Demand learning and dynamic pricing for multi-version products

We consider a capacity provider who offers multiple versions of a single product, such as different seat locations for an event. We assume that the different versions share an unknown core value and command a known premium or discount relative to the core value. Customers arrive at an unknown arrival rate during a finite sales horizon. We assume that the provider has a prior knowledge on the arrival rate which is updated using Bayesian rule. Estimates of the core value are updated using maximum likelihood estimation. We show how to simultaneously estimate the unknown parameters as the sales evolve and how to price the products to maximize revenues under a rolling horizon framework.