Discussion

Hi everyone,

I’m trying to understand intuitively when it makes sense to use a time series model like SARIMAX versus a simpler approach like linear regression, especially in cases of weak autocorrelation.

For example, in wind power generation forecasting, energy output mainly depends on wind speed and direction. The past energy output (e.g., 30 minutes ago) has little direct influence. While autocorrelation might appear high, it’s largely driven by the inputs, if it’s windy now, it was probably windy 30 minutes ago.

So my question is: how can you tell, just by looking at a “forecasting” problem, whether a time series model is necessary, or if a regression on relevant predictors is sufficient?

From what I've seen online the common consensus is to try everything and go with what works best.

Thanks :)

Upvote37Downvote32Go to commentsShare

OpenAI• Official•Promoted

Turn ideas into images with ChatGPT. Describe anything - like “a wizard reading in a moonlit forest” - and watch it appear. Try it out now ??