Forecasting intermittent demand: a comparative study

摘要: Methods for forecasting intermittent demand are compared using a large data set from the UK Royal Air Force. Several important results found. First, we show that traditional per period forecast error measures not appropriate demand, even though they consistently used in literature. Second, by comparing ability to approximate target service levels and stock holding implications, Croston's method (and variant) Bootstrapping clearly outperform Moving Average Single Exponential Smoothing. Third, performance of Croston can be significantly improved taking into account an order is triggered period.