How should additive Holt–Winters estimates be corrected?

An important by-product of a good forecasting procedure is to gain some understanding of how the series behaves. This paper looks primarily at how well the Additive Holt–Winters method succeeds in this task. Additive Holt–Winters is a widely used technique which has had some success in forecasting competitions. The paper shows that the basic form of the method does not give good estimates for the level and seasonal features of a time series. The paper then looks at various ways of correcting the estimates, looks at their performance and recommends a particular approach.