There is a growing recognition that e-market planners and various planning agencies in Information Technology sectors have a significant interest in measuring and forecasting the growth of e-commerce. The difficulties lie in finding a forecasting model that can incorporate both internal and external influences on diffusion, as well as an acceptable measure for e-commerce growth. This study uses models based on the knowledge of traditional diffusion theories as well as artificial neural networks. Additionally, it integrates the two into a hybrid model in order to study e-commerce growth. A count of dot-com hosts is used as a reliable measure of e-commerce growth in all the models. Our study demonstrates that a simple Neural Network model, if properly calibrated, can create a very flexible response function to forecast e-commerce diffusion growth. The neural network model successfully modeled both the internal and external influences in the data, while the traditional formulations could only model the internal influences. The predictive validation of the results was enhanced by replicating the comparisons on simulated data with various degrees of external influence. The study suggests that when external influences are present, the neural network model will be superior to the best traditional diffusion model.
Keywords: E-commerce, Dotcom, Forecasting, Neural Network, Diffusion models, and E-market Planning
(ProQuest: ... denotes formulae omitted.)
Studying the diffusion of e-commerce is extremely important for both government and business investors and policymakers for effective planning [Press 1997; Yao 2004]. However, industry and academic researchers found that measuring, forecasting and tracking the global diffusion of e-commerce faces two hurdles. The first problem is one of appropriately modeling the diffusion, both to understand the phenomenon and to forecast the diffusion for planning purposes. The process of innovation diffusion has been extensively researched [Rogers 1983], and several traditional diffusion models have been used to explain and forecast the phenomenon. Significant research in the past has used such models for the explanation and prediction of the diffusion of different technological innovations - e.g., the Bitnet [Gurbaxani 1990] organizational forms [Mahajan et al. 1998], corporate governance mechanisms [Venkatraman et al. 1994], the Internet [Rai et al. 1998], web-based shopping system [Changsu & Galliers 2004] and so on. These models are therefore a logical first choice in any attempt to understand and forecast e-commerce growth, but they do have some limitations.
The second problem is the difficulty in measuring the growth of diffusion of innovation. For some technological innovations, measurement of diffusion can be obvious. For instance, measuring the diffusion of cell phones is simply a matter of tracking sales of cell phones or subscriptions. Tracking diffusion of e-commerce is much more difficult. One obvious measure, sales dollars, is much less clear, since the portion of sales of each firm attributed to the Internet is not precisely tracked by most firms. In Section 2 we elaborate on the issue of e-commerce and the measurement of its growth.
The key idea of traditional diffusion modeling (discussed in greater detail in section 3) is to assume that there are a fixed number of potential adopters of new innovations. Therefore, this adoption process targets a decreasing number of adopters as time goes by. The growth rate of adoption can be constant [Fourt & Woodlock 1960]. [Mansfield 1961] proposes that the diffusion process follows a simple logistic curve (s-shaped) over time through imitation. [Bass 1969] suggests two main factors that are responsible for the diffusion growth process: imitation (or contagion) and innovation. [Mahajan & Muller 1979] later called them internal and external influences. Despite the popularity of the Bass model in explaining the diffusion of innovation of new products, research identified its drawbacks in forecasting growth in the near future. …