APPLICATION OF TRAMO-SEATS AUTOMATIC PROCEDURE FOR FORECASTING INTERMITTENT DEMAND PATTERNS

Managing intermittent demand represents a very critical task in term of forecasting and stock control due to the variability both of demand sizes and demand arrivals. In this pa-per the forecasting issue is tackled by comparing different extrapolative forecasting ap-proaches. In particular, the SARIMA model (Seasonal Autoregressive Integrated Moving Average) is applied on 60 real time series by means of the TRAMO-SEATS procedure, which is a versatile and automatic procedure allowing a quick identification of the best performing SARIMA model for each item. The forecasting performances are then compared with those obtained by the well-known methods of Croston and Syntetos-Boylan, which represent two modified versions of the simple exponential smoothing specifically introduced for estimating the mean demand per period in case of intermittent demand profiles. Furthermore, the aggre-gation of forecasts in lower-frequency ‘time buckets’ is implemented in order to evaluate how these methods behave on aggregated time horizons.