out-of-sample forecasts with state-space
Posted: Thu Aug 05, 2010 1:58 pm
Hi there,
I have a state-space model estimated from 1981Q1 to 2007Q4. The model converges and the coefficients are as expected.
Now I would like to perform forecasts until 2011Q4. I’ve tried the built-in forecast available (dynamic, dynamic & smoothed, etc) but I am not sure of how these forecasts are implemented and thus I am not sure whether the results is what I want.
My question is:
Is it possible to implement a “rolling regression” while estimating the state-space and generate 1-step-ahead “out-of-sample” forecasts? For instance, if I was to forecast values from 2008Q1 to 2011Q11:
1st estimation sample: 1981Q1 – 2007Q4 forecasts 2008Q1
2nd estimation sample: 1981Q2 – 2008Q1 forecasts 2008Q2
last estimation sample: 1984Q4 – 2011Q3 forecasts 2011Q4
Are any of the built-in forecasting options doing this implicitly?
Thanks for your help,
Mara
I have a state-space model estimated from 1981Q1 to 2007Q4. The model converges and the coefficients are as expected.
Now I would like to perform forecasts until 2011Q4. I’ve tried the built-in forecast available (dynamic, dynamic & smoothed, etc) but I am not sure of how these forecasts are implemented and thus I am not sure whether the results is what I want.
My question is:
Is it possible to implement a “rolling regression” while estimating the state-space and generate 1-step-ahead “out-of-sample” forecasts? For instance, if I was to forecast values from 2008Q1 to 2011Q11:
1st estimation sample: 1981Q1 – 2007Q4 forecasts 2008Q1
2nd estimation sample: 1981Q2 – 2008Q1 forecasts 2008Q2
last estimation sample: 1984Q4 – 2011Q3 forecasts 2011Q4
Are any of the built-in forecasting options doing this implicitly?
Thanks for your help,
Mara