The last decade has seen rapid advances in telecommunications technology in an increasingly deregulated and competitive market place. Companies operating in these various markets have relied on demand forecasts to justify the considerable investment needed to ensure capacity availability at the right time. These new markets are typically composed of new consumers taking up a product or service for the first time, established users changing their usage patterns, users of competing services shifting to the alternative service and those exiting from this segment of the market altogether. This paper describes various models that have been used to understand market dynamics. Markets discussed include both established and new: mobile, the internet, and PSTN (public switched telephony network). Cross-sectional choice models of the mode of accessing the service are discussed along with models for usage in established markets. These models typically include price (and perceived price) differentials and use standard econometric methods, focusing on price elasticity estimation. Forecasting accuracy has been neglected. New product models may include additional 'drivers' such as aspects of service quality and the attributes of the products themselves. Both choice models of adoption of new products and Bass-type diffusion models have been used in forecasting. Because of the complexity of the 'drivers' of the adoption process, the successful modelling of these new markets has been limited, not least by inadequate data. Simulation models have been proposed to structure the problem more completely and overcome these inadequacies. Both these classes of model have not been effectively validated, researchers having been content just to propose a new approach without thoroughly testing it against alternatives. The only class of telecommunications forecasting problem that has been more thoroughly analysed are those needed to support operations such as call centres. This review paper describes the research that has been carried out on the three problem areas of established products, new products and operations, highlighting areas where further research is needed. The paper also serves as an introduction to the Special Issue on Telecoms Forecasting by describing how the papers contribute to the developing research agenda.

The research provides an assessment of the relevant literature on market share models and identifies the need for further research. Additional insights are generated in this study by using point-of -sale scanners for evaluating the forecasting performance of market share models under various conditions. Specifically, weekly store-level scanner data for four frequently purchased product categories-saltine crackers, baking chips, diapers and toilet tissue, and simulated data are used in this study. Consistent with theoretical expectations, the attraction models estimated by GLS produce the best forecasts even (1) at the brand level, and (2) when competitors' actions are predicted. However, the superiority of the attraction models is diminished when systematic errors are introduced to the values of the competitors' predictor variables in the holdout sample. In fact, naïve models outperform all types of econometric models when large errors are present in the competitors' predictor values, and among the econometric models, linear models produce better forecasts than attraction models. The need for estimating the models with GLS (as opposed to OLS) with the use of cross-sectional time-series data is also illustrated. Finally, guidelines are developed for practitioners and researchers on the usefulness of market share models for forecasting.

Prior research assessing the predictive validity of alternate market share models produced conflicting results and often found that econometric models performed worse than naïve extrapolations. However, contributors to IJF's recent issue on market share models suggested that such models are often misspecified, in part because they exclude promotional variables and are estimated on aggregate data. Thus, we used weekly scanner data to assess full, reduced, and naïve form for linear, multiplicative, and attraction specifications across different levels of parameterization. Consistent with specification-based arguments (1) econometric models were superior to naïve models, (2) GLS estimates of attraction models were superior when models were fully specified, (3) OLS estimates of linear models were superior when models omitted important variables, and (4) attraction models predicted best overall. Moreover, in general, unconstrained models yielded superior forecasts relative to constrained models because brand-specific parameters were heterogeneous for the product category tested.

The ability to forecast market share remains a challenge for many managers especially in dynamic markets, such as the telecommunications sector. In order to accommodate the unique dynamic characteristics of the telecommunications market, we use a multi-component model, called MSHARE. Our method involves a two-phase process. The first phase consists of three components: a projection method, a ring down survey methodology and a purchase intentions survey. The predictions from these components are combined to forecast category sales for the wireless subscribers market. In the second phase, market shares for the various brands are generated using the forecast of the number of subscribers tat are obtained in Phase I and the share predictions from the ring down methodology. The proposed methodology produces the minimum Relative Absolute Error for each market as compared to the forecasts from each individual component in the first phase. The value of the proposed model is illustrated by its application to a real world scenario. The managerial implications of the proposed model are also discussed.

In order to effectively make forecasts in the telecommunications sector during the growth phase of a new product life cycle, we evaluate performance of an evolutionary technique: genetic algorithms (Gas), used in conjunction with a diffusion model of adoption such as the Bass model. During the growth phase, managers want to predict (1) future sales per period, (2) the magnitude of sales during peak, and (3) when the industry would reach maturity. At present, reliable estimation of parameters of diffusion models is possible, when sales data includes the peak sales also. Cellular phone adoption data from estimates obtained from Gas exhibit good consistency comparable to NLS, OLS, and a naïve time series model when the entire sales history is considered. When censored datasets (data points available until the inflection point) are used, the proposed technique provides better predictions of future sales; peak sales time period, and peak sales magnitude as compared to currently available estimation techniques.

Purchase intentions are routinely used to forecast sales of existing products and services. While past studies have shown that intentions are predictive of sales, they have only examined the absolute accuracy of intentions, not their accuracy relative to other forecasting methods. For example, no research has been bale to demonstrate that intentions-based forecasts can improve upon a simple extrapolation of past sales trends. We examined the relative accuracy of four methods that forecast sales from intentions. We tested these methods using four data sets involving different products and time horizons; one of French automobile sales, two of US automobile sales, and one of US wireless services. For all four products and time horizons, each of the four intentions-based forecasting methods was more accurate than an extrapolation of past sales. Combinations of these forecasting methods using equal weights lead to even greater accuracy, with error rates about one-third lower than extrapolations of past sales. Thus, it appears that purchase intentions can provide better forecasts than a study, the consistence of the results in our study suggest that intentions are a valuable input to sales forecasts.