Random Serial Dictatorship (RSD)

Updated 3 February 2026

Random Serial Dictatorship (RSD) is a randomized mechanism that assigns indivisible objects to agents based on a uniform random ordering of their strict preferences.
It guarantees key properties such as ex-post efficiency, equal treatment of equals, and strategyproofness, making it a robust choice in allocation problems.
Extensions like quota assignments and algorithmic approximations address computational challenges, though RSD may not achieve SD-efficiency or ordinal efficiency in all settings.

Random Serial Dictatorship (RSD) is a canonical randomized mechanism for allocating indivisible objects among agents with strict ordinal or lexicographic preferences. RSD operates by drawing a uniform random permutation of agents and, in sequence, allowing each to select her most-preferred available object. The mechanism is extensively studied in economics, computer science, and social choice, notable for its strong incentive properties and axiomatic simplicity. In lexicographic domains and under quota assignments, RSD admits further generalizations with salient envyfreeness and strategyproofness guarantees.

1. Formal Definition and Mechanism Design

Let $N=\{1,\dots,n\}$ be the set of agents, $M=\{1,\dots,m\}$ the set of indivisible objects, and each agent $i$ has a strict preference order $\succ_i$ over $M$ or bundles thereof. For the classic assignment problem, each agent is to receive exactly one object. Random Serial Dictatorship (RSD) is defined formally as follows:

Sample a permutation $\sigma \in S_n$ uniformly at random.
For $k = 1$ $k = 1$ to $n$ $n$ :
- Agent $\sigma(k)$ picks her most-preferred available object according to $\succ_{\sigma(k)}$ .
- Remove that object from the pool.
The outcome is the deterministic matching resulting from the chosen order; RSD is the uniform lottery over all $n!$ such deterministic assignments (Hosseini et al., 2015, Hosseini et al., 2015, Zaquen et al., 1 Feb 2026).

For quota problems, where agents may receive $q_i \ge 1$ objects each, the mechanism generalizes: the serial dictatorship quota mechanism (SDQ) permits each agent, in the order, to select her quota of objects, with RSDQ denoting the uniform lottery over all orderings and quota partitions (Hosseini et al., 2015).

2. Axiomatic Foundations and Characterizations

RSD is distinguished by three core axioms in the assignment context with strict preferences:

Ex-Post Efficiency (EPE): No deterministic assignment in the support is Pareto-dominated (Basteck, 22 Jun 2025, Brandt et al., 2023).
Equal Treatment of Equals (ETE): Agents with identical preferences must receive identical marginal distributions.
Strategyproofness (SP): Truthful reporting is a dominant strategy; no misreport yields a stochastically superior allocation in the agent's ordering.

Recent work fully characterizes the parameter domain in which these three axioms uniquely specify RSD: except for the “small” cases $(n,m)\in\{(n,1),(n,2),(3,m),(4,4)\}$ , RSD is the unique assignment mechanism satisfying EPE, ETE, and SP. In the exceptional cases, there exist explicit non-RSD rules satisfying all three axioms; even additional requirements such as bounded invariance, non-bossiness, or consistency do not restore uniqueness (Zaquen et al., 1 Feb 2026, Brandt et al., 2023). For randomized social choice, Probabilistic Maskin-monotonicity strengthens strategyproofness and, combined with EPE and ETE, characterizes RSD on the universal domain (Basteck, 22 Jun 2025).

3. Economic Properties: Efficiency, Fairness, and Incentives

Ex-Post Efficiency

RSD is always ex-post efficient by construction: each deterministic support is a serial dictatorship, and thus Pareto-optimal (Hosseini et al., 2015, Mennle et al., 2013).

Strategyproofness

RSD satisfies strong strategyproofness (sd-strategyproofness): for any agent and any misreport, the cumulative probability of receiving objects weakly preferred to any fixed object never increases. This results directly from the independence of the random order and the agent's position therein; truthful reporting maximizes available choices (Hosseini et al., 2015, Mennle et al., 2013, Hosseini et al., 2015).

Envyfreeness under Lexicographic Preferences

Under downward lexicographic (ld) preferences, RSD is envyfree: for all agent pairs, if the probabilities for all strictly more-preferred objects coincide, then no agent strictly prefers another’s lottery for any object (Hosseini et al., 2015, Hosseini et al., 2015). The proof leverages a coupling over paired orderings—swapping two agents and invoking lexicographic dominance ensures no agent envies another (Hosseini et al., 2015).

Limitations: SD-Efficiency and Ordinal Efficiency

RSD is not, in general, SD-efficient—there exist profiles and ex-post efficient lotteries (including RSD itself) that admit strict stochastic dominance improvements via trading cycles (Aziz, 2016, Hosseini et al., 2015). Probabilistic Serial (PS) achieves SD-efficiency but not strategyproofness. RSD is also not ordinally efficient, meaning it may be stochastically dominated by other assignments in the first-order sense (Mennle et al., 2013, Hosseini et al., 2015).

