Statistical BaselineClassical forecasting

Statistical Baselines

Reference rule-based baselines that every foundation model should beat. They have no learned parameters; they exist on the leaderboard so that absolute scores have an interpretable floor — if a TSFM cannot outperform Seasonal Naive, something is wrong.

Classical baselines sit on the leaderboard as honest reference points. They have no learned parameters — every forecast is a closed-form function of the most recent observations — so they impose a hard floor for absolute scores. A foundation model that loses to Seasonal Naive on a given challenge is almost certainly mis-configured for that series' frequency or horizon, or is suffering from a data issue rather than a modelling one.

Four baselines are evaluated here: Naive (carry the last value), Seasonal Naive (carry the value from the previous corresponding season), Simple Moving Average (mean of the last N observations), and Seasonal Average (mean of past seasonal periods). Seasonal Naive in particular is famously hard to beat on strongly periodic series like electricity load.

Versions on TS-Arena

Each version below corresponds to one registered model id in the leaderboard. Click through to its detail page for per-model rankings, forecasts, and history.

  • Naive
    naive

    Repeats the last observed value for the full horizon.

  • Seasonal Naive
    seasonal-naive

    Repeats the value from the most recent corresponding seasonal period. Surprisingly hard to beat.

  • Simple Moving Average
    simple-moving-average

    Mean of the last N observations, projected forward.

  • Seasonal Average
    seasonal-average

    Mean across past seasonal periods, projected forward.