Holt-Winters Exponential

This forecasting method is also an extension of Double Exponential Smoothing (see Holt Linear) that adds the ability to capture the seasonality of the time series data. This method is commonly referred to as Triple Exponential Smoothing. In addition to the alpha and beta parameters found in Double Exponential Smoothing, there is an added smoothing seasonal parameter gamma used to handle the influence of the overall seasonality. The method is broken into two variations based on the handling of the parameter gamma:

Seasonal Additive

The seasonal additive model is an extension of the Holt-Winters Exponential Additive Model. The Seasonal Additive model does not contain a trend equation, or parameter beta, and instead only applies the seasonal parameter gamma using the additive method.

NOTE: The Seasonal Additive model is only capable of handling data with a Seasonal component but no Trend Component.

Holt-Winters Exponential Multiplicative

In this variation, the seasonal parameter gamma is expressed in relative terms or as a percentage. In order to account for the seasonality, the trend equation of the series is seasonally adjusted by dividing each value by the seasonal parameter gamma. The multiplicative method is preferred when the seasonal variations are changing disproportionally to the level of the series.

Holt-Winters Damped Method

An extension of the Holt-Winters Exponential (Additive and Multiplicative), the damped method is used for long-term forecasts. For long-term forecasts, a Holts-Winters Exponential model will, by nature, increase or decrease indefinitely into the future. With the addition of a damping parameter between 0 and 1, this long-term increase or decrease can be minimized, and accurate long-term predictions can be maintained. The damping parameter is handled inside of the algorithm and thus is unspecified by a user.

NOTE: Holt-Winters Exponential Models are capable of handling data with both trend and seasonal components.

Seasonal Multiplicative

The seasonal multiplicative model is an extension of the Holt-Winters Exponential Multiplicative Model. The Seasonal Multiplicative model removes trend equation, or parameter beta, and instead only applies the seasonal parameter gamma, using the multiplicative method.

NOTE: The Seasonal Multiplicative model is only capable of handling data with a Seasonal component but no Trend Component.