EViews.com

This thread is about the Smooth Transition Regression (STR) add-in which performs testing, estimation and evaluation of STR models. The testing procedures includes the specific alternatives: LSTR, ESTR, LSTR-D and ESTR-D, transition variables significance and adequacy between STR structures. For the estimation of the STR models the equation or Logl objects are used. The evaluation consists on the usefulness of the transition function in the model, their dynamic properties and diagnostic tests of residuals. The add-in is based on Clive WJ Granger and Timo Terasvirta. Modelling non-linear economic relationships. Oxford University Press, 1993. Comments and suggestions of the add-in are welcomed.

Attached you can find the add-in documentation.

Regards, Nicolas Ronderos Pulido

This is a useful add-in. Thanks for the effort and sharing.

hi,

thanks for this useful add in.
could you provide some examples for use it? I was using JMulTI to estimate STR but it can't forecast, I suppose yours could do it. but it doesn't work to estimate by myself.

Do you check the add-in documentation? I provide two examples in there, one with simulated data and the other with USA inflation. The estimated model with the add-in can forecast if you choose LSTAR or ESTAR models, but if you choose LSTAR-D or ESTAR-D model it can not given that you need the transition variable (residuals) in more advanced periods. Remember that forecasting from nonlinear models for more than T+1 has some problems.

PD: note that the estimated model contain a dep variable as you dependent variable in the estimated models, you could create it as genr dep=%yourvariablename

Regards.

Hi!

First of all, thank you very much for this useful add-in!
I'm working with my post-graduate thesis in economics and I'm estimating an error correction model using an smooth transition approach (STECM).
The method I should use is pretty similar to the one that is used in Delatte's et. al. paper (only without the panel-approach):
http://papers.ssrn.com/sol3/papers.cfm? ... id=1917219

My question: is there a way I could include the short run effects of the ECM-model when using your add-in? This short run component is the one that is last term of the right hand side of the model, so the smooth transition function is not in direct relation with it. However, in order to estimate the model properly I should still include the short run part in it.

Thank you so much in advance!
-Petteri

The paper seems interesting. I don´t know what you mean exactly by the short-run effects, but you can estimate an Error Correction Model using the Engle-Granger procedure (from what I saw you have two variables CDS and bonds) and save the stationary error correction mechanism, then using the add-in twice estimate both equations including this mechanism. What is normally found is that in some periods the cointegration mechanism stops working, the transition variable will tell you this.

Good luck!

hello;
thank you for your program for estimating STR models. but I have a question about running it. would you please guide me what do you exactly mean about "vector object for starting value"? it is not clear for me and I don not know what i must insert in it? can you help me where I can find any description about your program?
thanks again for providing this program.

Hi,

STR models are nonlinear in parameters so to estimate them one must use an optimization algorithm, the vector object for starting values is a vector object that contains an starting value for each of the coefficients in the model respectively i.e. the position (1,1) in the vector corresponds to the starting value for the first coefficient and so on. The vector object must be at least of the same size than the number of parameters you want to estimate.

Check the examples in the add-in documentation, it will give you a better idea .

Regards.

Hi;
I am using eviews 8 to running STR add-in, but I can not find any add-in documentation and your examples.
Now I am trying to insert an item in " vector object for starting value", I have 5 coefficient to be estimated, can you please exactly describe for me what I write there for starting value?
Thanks

If you click Add-ins->Manage add-ins-> select the star add-in and then Docs the add-in documentation will appear (now is attached in the first comment just in case). You have to create a vector of size 5 or greater.

In this case all the starting values will equal one, if you want to change any then just fill the starting value in the corresponding coordinate.

Hi;
Thank you so much for your helping, and please excuse me for interrupting you. But when I running the program, I face with this massage error:
" Syntax error in control statement in " If 2*@wcount("inf(-1) ex y pim")+1>@ROWS(VECTOR(10) STARTING_VALUE=1) THEN"."
I do not understand what is the problem!!
I will be appreciate you if explain this massage error for me.
Best Regard

In the edit box you only enter the name of the vector object i.e. starting_values

Hi,

Thank you for this useful add-in. I've trying to use it with my data, however I get the following message WARNING: Singular covariance - coefficients are not unique and in the estimation model, the coefficients appear but I only get NA for the t-stat and prob.

Could you help me ? Thank you !!

Souad

Check the appendix c of user guide II, estimation and solution options.

Regards.

Dear Nicolas, thanks for the STAR* add ins! I have installed the add in and tried to use it. I read the PDF help file you wrote but still not clear how to implement. For example, how do you input the commend for example 1 in the PDF file? If I input the equations, variables in the STAR window, what are the definitions of all the fields in the "STAR models" window? Many thanks and I look forward to hearing from you! /Louisa