Abstract

While the adoption of electric vehicles can reduce CO2 emissions, the extent of this reduction depends on the growth of the EV market. Sigmoid growth models, such as logistic or Gompertz function models, can be used to predict expected EV sales trends; however, their quality has not yet been comprehensively analyzed, as this would require looking into the future to compare today’s predictions with future data. Since this is obviously not possible, this study takes a retrograde approach. It uses the available historical data to create forecasts that are then compared with the actual values from subsequent years. For example, a forecast based on data from 2010 to 2014 can be compared with the values achieved in years from 2015 to 2025. The quality of the functions is assessed using fit indices. Among the ten distinct functions tested, including two equivalent Gompertz functions, and under the stated saturation assumptions, the Gompertz family offers the most stable retrospective forecasts of EV stock (prediction period MAPE of 16.0% for the global data, against 15.1% for the generalized logistic, which performs comparably). The generalized logistic attains marginally better global point accuracy, whereas Gompertz is preferred for its greater stability across forecast origins and its more interpretable parameters.

IPC Classification

Keywords