This
technique is an exponential smoothing method that is used when the data is characterized
by linear trend and multiplicative seasonal variation. The form of the model is stated as:
yt =
(Bo + B1t) (SNt) + et
The
method itself is similar to the Holt two parameter methodology. Basically, the methodology is extended by an
additional equation to estimate the seasonal component to the time series. The following seven data items are required
for this method:
1 Alpha; where alpha is between 0 and 1 is associated with exponential adjustments to the permanent component (new value) of the next period forecast.
2 Beta; where beta is between 0 and 1 and is associated with adjustments to the linear trend in the calculation.
3 Gamma; where gamma is between 0 and 1 and is associated with the internal calculation updates to the seasonal adjustment factor.
4 Time Period: 1, 4, or 12.
5 Initial Base Value for Time Series - optional.
6 Value of next observation in the Linear Trend (slope): optional.
7 Number of Periods over which to forecast.
Simulation
is used to determine the optimal combination of smoothing constants alpha,
beta, and gamma. Try forming a set of
combinations with the three parameters. Sequentially range each from 0.01 to 0.30 in
steps. The combination that minimizes
the forecast error is the combination to use for actual forecasting.