Following to Lead: Dominant Followership in Sequential Price Competition
Abstract
A dominant platform might be expected to use its market position to charge more than rivals, even for the same products. Amazon does not. Using synchronized high-frequency cross-platform pricing data, I find that Amazon prices at or below the lowest rival price. Its distinctive role appears instead in how it responds to rival price changes, where I document three patterns. First, after a rival raises its price, Amazon's probability of raising its own price more than doubles, whereas rivals show no detectable response when Amazon moves first. Second, when Amazon responds, it tends to match the rival's new price. Third, matched rival price increases remain in place longer than unmatched ones, while matched decreases show no comparable persistence pattern. I call this strategy dominant followership. In addition to documenting this pattern, I provide a stylized model as a potential explanation. A sequential Bertrand model modified to break ties in favor of the dominant firm shows why dominant followership can raise prices above the simultaneous-move Nash benchmark. Because the dominant firm keeps most demand at price parity, it may prefer matching a rival's higher price to undercutting it. The rival is willing to raise price first when that match is predictable; once the price is matched, rival response friction can help preserve it by limiting later undercutting. Using simulation studies, I show that Q-learning algorithms learn dominant-followership behavior endogenously. The learning result separates initiation from incidence: the rival's algorithm learns to initiate higher prices, nearly tripling the market price, while the dominant firm captures 91 percent of the resulting gains. This speaks to the FTC's Project Nessie allegation: individually optimal responses can teach rival algorithms that raising price is profitable, even without explicit coordination. The main insight of this paper is that market power need not operate through price leadership: it can operate through anticipated response. Modeling such sequential responses is important for capturing the essence of modern algorithmic price competition.
Can Consumer AI Discipline Algorithmic Pricing?
Work in progress
This project studies whether consumer-side AI shopping agents can weaken dominant-platform pricing power in markets where firms use pricing algorithms. Many consumers begin product search inside a dominant platform and compare only a limited set of alternatives, allowing firms to benefit from search frictions and default position. AI shopping agents can change this environment by searching across platforms, tracking recent price histories, and making price and non-price tradeoffs easier to compare. However, lower search costs need not eliminate platform advantage, because consumers may still value the dominant platform's non-price advantages, such as delivery speed, return convenience, and trust.
I model consumer-side AI as lowering search frictions and expanding consumers' consideration sets, which reduces the dominant firm's demand advantage. Starting from a market in which the dominant platform captures a large share of demand, I ask how pricing incentives change as AI agents make consumers more willing to consider rival sellers. The main exercise is a threshold analysis: how much must the dominant firm's demand advantage fall before the dominant-followership mechanism breaks down? I then examine the implications for consumer welfare. The project speaks to a policy question in algorithmic pricing: whether consumer-side AI can discipline dominant platforms through better search, or whether dominant-platform advantages are too persistent for the market to self-correct.