Decentralized Decision-Making in Retail Chains: Evidence from Inventory Management

This paper examines the effects of decentralizing decision-making in multi-establishment firms. Using a unique dataset from the Liquor Control Board of Ontario (LCBO), we assess the impact of allowing store managers to control the inventory replenishment decisions of their stores. We first present evidence of strong heterogeneity across the inventory decisions of the 634 store managers in the retail chain. We then study the sources of this heterogeneity by focusing on differences across stores in the structure of the profit function. We estimate a separate dynamic structural model of inventory management for each store manager in the retail chain using two years of daily data. We find very substantial heterogeneity across stores in unit costs of holding inventory, placing orders, and stockouts, and in fixed ordering costs. Using counterfactual experiments based on the estimated model, we find that a centralized inventory management system would yield a 0.4% increase in annual profit for LCBO, equivalent to $6.8 million. This modest effect is the result of combining two substantial effects with opposite signs: the negative effect of delaying the processing of information is more than compensated by the large reduction in ordering and storage costs from eliminating store managers' behavioral biases and heterogeneous skills (i.e., 25.5% on average, 6.3% for the median store). Furthermore, the gains from centralization are very heterogeneous across stores in the retail chain, with both very substantial losses and gains. These distributional effects within multi-divisional companies can be relevant when deciding whether to adopt an organizational change.