Accurate sales projections are of vital importance to the profitability and long-term survival of high-tech companies. This is especially true in the growth stage of product innovation, because major investments and marketing decisions are made in this phase. By examining recent empirical studies focusing on consumer behavior in high-technology markets, several factors are identified that can affect individual buying decisions and aggregate sales, namely, interpersonal communication, democratization of innovation, direct and indirect network effects, forward-looking behavior, and consumer heterogeneity. Against this background the diffusion-modeling and the utility-based approach are reviewed in terms of their basic conception and their applicability to the markets concerned. Based on this investigation a sales forecasting model for high-tech products, specifically in the growth stage, is developed. The model has a utility-theoretic background and a logistic structure. Since data are scarce in this early stage of the product life cycle two versions of the new approach are discussed, an extended version considering forward-looking behavior and a more parsimonious (''myopic'') one. The performance of the new model is demonstrated using real sales data on the CD player, the DVD player/recorder, and the digital camera market. The empirical comparisons include alternative specifications of the Bass diffusion model as well as a proportional hazard model and consist of two steps. First, the models are checked as to whether they are able to represent the sales data at all. It is shown that both versions of the proposed model are at least equivalent, if not superior, to the models used as benchmarks in terms of fit. In the next step, the models are applied to predict future sales in the three markets. The resulting forecasts show that the proposed model performs significantly better than its benchmarks. Its parsimony enables reliable predictions, even in cases, where only short time series are available for parameter estimation. The model is able to anticipate decreasing diffusion rates as they occur at the end of the growth stage and, thus, helps to avoid overoptimistic sales forecasts, which may cause severe economic damages. The new approach is easy to calibrate and can be applied without specialized econometric expertise.

Starting from a comprehensive examination of recent empirical studies focusing on consumer behavior in high‐technology markets and the resulting identification of factors probably affecting individual buying decisions as well as aggregate product sales, Decker and Gnibba‐Yukawa developed and empirically verified a utility‐based sales forecasting approach in their earlier work. Based on data for 14 consumer electronic products and using the Gompertz curve as a benchmark, Goodwin and Meeran carried out a “more extensive testing” of this proposal. However, at least from a practical point of view, the plausibility of their testing framework regarding the market potential m is not unquestionable. This paper, therefore, first discusses some theoretical aspects of both approaches by addressing issues challenged by Goodwin and Meeran, especially regarding the use of short time series and the consideration of replacement purchases. Then, the quasi‐endogenous estimation method for m favored by Goodwin and Meeran for the Gompertz curve is examined in terms of sensitivity to better understand its influence on sales forecasts, and the adequacy of the suggested range for m in the case of the approach by Decker and Gnibba‐Yukawa is investigated. In addition, the results presented in Goodwin and Meeran are considered from a more distant perspective, and possible causes of the variations in forecasting accuracy are discussed, which finally reveals that the forecasting performance of the utility‐based approach is not that “disappointing” as claimed. It provides more accurate (or at least equivalent) forecasts than the Gompertz curve approach in 64% of the cases considered. Furthermore, if product 14 (portable MP3 players) is excluded from the analysis because of the nonconsideration of probably existing product improvement effects, then the utility‐based approach, on average, outperforms the benchmark in all forecasting years. Altogether, this suggests that the approach by Decker and Gnibba‐Yukawa could achieve more accurate forecasts when applying a more reasonable range for m, rather than varying it between 2 and 15 times the cumulative sales by the end of year 7 as proposed by Goodwin and Meeran. It turns out that the Gompertz curve approach can perform on a par with the utility‐based approach in high‐technology product sales forecasting based on short time series if the market potential m is estimated exogenously. A combination of the outcomes of both approaches can even lead to more accurate forecasts as when being used individually insofar as composite forecasting seems to be a practicable approach to the problem of shorter time series compelled by the accelerated diffusion speed in high‐technology markets, rather than relying on one presumably “best” model.

No abstract