Smooth Transition Regression - transition with time

[hyo]

Hi everyone,

I am looking for help on how to estimate a Smooth Transition Regression for a monetary policy rule. In the setup there are two different regimes, and the policy rules in any one period is a weighted average of those two rules. I am looking for just one switch in the model.

The model is as follows:

Code: Select all

i = (1-w)*(c(1) + c(2)*inflation + c(3)*outputgap) + w*(c(4) + c(5)*inflation + c(6)*outputgap)) with w = (1 + exp(-c(7)*(t - c(8))))^-1 0 ≤ w ≤ 1

The t stands for a time trend. Now one major problem I have is, I don't know how to model this time-trend in eviews so that c(8) will give me the switching date (or observation).

Also I am not sure how to estimate this in eviews in general. I tried following the advice from this topic, http://forums.eviews.com/viewtopic.php?f=4&t=3712 , and with exchanging t with a lagged observation of i (=interest rate), but I only received non-significant and nonsensical coefficients.

Any help would be appreciated.

[EViews Glenn]

Can you provide more detail on what you tried to estimate and what happened? Perhaps you could post your workfile.

For a transition with time, you can just put a @trend in place of the t in your W so that c(8) will give you the threshold in terms of an offset from the start of the workfile.

[hyo]

Thanks for the quick reply.

I am trying to estimate a reaction function for the ECB and want to find out if a structural change occurred with this monetary policy rule.
So the result should be two different functions, one explaining the ECB rate-setting before and one after the change (~ the financial crisis in 2008).
The reaction function states simply that the nominal interest rate (that the ECB sets) can be explained by the current inflation and outputgap.
interest = c1 + c2*inflation + c3*outputgap


Through the STAR model I am trying to find the date of the switch between two of those reaction functions.

I attached my workfiles: ecb_star, where I just estimated the whole equation with a least squares approach, and ecb_star_logl, where I tried to run a program code (which I got from the forum post http://forums.eviews.com/viewtopic.php?f=4&t=3712 )
I am new to EViews so I still need to learn a lot about this program.

[EViews Rebecca]

Your program code looks fine. What is your question?

[hyo]

The resulting coefficients don't make any sense in the model framework and the p-values are not-significant.
For the approach via least-squares I don't get any numbers for the t-statistic and p-values.
I know the model can be estimated since I am following another paper on this issue (http://link.springer.com/article/10.100 ... 013-0713-6).

Any advice to why i get strange results?

[EViews Rebecca]

These models are very sensitive to starting values. Try playing around with those.

[hyo]

By that you mean I should give the coefficient-vector C different values before the estimation?

Thank you for your help again.

[EViews Rebecca]

They get set in your logl program.

[hifinecon]

Hi, I am an Eviews user for many years but only for basics. Now I want to estimate following equations:

Logistic smooth transition regression model:
Y(t)= b(0) + b(1)Y(t-1) + b(2)X(1t) + b(3)X(2t) + b(4)X(3t) + θ(b(0) + b(1)Y(t-1) + b(2)X(1t) + b(3)X(2t) + b(4)X(3t)) + ε(t) (1)
Θ = {1 + exp[-γ(Y(t-1) – c)]}^(-1), γ≥0

Exponential smooth transition regression model:
Y(t)= b(0) + b(1)Y(t-1) + b(2)X(1t) + b(3)X(2t) + b(4)X(3t) + θ(b(0) + b(1)Y(t-1) + b(2)X(1t) + b(3)X(2t) + b(4)X(3t)) + ε(t)
Θ = 1 - exp[-γ(Y(t-1) – c) ^(2)] , γ≥0

Because I am not an econometrician, I just use Quick/Estimate Equation and select THRESHOLD - Threshold Regression from the main Method dropdown menu near the bottom of the dialog, and click on the Smooth radio button in the Threshold type setting. Can I just write the equation (1) for the dependent variable and the threshold varying regressors In the Equation specification section? And leave nothing in the second edit field. are there any other details have to adjust in the option under estimate equation? I assume that I could choose either logistic or exponential in the option of estimate equation.

Sorry about long question since I am not an econometrics guy.

[EViews Rebecca]

See the documentation on this subject at:

http://www.eviews.com/help/helpintro.ht ... Vi.html%23