This paper presents a new modelling framework for the freight demand forecasting problem (FDFP), and shows the effect of ignoring spatial interactions in a model's specification. This framework provides a new approach to assessing the probability of availability of load (pickup or delivery) over space. The FDFP problem is defined for several activities centres distributed over a bounded region, whose inhabitants exchange goods and services. Demand for freight transport is here assumed to be a stochastic process with both time and space interactions, whose underlying process is compound Poisson. During a given time interval, each site has an arrival rate, depending on its past and on the history of arrival rates at other locations. The dynamic dependence between arrival rates is assumed to be linear in a space-time autoregressive moving average (STARMA). The general mathematical form of the STARMA model is specified, together with its `optimal linear forecast'. The relationship between the STARMA and autoregressive moving average (ARMA) models is analysed. The methodology has the stages of simulation, estimation, forecasting, and comparison. The STARMA process was simulated for a specific numerical example; one time-series was generated for each of the example's nine sites for each of nine spatial parameter values.
Samenvatting