TAR model

[hsmellin]

Hi,

I am looking to estimate a threshold autoregression (TAR) model. I was wondering how one would create an indicator function such that if the value of the transition variable is less than or equal the threshold, the indicator is equal to one (and zero otherwise)

thanks,

M

[EViews Glenn]

http://www.eviews.com/Learning/dummies.html

[hsmellin]

Thanks Glenn.

This was a useful link. However, i wish the threshold to be determined by the data. I have tried using the @inlist function yet this does not report the threshold parameter. So then i have also try the @between function which reports a parameter estimate of zero so again i feel this may not be correct either since i have tried it with two different threshold/transition variables. So i assume what i am trying to do requires something more sophisticated?

thank,
M

[EViews Glenn]

I don't know what you mean by "determined by the data". If you have a threshold variable and a threshold then the simple comparison:

Code: Select all

series thresh = y(-2)>37

produces the corresponding dummy. For estimation of the threshold as part of your procedure you'll have to move the "37" around as needed, but the principle is the same.

[trubador]

I think s/he is asking something along the lines of Chan (1993)*, where the threshold is determined searching over the data itself through minimizing the sum of squared errors from the fitted model. So yes, it requires a more sophisticated but a doable analysis. Those dummies will still be of use, when programming such a model.

*Chan, K. S. (1993), "Consistency and Limiting Distribution of the Least Squares Estimator of a Threshold Autoregressive Model," The Annals of Statistics, 21, 520-533.

[EViews Glenn]

Yes, I figured that the estimation was part of the procedure, hence my comment that if the threshold finding was part of the estimation, he/she would have to move the threshold value around as required. I was merely pointing out the construction of the dummy variables used in forming the objective was straightforward.

[hsmellin]

Thanks for the inputs Glenn and Trubador, also thank you for the reference.

i will take this to my data and program accordingly.

Mike