Forecasting market diffusion of innovative battery-electric and conventional vehicles in Germany under model uncertainty

In this research paper accuracies (percentage errors, MAPE) of different procedures (growth, ARIMA(X), exponential smoothing and deterministic trend models) in forecasting new passenger car registrations in Germany are presented. It is found that the Logistic Growth Model provides rather accurate predictions of the number of new registrations (total number, which still refers to predominantly conventional gasoline and diesel vehicles) for the forecast period of the study. However, the Bass diffusion model is recommended for predicting the new registration numbers of the innovative battery-electric technology. Furthermore, it is exemplarified that the Bass coefficient of imitation q, in contrast to the coefficient of innovation p, is robust to a variation of the assumed market potential M. Therefore, q should also contribute to a stable short-term forecast (given a variation of M), provided that a period in the early phase of the product life cycle is considered. The study also shows that with the bulk of the procedures, percentage forecast errors are obtained which lie in a narrow margin for the established product passenger car, but not for the innovative battery-electric propulsion technology. So while the careful selection of the forecasting model seems rather negligible for the established product, it is essential for the innovative product. In addition, new registration figures in the German federal states were forecasted, which in turn were used to calculate pooled forecasts for Germany. In general, no increase in forecast accuracy was achieved by means of pooling compared with direct forecasting (i.e. from the national time series).