Mail-order Demands for Style Goods: Theory and Data Analysis

A number of alternative data-generating processes are explored for mail-order demands for seasonal style-goods. Weekly demands for 126 items over an 18-week selling season provide the empirical data. Arguments are presented which result in the following candidates for data-generating processes: (1) ratios of successive forecasts are distributed lognormally; (2) ratios of successive forecasts are distributed as t_ (Student); (3) actual demands during unequal time periods are distributed as negative binomial. Analysis of the data suggests the negative binomial data-generating process as both most closely representing the underlying process and being simple to adapt to a decision model. The paper concludes with an example of the use of the chosen data-generating process in a decision framework, and deals briefly with some issues of implementation.