Refining and Operationalizing Entropy‐Maximizing Models for Business Applications. 商业应用模式下熵最大化模型的应用与改进

The aim of this article is to explore the main issues involved with operationalizing entropy‐maximizing models in a retail business context. We draw on our experiences of work undertaken with a variety of international blue‐chip clients. First, we examine issues relating to demand estimation, focusing on practical considerations involved in different types of application. These issues include choice of demand estimation methodology, dealing with nonresidential flows and demand elasticity. Second, we investigate the supply‐side or attractiveness component in detail, including a review of how such factors as brand preference, brand loyalty, and the so‐called network effect can be addressed within a spatial interaction modeling framework. We then explore the distance decay variable in more detail before examining more specific measures of model performance. Historically in the literature, goodness of fit has been used as a measure of model performance, but we shall argue that goodness of forecast is the metric by which model performance should be measured. The difficulties of trying to achieve this are outlined and some examples discussed.