Hansen-Seo(2002), Threshold cointegration

[TomDoan]

This is a replication file for Hansen and Seo(2002), "Testing for two-regime threshold cointegration in vector error-correction models", J of Econometrics, vol 110, pp 293-318. Compared with Balke-Fomby(1997) this doesn't take the cointegrating vector as known and unlike Enders-Siklos(2001), it estimates the cointegrating vector and threshold together, rather than using a two-step process. That makes for a rather complicated estimation process, requiring a double grid search (over beta's, then, given beta, over threshold values). The likelihood function is highly discontinuous, so slightly different grids can produce quite different results.

Updated 28 October 2011 to add an option from graphing the beta vs log det sigma and to correct an error in the grid search for beta.

hansenseo.rpf

Program file

(13.94 KiB) Downloaded 1840 times

zeroyld.dat

Data file

(241 KiB) Downloaded 1676 times

[mskare69]

Hello,

How can I modify Hansen and Soe 2002 rats code to run three regime threshold cointegration model (if two thresholds are identified) and perform impulse response analysis in such three regime model?

Any help is much appreciated.

Fondly

Marinko

[mskare69]

I forgot to ask how to test If TVECM3 is more appropriate then TVECM2?

Thanks

[TomDoan]

Hello,

How can I modify Hansen and Soe 2002 rats code to run three regime threshold cointegration model (if two thresholds are identified) and perform impulse response analysis in such three regime model?

Any help is much appreciated.

Fondly

Marinko

That would require a third loop over a second value of "gamma" inside the existing ones. However, I wouldn't recommend that. Even the one break Hansen-Seo doesn't work very well, and I can't even guess how badly behaved a double break model would be.

[mskare69]

Thanks for a quick response. Let me rephrase the question. Is there a completed code in Rats that I can use to run TVECM with two thresholds (three regimes). Can I use some of the codes (5.1, 5.2, 5.3) from the last Estima course study to run the whole thing - testing for threshold cointegration, estimating TVECM and doing impulse response of the model.

Fondly,

Marinko

[TomDoan]

Thanks for a quick response. Let me rephrase the question. Is there a completed code in Rats that I can use to run TVECM with two thresholds (three regimes). Can I use some of the codes (5.1, 5.2, 5.3) from the last Estima course study to run the whole thing - testing for threshold cointegration, estimating TVECM and doing impulse response of the model.

Fondly,

Marinko

The Tsay example (SMS_5_3.RPF) estimates two and three regime models. It's not strictly for a VECM, but there's no real difference between a VECM with the cointegrating vector treated as known and what that's doing. The difficulty with the Hansen-Seo approach is that the likelihood when the cointegrating relation is also unknown is very, very badly behaved.

[ege_man]

Dear Tom
Is it possible to modify this code to estimate cointegration in a five-variable model?
Thanks

[TomDoan]

Dear Tom
Is it possible to modify this code to estimate cointegration in a five-variable model?
Thanks

No. That would require that the one beta loop be replaced with four. Even if that were feasible (it would take too long), there is no way that the results would be usable. As I mentioned in a response to an earlier question, this really is a badly behaved model even with two variables and two regimes.