Pedroni, cointegration, heterogeneous cross-sections

[tvonbrasch]

Hi,

the Pedroni test (1999) allows for beta to vary across units, but as you write in the manual "Notice that the cointegrating relationship between y and x is assumed to be homogeneous
across cross-sections" (p.797). can you change Eviews to also allow for the cointegrating relationship between y and x to be heterogeneous ? maybe you can make this option available as an addin? it would be highly appreciated

Thomas

(also, when writing about the Pedroni test for cointegration in the manual you state that "In this case, eight of the eleven statistics do not reject the null hypothesis of no
cointegration at the conventional size of 0.05." (p. 843). But on the picture I see only seven statistics, do you mean five of the seven, or have i misunderstood?)

[EViews Glenn]

The Pedroni (1999) heterogeneous slopes are for the test, but not the FMOLS estimator. The Pedroni FMOLS estimator explicitly restricts B to be common.

I haven't looked at this stuff in a while, but my recollection is that under heterogeneity, the Phillips and Moon results can be used to show that the various estimators provide consistent estimates of the long-run average coefficient. If there is concern about heterogeneity, one case use the grouped mean estimators to estimate the average of the cointegrating vectors which will be robust to that form of heterogeneity. So there's nothing really to add to EViews. See Pedroni (2000) Fully Modified OLS for Heterogeneous Cointegrated Panels, for details.

There are 11 statistics, 8 pooled weighted and unweighted, and 3 group.