What regressor should I use? Lagged or Future?

[kkckk1110]

Hello, I found neural prophet quite useful in my task and would like to thank the authors. However, I have a question about the selection of lagged and future regressors. In my task, when I forecast y at time t, it is expected that the regressors at time t, t-1, t-2..., t-p, are included. But the future information of the regressor is unknown. I read the tutorial and noticed that, if I treat it as a lagged regressor, I can only access to its value up to t-1. So I am confused that, what regressor should I use in this case? I am looking forward to your reply. Thanks a lot.

[quant-exchange]

@kkckk1110 the way I've used it is for forecasting a future 1 hour value only*. With that said, I've chosen 'lagged' regressor/s: as I 'know' the actual values I want to pass as an argument are relevant on quantifiable current data. However, you may find that you can 'forecast' future values you can use as input for your 'future' regressors; making them known.

I've not implemented a setup around this here, but this is how I've thought about using it*.