Feature request: Direct forecast

[itamarcaspi]

If I understand correctly, fable (and forecast) only deal with iterated forecasts, i.e., multiperiod-ahead time series forecasts are made using a one-period ahead model. Am I right? If so, would it be possible to add the option for direct forecasts, made using a horizon-specific estimated model, where the dependent variable is the multi-period ahead value being forecasted?

Reference

Marcellino, M., Stock, J. H., & Watson, M. W. (2006). A comparison of direct and iterated multistep AR methods for forecasting macroeconomic time series. Journal of Econometrics, 135(1-2), 499-526.

[mitchelloharawild]

This option would require a model specific implementation, of which none of the currently implemented methods are. However as far as implementation is concerned, there is nothing preventing a newly implemented model from being optimised for a specific horizon.

[edgBR]

Hi @itamarcaspi

For direct forecast there is a nice package available:

https://cran.r-project.org/web/packages/forecastML/vignettes/package_overview.html

However is more focused into transforming the time series problems into a machine learning regression problem with lagged predictors.

[itamarcaspi]

Thanks, @edgBR ! This package seems to be very promising and up-to-date.

[edgBR]

An update maybe useful for the creators of tidyverts:

https://github.com/nredell/forecastML/issues/35

Next release is planned to be as compliant as possible with the fable API.

BR /Edgar

[mitchelloharawild]

Thanks for the mention - useful update indeed.