Understanding Strategic Platform Entry and Seller Exploration: A Stackelberg Model

A team from Harvard University and Google Research has published a significant paper, 'Understanding Strategic Platform Entry and Seller Exploration: A Stackelberg Model,' accepted at The Web Conference 2026. The research tackles the pervasive empirical phenomenon where dominant platforms like Amazon, Apple, and DoorDash observe successful third-party products on their own marketplaces and then launch competing, often similar, offerings. The authors formalize this dynamic using a Stackelberg game theory model, where the platform acts as the strategic leader, committing to an 'entry policy' that dictates when it will copy and compete on a product.

For a single seller scenario, the team characterized the seller's optimal 'explore-exploit' strategy using a Gittins-index policy and provided an algorithm to compute the platform's optimal entry timing. However, the more complex and realistic scenario involves multiple sellers, where competition and information spillover occur. Here, the Gittins-index approach failed, prompting the researchers to employ deep reinforcement learning (deep RL) to examine and simulate seller equilibrium behavior. This computational framework allows for analyzing how different platform policies—from aggressive copying to delayed entry—directly impact sellers' willingness to invest in innovation and exploration of new products.

The findings provide a formal, data-driven model for a major tension in the digital economy, offering insights consistent with observed behavior in markets like Amazon and Google Play. The research has substantial implications for ongoing antitrust and regulatory efforts aimed at preserving market diversity and innovation, providing policymakers with a theoretical and computational tool to assess the effects of platform 'copycat' strategies.