Decision models for e-hailing texi platforms

Abstract

Uber, Ola, Didi, and Lyft are examples of successful e-hailing taxi platforms. In this thesis, we aim to develop decision models for such platforms. In the rst essay, for an e-hailing taxi operation, we analyze the driver's pro tmaximizing reactive strategy (best evaluation criteria to either accept or refuse a ride request) in response to the ride request characteristics broadcasted by the platform and an associated driver penalty for refusing a ride request. We analyze six operating modes, which are a combination of three reactive strategies (no refusal, refusal based on proximity, and refusal based on pro tability index), and two broadcast methods (no matching, and matching). Also, we consider three types of service region topology: straight line, square, and circle. We provide structural properties of operating modes that show the similarities and di erences among the operating modes. We develop two models: an analytical model of a single taxi operation and an agent-based simulation model of multiple taxi operations. Using real data, we nd that a driver who follows refusal based on proximity strategy can earn approximately 25% more than the baseline - no refusal strategy. We also nd that the single taxi based analytical model is a good approximation of a more realistic problem. An e-hailing platform faces uncertainty from both the supply and demand side. Our second essay studies the platform's capacity decisions - the number of drivers to serve the demand - which a ects all the stakeholders (platform, driver, and customer). An e-hailing driver is characterized by ride capacity, ride acceptance rate, driver incentive, and absenteeism rate. Platforms provide driver incentives to maintain a high ride acceptance rate. We study capacity decision models under di erent scenarios: a single supply source, a dual source of reliable and unreliable drivers, driver absenteeism, presence of supply constraints, and competition. We nd that it is optimal to select taxi drivers exclusively from either unreliable or reliable drivers pool. The above nding holds even if the drivers exhibit absenteeism. However, when there are supply constraints imposed by regulators, the optimal capacity decision is to have a mix of unreliable and reliable drivers. We identify driver incentives as a more dominant lever for the platform as compared to ride acceptance rates and absenteeism rates. When two platforms engage in two-stage capacity-price competition for a random demand, we nd the platforms are asymmetric in their capacity sizes at optimality.