The Predictive Power of Operational Demand Models
GUNN H F, Hague Consulting Group, HOORN A I J M van der, Ministry of Transport (AVV), The Netherlands
Between 1977 and 1984, research studies were undertaken by the Dutch Ministry of Transport aimed at development a predictive model system capable of producing detailed forecasts of travel demand, being multi-modal and capable of break-down by market segme
Between 1977 and 1984, research studies were undertaken by the Dutch Ministry of Transport aimed at development a predictive model system capable of producing detailed forecasts of travel demand, being multi-modal and capable of break-down by market segment, and carded through to the stage of assignment to networks in the car and rail.
The system covered the entire Netherlands, and is referred to as the National Model System (NMS).
In 1985, the prospects for the system ,were still unknown. The methodology had been tested, and the scientific groundwork had been laid to establish that model could reasonably be advanced as a forecasting tool with a basis in classical micro-economic theory. Backcasting was ruled out in the available timescale; model elasticities had been thoroughly checked, but no test of actual forecasting performance was possible.
The model system is still in use, having been extensively extended and refined in the intervening years. The authors retained a record of the assumptions and results from the 1985 projections. There are now published aggregate statistics against which to assess these.
The research reported in this paper uses these statistics to assess the accuracy of the first predictions, to analyse the major factors which have contributed to the discrepancies between forecast and outcome, to assess the degree to which later refinements of the NMS would have improved the predictions, and to identify the areas in which future research could lead to resolution of remaining problems.
The paper hopes to set out a structure within long-standing predictive tools can be monitored for shortcomings, from which potential for incremental improvements can be identified. In addition, it provides a quantification of the errors from different sources in the Case Study. These include consideration of the socio-economic data and network assumptions, in addition to pricing policies. The residual forecasting error is analysed in relation to different aspects of the model, with a view to recommending a research strategy leading to an improved capability.
Association for European Transport