EViews.com

The Eviews manual says that "This model takes the cointegration form of the simple ARDL model and adapts it for a panel setting by allowing
the intercepts, short-run coefficients and cointegrating terms to differ across cross-sections". It sounds like the intercepts here refers to the long-run intercepts, and in the dropdown trend sepcification menu it says "constant(level)", but - here is the odd thing - constant and trend appear in the short-run equation in the estimation results. How should I interpret this???

Furthermore, when translating the long-run trend specifications (restricted constant, constant and restricted trend) from a VECM/ARDL setting, what are the equivalent cases/specifications used in panel ARDL/PMG?

Best,
Lars

Hi Lars,

The statement from the manual is indeed correct. In fact, this is the premise of the Pesaran, Shin, Smith (1999) PMG paper. However, at the moment, the intercepts do NOT refer to the long-run intercepts. That is, the intercept is NOT part of the cointegrating equation. This is, at the moment, a limitation which we have decided to lift in the next major EViews release, namely EViews 13. In other words, users will be able to choose whether they wish the intercept to be considered as a short-run variable (this is the current "constant (level)") or as part of the cointegrating equation.

As far as a trend appearing in the short-run equation estimates, I can't replicate this behaviour. Choosing "constant (level)" should produce cross-sectional regressions with only a constant. If you're certain this is happening, please attach a workfile with this behaviour so that we can investigate further.

Finally, as far as the VECM/(univariate) ARDL trend specifications are concerned, the "Constant" and "Trend" options in univariate ARDL (VECM) correspond, respectively, to the "Constant (Level)" and "Linear Trend" options in PMG (panel). As I mentioned earlier, with EViews 13, the same set of options that are currently available for univariate ARDL will also be available in PMG estimation (panel).

I hope this clarifies things.

Thank you for an informative response! So, not including a constant refers to "restricted constant" in a regular ARDL(VECM) setting?