EViews.com

Hi Louisa,

For the example 1, after installing the add-in just entee in the command window:
¿Which version of Eviews are you using? ¿Of which fields to fill do you have questions? all of them are explained if the pdf.

hi
thank you so much for your useful add-ins.
i have some question about running it: 1.while writting linear variables in the program box, and estimating the model, why is omitted the two first linear variables?
2.how can i solve the over flow..... error?
best regard

Masume:

What do you mean by ommitted?, the first two linear variables are the dependent variable and the constant. ¿What did you tried to do to get the over flow error?, please post your workfile.

Hi Nicolas
i don't have two dependent variables. my dependent variable is [inf] and independant variables are [inf(-1), y, pim, oil, neer], and my transition variable is [neer(-1)].
can you please guide me that how can i estimate my model? does it have a special arrangement for writting the variables in star box?
i sent you a file. please check it and tell me what is wrong?
i sent you the work file too.
you know, i estimate the model in star program not in eviews command line. is it true?
about the over flow error: when i add an additional variable, this error occures.
best regards

The specification you posted does not run because you only have 35 observations, since the tests are based on expansions that use variables combination and exponentials, like the RESET test, the testing procedure cannot be performed, you can bring out some variables, nevertheless I recommend you to increase the frequency of the data for two reasons, be able to run the tests and increase the possibility of reject linearity, there is evidence that the temporal aggregation hides the nonlinear component of a series.

Hi all, I encountered this problem while running my model: std error, t-statistic and Prob. returned NA and I really don't understand why. Here I attach my result screenshot and data workfile in case you can help me check. Thanks in advance.

You probably have bad starting values in the C vector.

Thanks for the great Add-In. However, I am not clear on whether this add-in can handle data with multiple regimes (3 or more regimes)? In the add-in documentation the provided examples do not show the threshold specification summary.

So does this add-in estimate the appropriate number of thresholds and the value of each threshold?

Hi,

In STAR models the transition function determines the regime. Therefore with a STAR model you estimate as many regimes as values that your transition function takes. The threshold "c", the "Gamma" (In the add-in documentation notation) and the transition variable determines the behavior of your transition function. If Gamma is large enough then you will have a TAR model with two regimes in the LSTAR context, otherwise you will have many regimes. The following code could help.
Best regards,

Hi,

In STAR models the transition function determines the regime. Therefore with a STAR model you estimate as many regimes as values that your transition function takes. The threshold "c" and the "Gamma" (In the add-in documentation notation) determines the behavior of your transition function. If Gamma is large enough then you will have a TAR model with two regimes in the LSTAR context, otherwise you will have many regimes. The following code could help.

Code: Select all

wfcreate u 1 1000 genr x=@trend/100 for !gamma=0 to 20 step 0.1 'Smooth regimes to quick regimes genr y=@logit(!gamma*(x-2.5)) y.line statusline !gamma next

Best regards,

Hi Nicolas,

Thank you so much for the kind and helpful reply! The above code is indeed helpful and most of all, helps me understand that STAR models can have many regimes and the number of regimes is dependent upon the number of different values taken by the transition function. Thanks!

Hi,

In STAR models the transition function determines the regime. Therefore with a STAR model you estimate as many regimes as values that your transition function takes. The threshold "c", the "Gamma" (In the add-in documentation notation) and the transition variable determines the behavior of your transition function. If Gamma is large enough then you will have a TAR model with two regimes in the LSTAR context, otherwise you will have many regimes. The following code could help.

Code: Select all

wfcreate u 1 1000 genr x=@trend/100 for !gamma=0 to 20 step 0.1 'Smooth regimes to quick regimes genr y=@logit(!gamma*(x-2.5)) y.line statusline !gamma next

Best regards,

Dear NicolasR,

I hoping you can be of some additional help. I have estimated a STR model that has resulted in a large estimate of gamma. Does this indicate that quick regimes such as TAR might be a better fit?

Maybe. To check how your transition function behaves you can select the option "Evaluation of the transition function" and this will give you the values that your transition function takes. If the function only takes two values then you will have a TAR model. Do you want the model to calculate forecasts? or for what purpose?

Best regards,

Dear NicolasR,

Thank you for your response above. I just saw it there, as I am posting another issue.

I am having a problem with estimation. I have tried changing my starting values and I can't seem to resolve the problem here. Here is an upload of my output. But what is strange is that the transition variable I am using has some very strong results in the tests for significance. Also it seems I am correct in estimating an LSTR, as indicated by the structure tests.

However there is clearly something wrong with my standard errors, t-stats and p-values. Do you have any idea how I might resolve this issue?

Maybe. To check how your transition function behaves you can select the option "Evaluation of the transition function" and this will give you the values that your transition function takes. If the function only takes two values then you will have a TAR model. Do you want the model to calculate forecasts? or for what purpose?

Best regards,

It does appear to take on more than two values. Primarily I am not interested in forecasting, but rather in explaining observed patterns in two time series. For example I have two time series which have exhibited a coupling behavior over some time, meaning they nearly converged in value. My theory is that this coupling pattern was triggered by an exogenous observed variable crossing some threshold. So I guess my interests lie in model selection primarily, because I'd like to argue that some particular transition/threshold variable is the most significant variable in determining the regime change.

I looks like that the optimization algorithm does not reach convergence. You can try different strarting values. When I have no idea what values, what I usually do is to try random numbers for the starting values and I keep the ones that reach the best values of some criteria (BIC, AIC, R2,SSR). If this does not work you can change the convergence parameter of the optimization algorithm, the default can be changed in Options->General Options->Estimation Options.

For what I can see the model does not differ to much from a TAR. You can graph the transition function and the two time series over the same axis to see if the transition variable explains the divergence between the series.