The RATS Software Forum

Hi all,

Chang and Nguyen (Journal of Econometrics, 2012, vol 167, pp. 504-520) present a panel cointegration test and estimation framework which enables (among other things) to relax the assumption of homogenous coefficients on the stochastic variables across cross-section. That is, their approach allows to "have heterogeneous individual fixed effects, and cointegrating relationships defined with different cointegrating vectors and variables involved".

I wonder whether anyone has programmed (or is planning to program) a RATS code that would allow for testing and estimation in this framework, also allowing for different cointegrating vectors?

Cheers,

Does this differ in any substantial way from Pedroni's work? @PANCOINT allows for heterogeneity in the cointegrating vectors (and the intercepts and short-run dynamics...).

Tom,

I was a bit unclear with my initial message, apologies for that.

The problem is that while the Pedroni method allows for cointegration testing in the presence of heterogenous cointegrating vectors, it does not allow for estimating those (heterogenous) vectors (as far as I have understood). I thought that the Chang-Nguyen approach (Journal of Econometrics, 2012, vol 167, pp. 504-520) allows for that kind of estimation of the vectors as well, but noticed now at another reading that they do not estimate and report the vectors after all - my apologies for that mistake too.

The Pedroni2007 (Journal of Applied Econometrics, 2007) example seems to be helpful in many ways, but even that does not allow for estimating the heterogenous cointegrating vectors (again, as far as I have understood - please correct me if I am wrong ). I wonder whether finding cointegration (only) allowing for heterogeneity in the cointegrating vectors should just be 'handled' by estimating a cointegrating vector for each 'individual' (or 'subgroup' as in Pedroni, 2007) separately?

Anyhow, Chang and Nguyen approach (Journal of Econometrics, 2012, vol 167, pp. 504-520) seems to be more flexible in some other 'directions'.

The two Pedroni "estimation" procedures (@PANELFM and @PANELDOLS) both estimate the heterogeneous C.V.'s and also a grouped common estimator.

Thank you Tom and sorry for the misinformation I had - indeed, panelfm provides the flexibility and information I was looking for.