The Holt Winters Method
Holt Winters is exponential smoothing extended to capture both trend and seasonality at once. It is sometimes called triple exponential smoothing.
Three smoothed states
The method maintains three evolving quantities, each with its own smoothing factor.
- Level: the current baseline value, smoothed with alpha.
- Trend: the current slope, smoothed with beta.
- Seasonal: a set of seasonal offsets, one per position in the cycle, smoothed with gamma.
A forecast adds the level, the projected trend, and the matching seasonal component for the target step.
Additive and multiplicative seasons
- Use the additive form when the seasonal swing stays a constant size.
- Use the multiplicative form when the swing grows with the level of the series.
Why it is useful
Holt Winters updates cheaply as new data arrives and handles the most common business patterns of a rising trend with a steady cycle. It is a strong baseline that often rivals heavier models on clean seasonal data. Its weakness is that it needs a known period and assumes a fairly regular season.
Key idea
Holt Winters smooths level, trend, and seasonality with three factors, giving a cheap and strong baseline for regular seasonal series.