Abstract
Ride-sharing platforms such as Lyft and Uber are among the fastest growing online marketplaces. A key feature of these platforms is the implementation of fine-grained, fast-timescale dynamic pricing --- where prices can react to instantaneous system state, and across very small geographic areas. In this talk, we explore the value of such fast-timescale dynamic pricing. To do so, we develop a new queueing model for ride-sharing platforms, which combines the stochastic dynamics of the platform's operations with strategic models of both passenger and driver behavior. Using this model, we demonstrate that dynamic pricing may not be better than the optimal quasi-static price in most settings. However, finding the optimal static price requires exact knowledge of system parameters; we show that dynamic pricing is much more robust to fluctuations in these parameters as compared to static prices. Thus, dynamic pricing does not necessarily buy more than static pricing, but in fact, it lets platforms realize the benefits of optimal static pricing with imperfect knowledge of system parameters.
Joint work with Siddhartha Banerjee (Cornell) and Carlos Riquelme (Stanford).