Work out exponential smoothing instantly with clear inputs, formula shown and shareable results.
Simple exponential smoothing updates the level as previous forecast plus alpha times the last error, which is equivalent to an exponentially decaying weighted average of all past observations. Because there is no trend term, the forecast is flat for every horizon. Alpha controls responsiveness: the weight on an observation k periods old is alpha(1-alpha)^k, giving a half-life of ln(0.5)/ln(1-alpha).
Simple exponential smoothing
L(t) = F(t) + alpha x (A(t) - F(t)); forecast for any horizon = L(t); half-life = ln(0.5) / ln(1 - alpha)
Fit it by minimising one-step-ahead squared error on historical data. Values from 0.1 to 0.3 suit stable series; higher values track volatile series but amplify noise.
Simple smoothing models level only. Add Holt's trend term for a sloping forecast and a seasonal term for repeating patterns.