Essays on pricing and allocation of perishable goods

Abstract

In the first chapter, I examine how a monopoly service provider prices and allocates their multi-server queuing system to handle the large volume of computing jobs, jobs that differ along several dimensions. Modern network operators, be they cloud computing providers or telecommunication carriers, must carefully design their pricing and allocation plans in the face of the surging demand for their perishable good, namely, computing resources or wireless connections. Motivated by the market for cloud computing, we examine how a monopoly service provider prices and allocates its multi-server queuing system to handle the large volume of computing jobs, jobs that differ along several dimensions. To study the regularities in such a large and diverse system, we construct a service model with a large user base, one whose large numbers make the aggregate statistics of queuing behavior nearly deterministic. Even if the provider has resources that are more than sufficient to serve the entire market, variability in the users' willingness to pay and delay sensitivity means that segmenting the queue's capacity into different tranches to serve users with different preferences increases profits, giving rise to "strategic throttling". In parallel with the classical analyses of durable goods monopolies, the inability to perfectly discriminate between consumers means that the allocation is inefficient. To implement the optimal allocation, we propose an appealing spot market auction mechanism in which users find bidding a supplier's optimal prices to be a weakly dominant bidding strategy. In the second chapter, I develop a theoretical model for the optimal design of airfare as options under capacity constraint to study the price dispersion in U.S airline industry. Monopoly seller uses a menu of refund contracts to sequentially screen heterogeneous travellers who purchase the tickets before travel uncertainty is fully resolved. Airline finds it optimal to sell fully refundable fares to high-value customer when capacity is limited and to sell partially refundable fare when seats are abundant. The change of capacity accounts for the distinct jumps of non-refundable fare and the variation of refundable fare over booking periods. The model predicts that the price dispersion might fluctuate and then diminish as the number of available seat shrinks. An original data, which contains both high-frequent price and seat availability information for differentiated fares, provides strong empirical evidence that the price gap experiences stepwise drops before diminishing as departure date nears. In the final chapter, I study an optimal information design for efficient fleet redeployment in an urban area. Inefficiency in the form of mismatch between demand and supply could arise in the traditional taxi industry where drivers receive no additional information about market demand. Yet, providing complete public information, such as Uber surge map or Lyft prime time zone, to entire fleet might incur inefficiency as well ([17]). A market designer, who observes both the demand and supply level, could send private message to influence beliefs and thus actions of multiple drivers whose payoff depends on the state of demand level. We impose consistency condition requiring the signaled matching probability must be equal to the realized one for credible persuasion in the Bayesian persuasion game.