https://github.com/nixtla logo
#neural-forecast
Title
# neural-forecast
r

Rachel Yee

04/27/2023, 1:27 PM
Hello, got a quick question to ask. I am trying to use exogenous variables in NHITS but am not clear on the definition of historic vs. Future. I'm doing a daily fuel sales forecast, with price of our own fuel and competitors price as exogenous variables. Both own prices and competitors' prices are only known on the day of the sales itself. Are they considered historic or future? Also I'm not quite sure what future and historic variables mean from the electricity price example. Is there another example to explain what is considered historic/future?
c

Cristian (Nixtla)

04/27/2023, 1:54 PM
Hi @Rachel Yee! In time series forecasting it is crucial to avoid information leakage, that is, when doing a forecast we can only use the available information at that time. We have the historic/future variables division to allow for different types of covariates, depending if we know their future values or not.
For instance, if today (04/27) you are doing the daily fuel sales forecast for tomorrow (04/28), and you know the prices for tomorrow, then they are future exogenous variables. The models can use the tomorrow's price to produce the forecast. If you only have past values of the prices (up to 4/27), then they should be historic variables
r

Rachel Yee

04/27/2023, 1:59 PM
Ah I see! Thank you!