Holt-Winters Forecasting Online: Triple Exponential Smoothing
What is Holt-Winters?
Holt-Winters (triple exponential smoothing) models level, trend and seasonality simultaneously, making it a classic method for series with both trend and seasonal patterns. It updates three smoothing equations for level, trend and seasonal components, controlled by smoothing parameters α, β and γ, where values closer to 1 respond faster to new data. It is simple, suited to online updating and a reliable choice for short-term forecasting once the parameters are optimized automatically.
When to Use It
Use it for demand and sales forecasting, inventory planning, and establishing a seasonal baseline for anomaly detection. It suits series with a clear trend and stable seasonal pattern; for series with structural breaks, consider intervention analysis or ARIMA. Compare candidate models with RMSE and MAPE, and prefer the model with the lower MAPE for cross-series comparison.
How to Use It (Step by Step)
Paste data covering at least two to three full seasonal cycles, specify the season length (for example 12 months or 7 days) and the forecast horizon, and choose the additive or multiplicative model. The tool optimizes the smoothing parameters, then outputs the fitted curve, forecast values and forecast intervals. Forecast intervals widen naturally as the horizon extends.
Key Formulas / Example
The additive model assumes y = level + trend + seasonal, while the multiplicative model assumes y = (level + trend) × seasonal. The additive form suits a stable seasonal amplitude; the multiplicative form suits series where the seasonal amplitude grows with the level, such as sales whose seasonal peaks scale with growth. Parameters α, β, γ are optimized in [0, 1] by minimizing fit error such as SSE.