Pricing algorithms that ignore rivals don't sustain collusion, Wu & Zeevi find

In a new paper accepted at EC 2026, Yuhang Wu and Assaf Zeevi tackle a hot-button question in algorithmic pricing: should demand models incorporate competitor prices? The authors model a stylized market where multiple sellers repeatedly set prices, using iterated least squares to learn demand curves. Some sellers choose an “oblivious” strategy—deliberately ignoring competitors’ prices in their model—while others are “informed” and include that data. The conventional wisdom from classical learning theory says obliviousness causes misspecification and inefficiency, but recent work on algorithmic collusion suggested it might foster cooperative pricing.

Wu and Zeevi show that this is not robust. Oblivious sellers must engage in more aggressive price exploration to compensate for the missing dynamic competitor information—a cost that informed sellers avoid. When all sellers are oblivious, prices can exhibit an “excursion phenomenon” with transient collusive spikes, but these fade as learning progresses. In mixed markets, informed sellers consistently out-earn the oblivious. Plugged into a game-theoretic lens, the modeling choice has a unique Nash equilibrium: an all-informed market, which efficiently drives prices to competitive outcomes. The takeaway is clear—incorporating competitor information plus sufficient price exploration is a reliable, pro-competitive strategy for sellers, and worries about algorithmic collusion via oblivious learning are overstated.