Double exponential smoothing (DES), also called Holt’s linear trend model, forecasts time series by exponentially weighting recent observations while modeling both the level and a linear trend. It suits data with a trend but no seasonal pattern. section 7.4.4 section 7.5
Use DES when the series has these characteristics:
- A visible linear trend that simple exponential smoothing cannot represent. section 7.4.6
- No seasonal component; Holt–Winters is the related extension for trend plus seasonality. section 7.4.4
- Recent observations should receive more influence than older observations through exponentially decreasing weights. section 7.4.6
Fit DES in Python
With a training series named train, import ExponentialSmoothing, specify an additive trend, and call fit(): des_model = ExponentialSmoothing(train, trend = 'add').fit(). The fitted model optimizes its parameters to capture the underlying trend. section 7.4.6
>>> from statsmodels.tsa.holtwinters import ExponentialSmoothing >>> des_model = ExponentialSmoothing(train, trend = 'add').fit()
Open the runnable statsmodels listing for the additive-trend DES fit. section 7.4.6
