Forecasting with growth curves: An empirical comparison

Abstract

The primary objective of this paper is to compare the forecasting performance of the increasingly wide range of growth curve models. Seventeen models are used to forecast the development of telecommunications markets, represented by 25 time series describing telephone penetration in 15 different countries. Forecasting performance is measured by root mean square error and mean absolute percentage error over the last 10 or 11 years of the series, the model parameters having been fitted over the previous 20 years. Note is taken of the convergence of the estimation process, the significance of parameters and the plausibility of the estimated saturation level. The local logistic, simple logistic and the Gompertz models are shown to significantly outperform more complex models such as the extended logistic and FLOG models.