Essays on spatial and behavioral analytics for platform design

Abstract

The design and operation of a two-sided platform require a variety of decisions to facilitate a match between sellers (capacity) and buyers (demand). Many platforms deploy analytic capabilities to leverage rich information, on both demand and capacity, that is available in real-time. This dissertation research explores design decisions, such as price structure and quality controls, and allied analytic capabilities in order to document their impact on platform governance. These decisions are tested in the context of ride-sharing platform by positing three fundamental challenges that must be accounted for effective design: (1) spatial distribution of capacity and demand that allows for capacity spillovers, (2) buyer’s sentiment biases, and (3) seller’s relocation biases. These challenges are assessed in three separate but related essays. The first essay investigates how the policies for setting surge prices should be designed under capacity spillovers. Using a data set from Uber’s operations, we estimate a spatial panel model to reveal its surge pricing structure that accounts for spatial dependency. Allied counterfactual analysis illustrates the limitations of a spot pricing policy (i.e., a policy that does not account for spillovers). The second essay assesses the impact of buyer’s sentiment bias, ranging from optimism to pessimism, on the platform’s decision to control seller quality. Platforms face a trade-off between ensuring high-quality sellers and guaranteeing enough sellers such that wait-time is lowered. We formally characterize an optimal exclusion threshold on seller quality in the presence of sentiment bias. We also examine strategies that a platform can access to benefit from buyer’s behavioral biases. Results document the impact of seller quality on a platform’s profitability and social welfare. In the last essay, we focus on the seller’s relocation behavior. There is a debate in the literature on whether sellers’ willingness to relocate across demand zones in order to chase surging prices is rewarded in a ride-sharing platform setting. Using multiple machine learning algorithms, we classify rewarding behaviors with different pricing structures under a variety of circumstances. Results provide guidance on how to provide incentives while managing the dynamics of spatially distributed capacity.