Randomized Assignments for Barter Exchanges: Fairness vs. Efficiency

We study fairness and efficiency properties of randomized algorithms for barter exchanges with direct applications to kidney exchange problems. It is well documented that randomization can serve as a tool to ensure fairness among participants. However, in many applications, practical constraints often restrict the maximum allowed cyclelength of the exchange and for randomized algorithms, this imposes constraints of the cycle-length of every realized exchange in their decomposition. We prove that standard fairness properties such as envy-freeness or symmetry are incompatible with even the weakest notion of economic efficiency in this setting. On the plus side, we adapt some well-known matching mechanisms to incorporate the restricted cycle constraint and evaluate their performance experimentally on instances of the kidney exchange problem, showing tradeoffs between fairness and efficiency.