Optimal Mechansim Design and Money Burning

Mechanism design is now a standard tool in computer science for aligning the incentives of self-interested agents with the objectives of a system designer. There is, however, a fundamental disconnect between the traditional application domains of mechanism design (such as auctions) and those arising in computer science (such as networks): while monetary transfers (i.e., payments) are essential for most of the known positive results in mechanism design, they are undesirable or even technologically infeasible in many computer systems. Classical impossibility results imply that the reach of mechanisms without transfers is severely limited. Computer systems typically do have the ability to reduce service quality–routing systems can drop or delay traffic, scheduling protocols can delay the release of jobs, and computational payment schemes can require computational payments from users (e.g., in spam-fighting systems). Service degradation is tantamount to requiring that users burn money}, and such ``payments’’ can be used to influence the preferences of the agents at a cost of degrading the social surplus. We develop a framework for the design and analysis of money-burning mechanisms to maximize the residual surplus–the total value of the chosen outcome minus the payments required.

💡 Research Summary

The paper tackles the problem of mechanism design in environments where monetary transfers are either undesirable, infeasible, or technologically impossible—situations common in many computer systems such as networks, scheduling platforms, and spam‑filtering services. Instead of using money, the authors propose “money‑burning” mechanisms, where the system degrades service quality (e.g., delays traffic, postpones job execution) and treats the resulting loss of utility as a payment that is permanently “burned.” The objective is to maximize residual surplus, defined as the total value derived from the chosen allocation minus the sum of these burned payments.

The authors first develop a general template for prior‑free (worst‑case) optimal mechanism design. The template consists of four steps: (1) characterize the Bayesian‑optimal mechanism for every i.i.d. valuation distribution; (2) interpret these mechanisms to obtain a distribution‑independent benchmark that any prior‑free mechanism should compete against; (3) design a single mechanism that approximates this benchmark on every valuation profile; and (4) prove lower bounds by exhibiting distributions where the benchmark is far from the performance of any Bayesian‑optimal mechanism. This template bridges the dominant‑strategy approach common in economics with the worst‑case analysis prevalent in theoretical computer science.

In the Bayesian setting, the paper provides a complete characterization of optimal money‑burning mechanisms for single‑parameter agents. The agents’ private valuations are drawn from a known distribution (not necessarily identical across agents). By extending Myerson’s virtual‑value framework, the authors show that the optimal mechanism is incentive compatible, individually rational, and can be expressed via a monotone allocation rule together with payments that exactly equal the agents’ virtual surplus. This result subsumes earlier work that was limited to multi‑unit auctions with monotone hazard rates, and it also handles asymmetric settings such as agents seeking disjoint paths in a network.

Using this characterization, the authors define a prior‑free benchmark: for any valuation profile, the benchmark value is the expected residual surplus of the Bayesian‑optimal mechanism when the valuations are drawn from the worst‑case i.i.d. distribution that makes this expectation smallest. This benchmark is distribution‑free yet captures the best possible performance achievable by any mechanism that knows the distribution.

For multi‑unit auctions (k identical items), the paper designs a simple, near‑optimal prior‑free mechanism called the k‑unit p‑lottery. Agents are randomly ordered; each is offered a take‑it‑or‑leave‑it deal with probability p of receiving an item, and the process stops after k items are allocated or all agents have been considered. By selecting p via a grand‑sampling technique, the mechanism guarantees a constant‑factor approximation to the prior‑free benchmark for every valuation profile, even when k=1. This result is notable because it avoids the “two‑or‑more winners” requirement that appears in many digital‑goods auctions.

Finally, the paper quantifies the power gap between mechanisms that can use arbitrary monetary transfers and those restricted to money‑burning. It proves that the worst‑case loss in social surplus due to burning payments is at most logarithmic in the number of participants (both in Bayesian and prior‑free settings). In other words, a money‑burning mechanism can achieve residual surplus within an O(log n) factor of the full surplus attainable by Vickrey‑Clarke‑Groves (VCG) mechanisms. This logarithmic bound is tight, and it dramatically improves upon the linear‑scale loss known for mechanisms without any payments.

Overall, the contributions are: (1) a full characterization of Bayesian‑optimal money‑burning mechanisms for single‑parameter agents; (2) a principled prior‑free benchmark derived from these optimal mechanisms; (3) a constant‑approximation prior‑free mechanism for multi‑unit auctions; and (4) a tight logarithmic bound on the efficiency loss when only money‑burning is allowed. These results open a rigorous pathway for designing efficient, incentive‑compatible protocols in systems where traditional monetary transfers are unavailable, by leveraging controlled service degradation as a “payment” instrument.