Models & methods

Foundation model

A forecasting foundation model is pretrained on many time series, then used zero-shot or fine-tuned on yours. The bet is that other series taught it shapes that transfer — season, spikes, slow trends — so you are not starting from random weights.

They shine when you have many related series or short history. They are not exempt from a backtest. A pretrained prior can be confidently wrong on a niche promo calendar.

Fine-tuning is extra fit on your workspace. Do it when the zero-shot path plateaus and you can afford the job. Do not fine-tune as a ritual.

Foundation does not mean 'no features.' Covariates you will have at inference still help. What you skip is weeks of architecture shopping for the first path.