Hi,

I'd like to do out-of-sample prediction after estimating a threshold model. I follow the examples in the manuals:
tsthresholdpostestimation.pdf (stata.com)

I tried (1) one-step ahead prediction and (2) dynamic predictions. The results are the same. Correct me if I am wrong -- I feel that these are actually in-sample fit for the model. What I want is the true "out-of-sample" forecasting. In the above example, all I want to do is
a. to estimate the model from 1950Q1 to 2002Q4, and predict 2003Q1
b. to estimate the model from 1950Q1 (recursive) or from 1950Q2 (rolling) to 2003Q1, and predict 2003Q2 .....
....

How can I do it?

What is the difference between dynamic forecasting here and in-sample fit?

Thanks,

J