There are many cases in which resources needs to be allocated over time and without resorting to monetary payments; for example, consider allocating cluster time to employees, or food to foodbanks. A popular approach is to distribute an artificial currency between the participants which they can use for bidding just as they would money. But how well does the use of artificial currency translate the guarantees of classical mechanism design, and when should we expect artificial currencies to help with efficient allocation?

In a series of papers with Sid Banerjee and Krishnamurthy Iyer (see below), we show in a general allocation setting that one can take any monetary mechanism and repeatedly run it using an artificial currency (with some additional precautions) to approximately realize the incentives and efficiency of the original mechanism. This result is formalized as a blackbox reduction that can be applied to any Baysian Incentive Compatible (BIC) mechanism.

My current work builds on this in several directions. First, in an effort to make these results more applicable, I am working on establishing similar guarantees under more frugal assumptions on what information is available to the designer. In another project, I am investigating whether the Nash Social Welfare, an objective that has attracted recent attention as a good trade-off between efficiency and fairness, can be approximated in dynamic non-monetary settings. Finally, I am working on generalizing our results to include exchange economies.

The majority of results in mechanism design focus on the property of individual strategy-proofness, but when applying these results in practice, coalitional deviations are also a major concern. The situation is particularly disconcerting when coalitions are allowed to exchange monetary payments (`bribes'), as almost no positive results exist under this assumption. A promising direction here is the recently-introduced notion of collusion-resilience, which instead of striving for truthful revelation, requires only that all (potentially unknown) coalitions have dominant strategies, and that the resulting outcome is efficient.

In joint work with my advisors, we show that collusion-resilience is achievable for a rather general class of settings, that, in particular, includes combinatorial auctions under GS preferences, and explicitely construct such a mechanism. We also argue that it is in fact a near-boundary for when this property is achivable, showing impossibilities for more general combinatorial auctions and collective decision-making problems.