Reference Dependence in Multilocation Newsvendor Models: A Structural Analysis

Reference Dependence in Multilocation Newsvendor Models: A Structural Analysis

0.00 Avg rating0 Votes
Article ID: iaor20108793
Volume: 56
Issue: 11
Start Page Number: 1891
End Page Number: 1910
Publication Date: Nov 2010
Journal: Management Science
Authors: , ,
Keywords: newsboy problem, preference modelling
Abstract:

We propose a behavioral theory to predict actual ordering behavior in multilocation inventory systems. The theory rests on a well-known stylized fact of human behavior: people's preferences are reference dependent. We incorporate reference dependence into the newsvendor framework by assuming that there are psychological costs of leftovers and stockouts. We also hypothesize that the psychological aversion to leftovers is greater than the disutility for stockouts. We then experimentally test the proposed theory in both the centralized and decentralized inventory structures using subjects motivated by substantial financial incentives. Consistent with the proposed theory, actual orders exhibit the so-called ‘pull-to-center’ bias and the degree of bias is greater in the high-profit margin than in the low-profit margin condition. These systematic biases are shown to eliminate the risk-pooling benefit when the demands across store locations are strongly correlated. Because the proposed model nests the standard inventory and ex post inventory error minimization theories as special cases, one can systematically evaluate the predictive power of each alternative using the generalized likelihood principle. We structurally estimate all three theories using the experimental data, and the estimation results strongly suggest that the proposed behavioral theory captures actual orders and profits better. We also conduct two experiments to validate the behavioral model by manipulating the relative salience of the psychological costs of leftovers versus that of stockouts to alleviate the pull-to-center bias.

Reviews

Required fields are marked *. Your email address will not be published.