Only two forecasting methods have been designed specifically for intermittent demand with possible demand obsolescence: Teunter-Syntetos-Babai (TSB) and Hyperbolic-Exponential Smoothing (HES). When an item becomes obsolete the TSB forecasts decay exponentially while those of HES decay hyperbolically. Both types of decay continue to predict non-zero demand indefinitely, and it would be preferable for forecasts to become zero after a finite time. We describe a third method called Exponential Smoothing with Linear Decay that decays linearly to zero in finite time, is asymptotically the best method for handling obsolescence, and performs well in experiments on real and synthetic data.
- Croston's method