Steering via Algorithmic Recommendations
54 Pages Posted: 13 Jan 2020 Last revised: 2 Apr 2020
Date Written: December 1, 2019
A platform owner can be an intermediator and a seller. Does the dual identity lead to steering incentive? We empirically examine a dual-identity digital platform's use of a product recommendation algorithm called "Frequently Bought Together" (FBT). We analyze unique high-frequency data for over 6 million popular products sold on the platform. First, we document that comparing to non-platform-selling products, platform-selling ones make a similar number of FBT recommendations but receive 73% more. Second, we observe patterns consistent with self-preferencing: controlling for price, sales, and rating, the same product is 8% less likely to be recommended during the platform's temporary absence. Third, we show that recommending platform-selling products may not maximize platform-level sales: "platform to platform" recommendations are 50% less effective than "platform to non-platform" ones. Lastly, focusing on products which both the platform and third parties sell, we document the existence of price competition and a strong positive association between platform dominance and the number of recommendations received.
Keywords: Digital Platform; e-Commerce; Algorithmic Bias; Big Data; Product Recommendation; Platform Design; Vertical Integration
JEL Classification: D22; D43; D83; L11; L43; L81; L86; M21; M31; M38
Suggested Citation: Suggested Citation