Spatial Information Sharing on On-Demand Service Platforms (with Basak Kalkanci)
Abstract: We investigate how an on-demand service platform's mechanism to share demand--supply mismatch information spatially affects drivers' relocation decisions and the platform's matching efficiency. We consider three mechanisms motivated by practice: the platform either shares demand--supply mismatch information about zones(s) with excess demand with all drivers (surge information sharing, common practice today), all zones with all drivers (full information sharing), or zone(s) with excess demand only with drivers sufficiently close by (local information sharing). We develop a game-theoretic model with three zones wherein drivers in two non-surge zones decide whether to relocate to the surge zone with excess demand. We incorporate two spatial aspects: drivers' relocation costs and initial supply across different non-surge zones. Theoretically, full information sharing can hurt the platform's matching efficiency compared to surge information sharing under low relocation costs because drivers in non-surge zones, facing high demand locally do not chase the surge as much. Local information sharing is strictly dominated by other mechanisms in terms of matching efficiency when the supply of drivers near the surge zone is limited, and weakly dominated otherwise by surge information sharing. We test these theory predictions in the lab with human participants as drivers, in an environment where theoretical matching efficiency is highest with surge and lowest with local information sharing. Experimentally, the platform serves fewer customers than predicted with surge information sharing because drivers relocate too often, compromising efficiency in non-surge zones. In contrast, the platform serves more customers than predicted with full and local information sharing, and these mechanisms perform at least as well in matching efficiency as surge. Therefore, sharing demand--supply mismatch information either fully or in a targeted manner (as in local) can help to alleviate coordination problems on a platform. A behavioral equilibrium incorporating loss aversion through mental accounting and decision errors describes drivers' behavior in our experiments better than the rational equilibrium.
Background Readings (in no particular order and not exhaustive):