A. Serdar Simsek

Research

• Variety and Inventory Tradeoff in Retailing: An Empirical Study (with G. Kok)
  
  Abstract
  We investigate the trade-off between variety and inventory on sales in retailing using a proprietary data set from a bookstore chain. In general, more variety is expected to generate higher traffic to store and higher conversion for incoming customers. Another argument is that too high a level of variety reduces sales due to the confusion it creates in the minds of the customers (choice paradox). At the same time, higher variety implies lower inventory per product due to limited shelf space in the stores. Hence, a critical question for a retailer is to determine where their stores are currently positioned on the variety-sales curve, and whether or not they should increase/decrease variety by extending/cutting the long tail of slow-moving products. We consider the number of subcategories in a store as a measure of macro-level variety and number of products in a subcategory as a measure of micro-level variety. We consider the average level of inventory (units per product) as a measure of inventory depth and inventory dispersion as a measure of inventory policy effectiveness (i.e., the degree of which inventory dispersion follows the sales dispersion across items). We exploit the differences between stores and variations over time in the data to investigate the net impact of variety on sales of total category and on sales of top-selling products. We introduce instrumental variables to control for endogeneity of variety and inventory. We also control for store traffic, seasonality, and forecasting of demand change effects, based on detailed knowledge of the dynamic inventory optimization policy of the retailer. We find that both micro and macro variety increase sales of the total category. More interestingly, variety increases the sales of the top selling products as well, a somewhat surprising positive effect of presenting the long-tail of products to the customer on the fast moving products. We also demonstrate how a retailer can operationalize our models and insights, and show that a retailer can significantly increase its revenue by optimally swapping shelf spaces across product categories based on the estimated category-specific variety effects.