Table: Core Properties of RSD

Property	General Preferences	Lexicographic Preferences
Ex-post efficient	Yes	Yes
Strategyproof	Yes	Yes
Envyfree	No	Yes
SD-efficient	No	—
Ordinally efficient	No	—

4. Computational Complexity and Algorithmic Techniques

Intractability

Computing marginal probabilities or the full output distribution of RSD is #P-complete for both voting and assignment settings (Mennle et al., 2013, Aziz et al., 2014). The reduction is via counting linear extensions in posets (voting) or perfect matchings in bipartite graphs (assignment).

Parametric and Approximate Algorithms

For settings with bounded agent or alternative types, fixed-parameter tractable algorithms with dynamic programming are available, yielding exact computation in $O(f(\text{parameter})\cdot\text{poly}(|R|))$ time (Aziz et al., 2014). For large-scale problems, approximate evaluation via random sampling of permutations delivers accurate estimates: with $m=O(n^2/\varepsilon^2\log(2/\delta))$ samples, the Hoeffding bound guarantees an additive $\varepsilon$ -approximation of expected welfare or cost (Caragiannis et al., 2024).

5. Welfare Analysis and Performance Bounds

Dichotomous preferences: RSD guarantees at least $1/3$ of the maximum offline welfare; up to $0.69$ by leveraging non-adversarial agent behavior and online bipartite matching (Adamczyk et al., 2014).
Normalized von Neumann–Morgenstern valuations: $E[\text{SW(RSD)}] \geq \frac{1}{e} \frac{\nu(\text{opt})^2}{n}$ (Adamczyk et al., 2014).
Egalitarian welfare: RSD guarantees proportionality ($1/n$) in worst case, and achieves close-to-optimal performance for typical utility distributions (Aziz et al., 2015).
Ordinal welfare factor: Asymptotically $1/2$; no mechanism exceeds this in the worst case (Bhalgat et al., 2011).

Dominance Relations and Comparisons

RSD and PS are nearly incomparable for $n=m$ , with stochastic dominance of PS over RSD occurring on a vanishing fraction of preference profiles as $n$ or $m$ increases. In practice, welfare and fairness are similar in balanced markets, with RSD outperforming on strategyproofness and robustness to manipulation (Hosseini et al., 2017, Hosseini et al., 2015).

6. Extensions, Robustness, and Application Contexts

Generalizations: Quota Settings and Externalities

Random Serial Dictatorship with Quotas (RSDQ): RSDQ retains strategyproofness, ex-post efficiency, and envyfreeness under lexicographic preferences for arbitrary quota systems and any agent/object configuration (Hosseini et al., 2015).
Externalities and Land Allocation with Friendship: Variants such as Online-Choose-Together RSD and Online-Choose-Adjacent RSD ensure universality of Pareto optimality and truthfulness, with explicit social welfare guarantees parameterized by friendship value $\alpha$ (Elkind et al., 2020).

Mechanisms with Transfers

When post-assignment transfers are allowed, Pareto efficiency and strategyproofness may be lost. Ex-post equilibrium trading can restore efficiency, but only with centralized clearing and low transaction costs, at the expense of strategic simplicity (Sundar et al., 2023).

Stability and Scalability

In large markets (school choice, housing), RSD cutoffs concentrate sharply around deterministic mean-field limits provided $m\ln m\ll n$ ; this validates the use of cutoff statistics for empirical analysis and mechanism design, even under adversarial preferences (Vijaykumar, 2021).

7. Domain-Specific Efficiency and Open Directions

RSD is ex-ante efficient on domains with strict or single-peaked preferences and most small-scale dichotomous environments. However, in general domains ex-post efficiency does not imply ex-ante efficiency. Precise geometric and linear-algebraic criteria (e.g., welfare-equalizing representation, rank test) determine when the notions coincide (Echenique et al., 2022, Brandt et al., 2023).

Recent axiomatization work formalizes RSD via Probabilistic Maskin-monotonicity, providing uniqueness among random assignment rules on the universal domain of strict preferences (Basteck, 22 Jun 2025). Several characterization problems remain open for large $n$ due to the exponential growth of matrix rank conditions.

For proofs, algorithms, and detailed treatment, see (Hosseini et al., 2015) for RSDQ characterization and lexicographic envyfreeness, (Adamczyk et al., 2014) for welfare analysis, (Zaquen et al., 1 Feb 2026) for full domain axiomatization, (Mennle et al., 2013) for complexity, (Aziz et al., 2014) for parameterized algorithms, and (Echenique et al., 2022) for efficiency theory in collective choice.