Game Theory Explains How Algorithms Can Drive Up Prices

Summary

Researchers report that even simple pricing algorithms can lead to higher prices without any explicit collusion. Game-theory experiments show that learning algorithms—especially those designed to avoid certain kinds of “regret”—can settle into equilibria where prices stay high because any unilateral change would reduce a seller’s profit.

Crucially, the new work finds that a very nonresponsive strategy (one that chooses prices randomly from a skewed distribution) can exploit a common class of well-behaved algorithms. The result is an outcome that looks like collusion to buyers but involves no agreement or threats in the usual sense, making regulation tricky.

Key Points

Classic collusion (human firms agreeing to raise prices) is illegal, but algorithms can produce high prices without any explicit agreement.
Learning algorithms adjust prices over repeated rounds; in some experiments they learn to punish undercutting, creating high-price equilibria.
No-swap-regret algorithms—designed so no simple swap of actions would have been better—generally lead to competitive prices when facing copies of themselves.
A seemingly harmless nonresponsive strategy (randomising with high probability on high prices) can steer a no-swap-regret opponent into a high-price equilibrium and reap profits by occasional undercuts.
These equilibria can be stable: neither seller has an incentive to switch, so high prices persist even though no explicit collusion occurred.
Policy options are unclear: banning algorithms except no-swap-regret ones is one proposal, but it won’t solve all edge cases and could be hard to enforce in practice.

Context and relevance

This research sits at the intersection of economics, computer science and competition policy. As online marketplaces and dynamic pricing systems become more common, outcomes generated by interacting algorithms matter for consumers, regulators and platform operators. The work highlights a gap in antitrust frameworks that depend on finding an agreement; algorithmic outcomes can look collusive while leaving no legal smoking gun.

For industry, it signals that seemingly simple or “dumb” pricing heuristics can have outsized market effects. For regulators and platform designers, it emphasises the need for new tools: either to certify algorithm behaviour, limit certain strategies, or monitor market outcomes rather than intent.

Why should I read this?

Because it explains, in plain terms, how bots can tacitly make your shopping bill bigger — and why existing laws may not spot the trick. If you buy online, design pricing systems, or work on regulation, this short read saves you the time of digging into dense game theory and shows what actually goes wrong.

Author style

Punchy: this is important. The paper exposes a real blind spot — high prices can emerge without anybody “doing” anything illegal. If you care about fair markets, this is worth following closely.