Dynamic duopoly with learning through market experimentation

GAICIL Insights

Sometimes the smartest move in a market is to stay ignorant.¹ Dynamic duopoly with learning through market experimentation, by Philippe Aghion, Maria Paz Espinosa and Bruno Jullien, shows how learning behavior can substantially modify the outcome of competition in an oligopolistic industry facing demand uncertainty.

The setup is a symmetric duopoly in which firms have imperfect information about market demand and learn by observing their sales volume. Prices do double duty: they generate current profits and produce information about demand. Firms may set prices partly to learn, a behavior known as market experimentation. Every price tag is also a probe.

The main result explains price dispersion. Even symmetric firms facing the same uncertainty may choose different prices, because experimentation sends them along different paths. The authors trace this and its dynamic evolution in a Hotelling duopoly, then extend the analysis to general demand functions and to oligopolies with many firms. Identical firms, different prices, no mystery required.

Information about demand is partly a public good. When one firm experiments, its rival may learn from the resulting market outcomes too. That invites free-riding on others' experimentation and may produce less experimentation than would be socially desirable. Why pay to learn what a rival will reveal for free?

The twist is that information can have negative value. Price competition raises the possibility that more demand information harms both firms. The authors construct an example where competition becomes tougher on average once firms learn the relevant demand parameter. Firms may then prefer to remain uninformed, because learning intensifies competition. Knowledge can be a weapon both sides would rather not hold.

Artificial intelligence transforms this game. AI dramatically improves firms' ability to learn about demand from data on sales, prices and customer behavior. Experimentation that once required costly price variation can now run rapidly and at scale. The paper's framework shows what happens when the probe gets cheap.

Spillovers sharpen. When demand information becomes widely available, every firm benefits, weakening any single firm's incentive to experiment. AI tools that process market signals may amplify these spillovers, changing the pace of learning across an industry. Cheap learning by one can become learning for all.

The negative value of information is good news for buyers. If better demand information intensifies competition, it can lower prices and benefit consumers. AI that sharpens firms' knowledge of demand may toughen competition even when each firm alone would prefer less information. The firms might wish the data away; consumers would not.

Dispersion becomes dynamic. Experimentation produces different prices across firms, and with AI enabling continuous experimentation, prices may vary more dynamically. Understanding these dynamics helps assess how AI-driven pricing affects markets and consumers. Price variation can be the visible trace of learning in progress.

Translated into policy, the framework anchors analysis of how learning shapes competition. As AI changes how firms acquire and use market information, effects on prices, dispersion and competitive intensity depend on the strategic considerations the paper identifies. Careful analysis can separate learning that intensifies competition from conduct requiring closer scrutiny.

The modeling path, from a Hotelling duopoly to general demand and many-firm oligopolies, shows the main insights are robust. Learning through experimentation is a general feature of markets under uncertainty.

Seen whole, the paper's message is plain: learning behavior substantially modifies competitive outcomes. Experimentation explains price dispersion, information has public good features, and more information can intensify competition. As AI transforms how firms learn about markets, these dynamics can benefit consumers and spur more efficient outcomes.

Here is the uncomfortable test for those worried that AI pricing tools will soften competition: the paper offers a scenario where the firms themselves would prefer to stay in the dark, because knowing demand precisely makes rivalry tougher. Before treating better information as a threat, ask who it actually hurts.