New Zealand
Statistical Association
|
||
|
Bill Bolstad Holt-Winters Forecasting with Trigonometric Seasonals
The additive Holt-Winters procedure is a short-term forecasting method for a time series that exhibits a structural pattern that combines a local linear trend and seasonal effects. It generalizes exponential smoothing in the sense that the smoothed structural parameter estimates are updated as each observation becomes available by taking a weighted average of a term based on the previous estimates and a term based on the current observation. These estimates are extrapolated into the future to give the forecast. This paper introduces a modification into the Holt-Winters procedure so trigonometric seasonals can be used instead of dummy seasonals. This has the advantage of reducing the number of parameters required which frequently improves forecasting accuracy. The additive Holt-Winters method is contrasted with the method of discounted least squares, another generalization of exponential smoothing also suitable for time series combining a local linear trend and seasonals. |
|
|
|
||
