Diffusion models in forecasting: A comparison with the Box–Jenkins approach

Diffusion models represent an interesting approach to the study of technical change and evolutionary processes. In the business area many models based on the diffusion approach have been developed particularly in forecasting, as in new product analysis. Some authors have also used these models extensively in technological forecasting applications. Despite the fact that at present diffusion models are a widely employed tool, their value compared to other forecasting techniques has not yet been well established. The aim of this paper is to carry out a comparative analysis on the descriptive and forecasting accuracy of the Box–Jenkins and diffusion models, on the basis of many different time series. The numerical procedures used in parameter calibration and the performance indexes employed in comparing the models' performance are explicated. Some general conclusions regarding the conditions of diffusion models' practical application are put forward.