At the end of each year, the German Federal Highway Research Institute (BASt) draws the balance of the road-safety development by forecasting the accident and casualty numbers of the closing year. They describe the development of 27 time-series of accident and casualty numbers disaggregated by road user types, age groups, type of road and the consequences of the accidents. However, at the time of publishing, these series are only available for the first eight or nine months. To make the balance for the whole year, the last three or four months are forecasted. The objective of this study was to improve the accuracy of these forecasts through structural time-series models which derive the forecasts for the remaining months from the trend observed in the first months of the year and the dynamics of the earlier development. To take the weather conditions into account, meteorological variables are included into the forecasting model. To test the models, the final data of the last 15 years were predicted from the preliminary data of the first months for each year. These predictions were compared to the actually observed final numbers of accidents and casualties. The results show that, compared to the earlier heuristic approach, 25 out of the 27 time-series are forecasted more precisely by the models presented. Only two series show a slightly increasing prediction error. When comparing models with and without meteorological variables, 23 out of 27 series were predicted more accurately when taking the weather into account. Apart from increasing the forecasting precision, the inclusion of meteorological variables also allows estimating to what extent changes in the observed numbers of accidents and casualties can be attributed to the specific weather condition in a particular month. We conclude that structural time series modelling and the inclusion of meteorological Variables clearly improves the forecasting precision in the year-end prognosis of the German accident and casualty numbers. The improved prediction due to the inclusion of meteorological variables confirms the dependence of these numbers on the weather condition throughout the year. (Author/publisher)
Abstract