Data Analysis with R, Second Edition - Second Edition

Double exponential smoothing

Where the simple (single) exponential smoothing method we've seen in the last section is only really appropriate for series with trend, double exponential smoothing forecasts – also known as Holt's linear method – can be used for series with trend (but no seasonality). Our temperature series would be a fine fit for this type of forecasting method.

Holt's linear method builds on simple exponential smoothing by introducing a term for the trend. So now, in addition to the alpha smoothness parameter for the level (trend-less series), we have a beta parameter which controls the smoothness of the trend component. We can perform Holt's linear method using the holt function from (you guessed it!) the forecast package. ...