Abstract
We examine self-preferencing in Amazon’s “Similar item to consider” box. We document a stark difference across domains: while Amazon frequently recommends products other than its own Amazon Basics (AB) brand on its US domain, it systematically suppresses these recommendations on its Canadian domain. Leveraging this difference, we estimate the causal effect of algorithmic exclusion on sales outcomes: we compare the Canadian sales of products that are and are not recommended in the US, using each product’s US performance to control for its underlying quality. We find that non-AB products that merit a recommendation in the US but are denied one in Canada suffer a sales loss of at least 12%. This finding quantifies the significant market distortion caused when gatekeepers favor their own inventory over rival goods.
