Mechanism Design for Coordinating Behavior

Abstract

The field of mechanism design studies incentive-aligned systems that aggregate the preference of individual participants and implement desired social outcomes, despite agents being strategic and acting to optimize their own interests. The theory and practice of mechanism design still falls short, however, for settings where the designer not only cares about the efficiency of the allocation of resources or tasks, but also has an interest in participants taking intended, downstream actions. Consider ridesharing platforms, where drivers' strategic behavior (e.g., cherry-picking trips, declining trips to chase surge prices or wait for price hikes) undercuts reliability. Consider also the problem of allocated resources going to waste: it is common for sport facilities to be fully booked, and yet relatively under used on any given day due to no-show's. Similar problems exist for demand response programs, which are helpful for the power system only if the consumers actually reduce electricity consumption in a reliable manner, during times when demand exceeds supply. In this thesis, I present my work on mechanism design for these settings, where there is the need to coordinate agents' behavior by shaping the downstream decision making environments. For ridesharing, I propose the Spatio-Temporal Pricing mechanism, which achieves envy-freeness, robustness, and aligns incentives for drivers without using penalties or time-extended contracts. This provides both reliability for riders and flexibility for drivers. For settings of shared resources or tasks, I design simple and indirect mechanisms based on novel penalty-bidding schemes that achieve optimal utilization for resource assignment, or guarantee reliability for demand response. The penalty bids provide signals on people's reliability, and the actual penalties charged align incentives and favor the intended actions. This also uncovers an interesting problem of social choice problems with non quasi-linear utilities, for which I provide a tight characterization of the maximal non quasi-linear utility domain for which "reasonable" social choice mechanisms (with VCG-like properties) continue to exist.