← back to tools

Sales Forecast Lab

Holt-Winters triple exponential smoothing on a simulated weekly retail sales series. Drag the level (α), trend (β), and seasonal (γ) knobs to see the model re-fit and forecast the next 12 weeks live. This is a stripped-down cousin of the cluster-specific forecasters I ship for retail clients.

Level smoothing α 0.45

Trend smoothing β 0.10

Seasonal smoothing γ 0.30

Seasonal period (weeks) 12

Forecast horizon 12

Noise level 0.08

MAPE

—

RMSE

—

Actual sales Fitted Forecast

Under the hood: additive Holt-Winters. Level ℓ, trend b, and season s each update from the previous step with their smoothing parameter. Forecast for horizon h is ℓ + h·b + s[t−m+h mod m]. MAPE and RMSE are computed on the in-sample fit. Try α≈0.5, β≈0.1, γ≈0.3 as a sensible starting point — then push γ up to see it over-fit the seasonality.