Hi everyone - I'm currently evaluating timeGPT for some use cases. I'd have a general question with regard to exogenous variables. With other models, like for neural-forecast there is the disctinction between historic and future exogenous variables. The values of historic exogenous variables are "not known" / cannot (reliably) be predicted for the future, but their historic values can in some cases provide significant information for the prediction. For timeGPT I "only" found the possibility for the inclusion of future exogenous variables, which I have historic information on and which I know / can predict for the predicted horizion (like weekdays as an easy example). Is there a way to incorporate historic exogenous variables with timeGPT - maybe I just overlooked some aspects? Would appreciate any help - thanks a lot in advance

Hi @GFa for TimeGPT you do need the future values of the exogenous variables. It's not possible to just use their historical values. I understand that they're hard to forecast (and you're adding more uncertainty to your problem), but you could try forecasting them with TimeGPT and then use that to forecast your target variable.

Thanks a lot!