Department seminar by Yongyi Guo

Please Note: This seminar will be given online.

A statistical approach to feature-based dynamic pricing

Dynamic pricing is one of the most common examples of online learning and decision problems with continuous action space. With the development of e-commerce and the massive real-time data in online platforms today, feature-based (or contextual) pricing models has become increasingly important. In this work, we study the feature-based dynamic pricing problem where the market value of a product is linear in its observed features plus market noise. Moreover, since market noise cannot be observed, we assume that the market noise density falls into a general (non-parametric) class. We propose a dynamic statistical learning and decision making policy that minimizes regret by combining online decision making and semi-parametric statistical estimation from a generalized linear model with an unknown link. Specifically, we provide non-asymptotic uniform error bounds for kernel type regression estimators, which enable us to control the regret while learning the model efficiently. Under mild conditions, our proposed algorithm achieves near optimal regret at the same order as the lower bound when the market noise distribution is parametric ($\Omega(\sqrt{T})$). The performance of the algorithm is also demonstrated through intensive simulations.