Forecasting with a Repeat Purchase Diffusion Model
提出一种方法，在无销售数据时预测处方药销量，并利用少量销售数据更新预测。通过医生对药品属性的感知估计模型参数，并在19种药品上验证，预测结果令人鼓舞。
A methodology for forecasting the sales of an ethical drug as a function of marketing effort before any sales data are available and for updating the forecast with a few periods of sales data is presented. Physicians' perceptions of the drug on a number of attributes, e.g. effectiveness, range of ailments for which appropriate, frequency of prescriptions, are used to estimate the parameters of a model originally proposed by Lilien, Rao and Kalish (Lilien, G. L., A. G. Rao, S. Kalish. 1981. Bayesian estimation and control of detailing effort in a repeat purchase diffusion environment. Management Sci. 27(May) 493–506.). This model conceptualizes the drug adoption process as a repeat purchase diffusion model; sales are expressed as a function of a drug's own and competitive marketing efforts and of word of mouth. The model is first validated in this paper via predictive testing on 19 drugs prescribed by three types of physicians. The forecasting methodology is illustrated using physicians' perceptions on these drugs. Forecasts obtained without any sales data, and updated forecasts using seven periods of sales data are presented, and are encouraging.