Dynamic Pricing with Unknown Nonparametric Demand and Limited Price Changes.

The article focuses on the dynamic pricing problem faced by retailers who must maximize cumulative revenue over a finite horizon while learning an unknown, nonparametric demand function and limiting costly price changes. It introduces the Stochastic Limited Price Experimentation (SLPE) policy, which uses second-order approximations of the demand function and leverages a two-point bandit feedback assumption to achieve a near-optimal regret rate of order \(\tilde{O}(\sqrt{T})\) while reducing the total number of price changes to order \(O(\log \log T)\), improving upon previous methods that required \(O(\log T)\) price changes. The paper also relaxes the two-point bandit feedback assumption, showing that similar performance guarantees hold under weaker conditions, and provides extensive numerical experiments demonstrating that SLPE substantially reduces price changes and price experimentation compared to benchmark algorithms, with comparable regret performance. This work addresses practical concerns in retail pricing by balancing exploration-exploitation tradeoffs without parametric demand assumptions and operational constraints on price adjustments.