Holt-Winters Forecast Explorer

Tune α, β, γ, φ live — watch how the model handles trend, seasonality, and structural breaks across four supply chain datasets

Dataset
Strong trend + multiplicative seasonality.
Model Structure
Smoothing Parameters
How fast the level adapts
How fast the slope adapts
How fast seasonal indices update
0.80 = heavy, 0.99 ≈ linear
Forecast Horizon
Actions
—
Training RMSE
In-sample fit quality
—
Holdout RMSE
Last 20% of data
—
Model (ETS notation)
Error, Trend, Seasonal
Forecast
Historical, fitted (training), forecast with 80% / 95% prediction intervals. Red line = train / holdout split.
Historical Fitted Forecast 80% PI 95% PI
Decomposed Components
Estimated level ℓt, trend bt, and seasonal st over the training portion.
Level ℓt
Trend bt
Seasonal st
Residuals
Observed minus fitted (training). Should look like noise around zero — patterns indicate model misfit.
Model Interpretation
Implements Holt-Winters (additive and multiplicative) with optional damping (Gardner & McKenzie 1985) following Hyndman & Athanasopoulos (2021), Forecasting: Principles and Practice, 3rd ed., Ch. 8. "Let ETS pick" minimizes in-sample SSE via grid search on the first 80% of the series. Prediction intervals use the σ√h approximation. All calculations run live in your browser.