60, 119-143) has results for that estimator.

[ryans]

ok thanks. Are the t-stats given in the FMOLS regression output in Eviews FM t-stats?

[EViews Glenn]

Yes. As described in the EViews manual, in FMOLS and CCR we scale the inverse moment matrix by the estimate of the long run varinace of u1t conditional on u2t. (In the dialog, we call this "rescaled OLS"). When estimating DOLS, we use this setting as the default, but support standard OLS, White, and HAC estimators. These latter estimators may be useful for performing various comparisons.

Note that the EViews manual is pretty good in discussing this particular part of the computation.