Estimating Regression Parameters in Inventory Management

• Activity Number: 364
• Type: Contributed
• Date/Time: Wednesday, August 10, 2005 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): Samaradasa Weerahandi*+ and Martin Koschat and Xiaolin Teng
• Companies: Time Warner, Inc. and Time Warner, Inc. and Time Warner, Inc.
• Keywords: Ordinary least squares ; estimation loss functions ; newsboy problem ; unbiasedness by attribute

Abstract

In applications such as Inventory Management, estimation of regression parameters by OLS leads to undesirable sales predictions and inefficient inventory allocations. For instance, they tend to underestimate the sales of large stores and overestimate the sales of small stores---highly undesirable properties in inventory management. In this paper, we propose a methodology to overcome this drawback. It is based on a natural estimation loss function we derive from the elements of the inventory problem. One also can apply the proposed method without any loss function by providing two parameters to achieve the desired level of unbiasedness observed in a usual goodness-of-fit chart. The methodology will be illustrated by an application in a newsboy type problem.