Forecasting

[nflp]

Hi Gregor, many thanks for an amazing package and all the great work!

Is there a way to use the models we estimate for out-of-sample forecasting (maybe some of the underlying filtering packages support that)? I haven't been able to figure anything out.

[gboehl]

Hi, thanks for your interest in the package!

Yes, this is a quite straightforward application similar to the counterfactual analisis here. You take the last state from the smoothed series, and then draw shocks from the estimated shock distribution to simulate the model forwards.

I hope this helps.

[nflp]

Thanks!
Not sure how in practice I would implement that... Starting with the counterfactual example, I would modify the epd object to also include n-periods ahead that I want to forecast (basically, append zeroes to epd['resid]), and extend the mask object accordingly. Does this sound right?

[gboehl]

Yes, sounds very right. This should be it.

[nflp]

Thanks! I made a try here in case it's interesting for you to take a look https://github.com/nflp/dsge_forecasting/blob/main/DSGE%20forecasting.ipynb

[gboehl]

Very nice, this seems to be pretty much how it is done. Thanks for this!

Any particular reason why you needed to install pyyaml manually?