Adaptive Smoothing

May also be known as adaptive control (Bowerman section 8.2) or adaptive updating of the discount factor (Montgomery section 4.6.4).

Based on simple (single) exponential smoothing with one discount factor (smoothing constant, lambda).

Simple exponential smoothing with a fixed discount factor (lambda) may have difficulty following changes to a time series. Adaptive control allows lambda to change or adapt to changes in the time series model. Decisions on when to change lambda and how to change it are based on a tracking signal, which is based on one-step-ahead forecasts.

Based on a procedure introduced by Trigg and Leach.

Related terms:

• Smoothed error

• Tracking signal