ENDERSSIKLOS—Asymmetric error correction

[TomDoan]

@EndersSiklos does various types of the unit root regressions with threshold breaks on the residuals from an Engle-Granger cointegrating regression (assumed to be done already---input are the residuals). A related procedure for threshold unit roots is @EndersGranger.

Detailed description

For an example, see Enders and Siklos(2001), "Cointegration and Threshold Adjustment," JBES, vol. 19, no. 2, 166-76.

[challenges]

Dear Tom,

Could you please tell me whether in the EndersSiklos procedure, the unit root tests may be done on the residuals stemming from a long-run relationship that would have been previously estimated with DOLS, rather than OLS?
Thank you a lot for your reply.

[TomDoan]

I don't see any reason why they couldn't.

[challenges]

I don't see any reason why they couldn't.

Dear Tom,
Thank you for your quick reply. Would it then be correct to use the critical values of the paper of Enders and Siklos (2001) for the unit root tests performed on the residuals stemming from the long-run relationship that would have been previously estimated with DOLS?

Moreover, could you please tell me if I should change the ECM representation in order to remain consistant with the use of the DOLS estimator in the long-run relationship, or may the ECM remain the same as in the paper of Enders and Siklos (2001)?

Thank you for your help.

[TomDoan]

I don't see any reason why they couldn't.

Dear Tom,
Thank you for your quick reply. Would it then be correct to use the critical values of the paper of Enders and Siklos (2001) for the unit root tests performed on the residuals stemming from the long-run relationship that would have been previously estimated with DOLS?

Moreover, could you please tell me if I should change the ECM representation in order to remain consistant with the use of the DOLS estimator in the long-run relationship, or may the ECM remain the same as in the paper of Enders and Siklos (2001)?

Thank you for your help.

I'm not sure whether the critical values would change. The critical values are generated by Monte Carlo using the prescribed method. While one would suspect that the difference between methods of estimating the cointegrating vector would be minor compared to the sources of randomness, it would probably be necessary to redo the simulations to tell for sure.