Sample-Based Optimal Pricing

Friday, October 5, 2018 - 11:45am - 12:30pm
Keller 3-180
Omar Besbes (Columbia University)
Pricing is central to many industries and academic disciplines ranging from Operations Research to Computer Science and Economics. In this talk, we study data-driven optimal pricing in low informational environments. We analyze how a decision-maker should price based on a single sample of the willingness-to-pay (WTP) of customers. The decision-maker's objective is to select a general pricing policy with maximum competitive ratio when the WTP distribution is only known to belong to some broad set. We characterize optimal performance across a spectrum of non-parametric families of distributions, $\lambda$-regular distributions, two notable special cases being regular and monotone hazard rate distributions. We develop a general approach to obtain parametric lower bounds on the maximin ratio as well parametric upper bounds. The results have implications on the value of samples in absolute terms but also viz., e.g., increased customer competition. (joint work with A. Allouah)