Algorithmic Collusion on Online Marketplaces
The use of algorithms to set prices is particularly popular in online marketplaces, where sellers need to take quick decisions in complex dynamic environments. In this article, I investigate the role of online marketplaces in facilitating or preventing collusion among sellers that use pricing algorithms. In particular, I investigate a platform that has the ability to give prominence to certain products and automates this decision through a reinforcement learning algorithm, that maximizes the platform’s profits. Depending on whether the business model of the platform is more aligned with consumer welfare or with sellers’ profits (e.g., if it collects quantity or profit fees), the platform either prevents or facilitates collusion among algorithmic sellers. If the platform is also active as a seller, the so-called dual role, it is able to both induce sellers to set high prices and appropriate most of the profits. Importantly, self-preferencing only happens during the learning phase and not in equilibrium. I investigate a potential solution: separating the sales and marketplace divisions. The policy is effective but does not fully restore the competitive outcome when the fee is distortive, as in the case of a revenue fee.