Experimentation, Biased Learning, and Conjectural Variations in Competitive Dynamic Pricing
New research reveals how simple A/B testing leads to inflated prices.
Deep Dive
A new paper shows that when competing sellers use simple A/B price experiments and basic learning algorithms, their actions can converge to a 'Conjectural Variations' equilibrium, leading to higher, supra-competitive prices. This happens due to a bias in demand learning caused by correlated experimentation schedules. The research provides a finite-sample guarantee of price error decaying at a rate of T^{-1/2}. Crucially, independent experimentation avoids this bias and leads to standard competitive pricing.
Why It Matters
This means automated pricing tools could inadvertently collude, raising costs for consumers without any illegal communication.