with Junhong Chu and Xueli Zhang
Abstract: Previous studies have shown that dispatch systems, which use innovative technology to match riders with taxis, can substantially reduce search frictions and are therefore more efficient than conventional street hails. However, how to design dispatch systems has not been fully investigated in the literature. Two dispatch systems are common in practice: a bidding system, in which drivers bid for a booking request, and an allocation system, in which drivers are allocated to riders by ride-hailing platforms based on algorithms. We developed and estimated a structural model for the ride-hailing market to compare the advantages and disadvantages of these two dispatch systems. On the demand side, we modeled riders to arrive stochastically in each location in each time period and make a discrete choice over the means of transportation, including hailing a taxi on the street, booking a taxi through the company’s booking system, and an outside option. On the supply side, we modeled taxi drivers’ decisions over a shift as a finite-horizon dynamic oligopoly game with a two-stage sequential decision in each period. In the first stage, drivers decide whether to pay an attention cost to enter the matching pool for booking trips; in the second stage, unmatched drivers make relocation decisions by deciding which location to head to in order to search for passengers in the next period. We applied our structural model to Singapore’s taxi market, where a leading ride-hailing company used a bidding system for booking trips during the data period. We estimated the structural model with detailed trip data and a parametric matching function that enabled us to back out unobserved demand for street-hail trips from the high-frequency, large-scale GPS data. Based on the structural estimates, we conducted counterfactual analyses by replacing the bidding system with an allocation system and made a comprehensive comparison across the two systems. We found that an allocation system could significantly improve drivers’ time utilization and earnings than a bidding system, but it would incentivize more drivers into the booking pool and reduce the number of drivers for street hails and hence increase riders’ unfulfilled demand for street-hail trips. Our results provide important managerial implications for the design of dispatch systems in the ride-hailing market.