MANAGEMENT SCIENCE Articles in Advance, pp. 1–18 ISSN 0025-1909 (print)  ISSN 1526-5501 (online) http://dx.doi.org/10.1287/mnsc.2014.1940 © 2014 INFORMS Push, Pull, or Both? A Behavioral Study of How the Allocation of Inventory Risk Affects Channel Efficiency Andrew M. Davis Samuel Curtis Johnson Graduate School of Management, Cornell University, Ithaca, New York 14853, adavis@cornell.edu Elena Katok Jindal School of Management, University of Texas at Dallas, Richardson, Texas 75080, ekatok@utdallas.edu Natalia Santamaría Samuel Curtis Johnson Graduate School of Management, Cornell University, Ithaca, New York 14853, nsantam@cornell.edu I n this paper we experimentally investigate how the allocation of inventory risk in a two-stage supply chain affects channel efficiency and profit distribution. We first evaluate two common wholesale price contracts that differ in which party incurs the risk associated with unsold inventory: a push contract in which the retailer incurs the risk and a pull contract in which the supplier incurs the risk. Our experimental results show that a pull contract achieves higher channel efficiency than that of a push contract, and that behavior systematically deviates from the standard theory in three ways: (1) stocking quantities are set too low, (2) wholesale prices are more favorable to the party stocking the inventory, and (3) some contracts are erroneously accepted or rejected. To account for these systematic regularities, we extend the existing theory and structurally estimate a number of behavioral models. The estimates suggest that a combination of loss aversion with errors organizes our data remarkably well. We apply our behavioral model to the advance purchase discount (APD) contract, which combines features of push and pull by allowing both parties to share the inventory risk, in a separate experiment as an out-of-sample test, and we find that it accurately predicts channel efficiency and qualitatively matches decisions. Two practical implications of our work are that (1) the push contract performs close to standard theoretical benchmarks, which implies that it is robust to behavioral biases, and (2) the APD contract weakly Pareto dominates the push contract; retailers are better off and suppliers are no worse off under the APD contract. Data, as supplemental material, are available at http://dx.doi.org/10.1287/mnsc.2014.1940. Keywords : behavioral operations management; inventory risk allocation; supply chain contracts History : Received November 15, 2012; accepted February 21, 2014, by Serguei Netessine, operations management. Published online in Articles in Advance. 1. Introduction Location and ownership of inventory is one of the key drivers of supply chain performance. Even in a simple supply channel—single retailer, single supplier, and full information—researchers and companies have found that common wholesale price contracts with different inventory allocations affect channel efficiency (e.g., Lariviere and Porteus 2001, Cachon 2003, Kaya and Özer 2012). Determining the best channel design and inventory allocation in supply chains involves difficult trade-offs that can have a direct effect on a firm’s survival. For instance, Randall et al. (2002) provide examples of companies in which the difference between success and bankruptcy may be attributed to different inventory allocation strategies, whereas Randall et al. (2006) identify empirical relationships between a firm’s decision to own the inventory and several key performance indicators. Traditional channels use a push structure in which the retailer makes stocking decisions, owns the inven- tory, and thus incurs the holding cost, as well as the cost of any unsold product. However, Internet- enabled technologies now permit other supply chain arrangements for allocating inventory ownership and risk (the cost of unsold inventory), which may affect channel profitability (see Cachon 2004, Netessine and Rudi 2006). One such arrangement is the pull inventory system. Under this system the supplier makes the stocking (pro- duction) decision and therefore incurs the holding costs and inventory risk. The retailer provides a storefront (real or virtual), but products flow from the supplier to the end customer with minimal exposure of the retailer to inventory risk. One practical implementation of the pull inventory system is a drop-shipping arrangement— the retailer is never exposed to the inventory at all; the 1