Algorithmic Competition, with Humans

Abstract

I model algorithmic pricing as an automated rule mapping a rival’s price to the firm's own. I introduce managerial override, whereby managers can override these rules after they are chosen, which erodes algorithmic commitment. Even if overriding is costless, supracompetitive prices with "override-proof'" algorithms are an equilibrium outcome. The exact nature of override-proofness depends on equilibrium selection. If there are small override costs, then Bertrand pricing is never an equilibrium and prices are always supracompetitive. Finally, if override is costless and managers can respond to changes in demand or cost conditions in ways algorithms cannot, prices are generically Bertrand. This highlights the role of algorithmic commitment and prediction as substitutes in sustaining supracompeitive prices.

Keywords: Pricing algorithms, collusion, conduct, managers

JEL Classification: L40, L13, D43