Reference-Dependence in Multi-Location Newsvendor Models : A Structural Analysis

We propose a behavioral theory to predict actual ordering behavior in multi-location 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 left-overs and stock outs. We also hypothesize that the psychological aversion to leftovers is greater than the disutility for stock outs. We then experimentally test the proposed theory in both the centralized and decentralized inventory structures using subjects motivated by substantial nancial 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-pro t margin than in the low-pro t margin condition. These systematic biases are shown to eliminate the risk pooling bene t when the demands across store locations are strongly correlated. Since 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 pro ts 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 stock outs to alleviate the pull-to-center bias