Outcome-Driven Dynamic Refugee Assignment with Allocation Balancing

Abstract

Dynamic Assignment Algorithms to Boost Refugee Outcomes Amid record-breaking forced displacement in recent years, researchers and policymakers alike have become increasingly interested in the idea of algorithmically matching refugees to geographic localities in order to optimize their employment or other integration outcomes. In “Outcome-Driven Dynamic Refugee Assignment with Allocation Balancing,” Bansak and Paulson propose new dynamic assignment algorithms for this context designed to maximize a given outcome while meeting the operational needs of refugee resettlement and asylum agencies. Using resettlement data from both the United States and Switzerland, they demonstrate how one algorithm (currently being piloted in Switzerland) can achieve near-optimal results compared with a hindsight-optimal matching. They also show that, because of nonstationarities in the arrival process, outcome maximization (even when subject to location capacity constraints) can result in an imbalanced allocation to localities over time, putting periodic strains on limited local resettlement resources. They account for this problem in a second algorithm that achieves near-perfect balance over time with only a small loss in average outcomes compared with the first algorithm.