Research
Work-in-progress
Localized Epsilon-Greedy Q-learning and Price Competition in Amazon-Like Buybox Markets
This paper examines the performance of Q-learning algorithms in a “winner-take-all” pricing environment, such as the Amazon Buybox. Motivated by the increasing use of algorithms for pricing and the potential for collusion, this paper makes two contribu- tions. First, it investigates whether collusive pricing trends observed in previous literature persist even in a setting designed to promote price competition, like the Buybox. Second, it proposes a refinement to the standard epsilon-greedy Q-learning algorithm called Localized Epsilon- Greedy Q-learning, where price exploration is focused around the cur- rent best price estimate. Through simulations, I demonstrate that the Buybox mechanism does not prevent collusive pricing and that local- ized exploration can lead to faster convergence to the monopoly price and higher profits for firms compared to the traditional epsilon-greedy approach.
Estimating “Reverse Payments” in Pay-for-Delay Settlements
So called pay-for-delay deals have been the subject of scrutiny by regulators, courts, legal-scholars, and economists. How- ever, the Supreme Court ruled that in order for a pay-for-delay deal to be anti-competitive, plaintiffs must demonstrate that there is a “large” and “unexplained” payment. However, many pay-for-delay deals have shifted away from cash payments towards business deals that heavily favor the generic entrant. Therefore, being able to estimate the value of these reverse payment deals is crucial. This paper proposes a method for estimating the value of these payments using event study abnormal returns.