Top-Trading-Cycles Mechanism (TTC)

• Top-Trading-Cycles (TTC) is a cycle-based allocation mechanism that reallocates indivisible objects to achieve Pareto efficiency, individual rationality, and strategy-proof outcomes.
• It operates by having agents point to their top remaining object, forming and executing directed cycles that yield the unique core allocation in strict preference environments.
• Extensions of TTC apply to school choice, multi-center, networked, and fractional assignment settings, demonstrating its robust axiomatic and computational adaptability.

Top-Trading-Cycles (TTC) is a family of allocation mechanisms for reallocating indivisible objects through directed cycles of mutually compatible claims. In its canonical form, introduced for the Shapley–Scarf housing market, each agent owns one object, points to the most-preferred remaining object, and each object points to its current owner; every directed cycle is then executed, assigned agents and objects are removed, and the process repeats until exhaustion. In that baseline environment with strict preferences and one-to-one endowments, TTC is individually rational, Pareto efficient, strategy-proof, and selects the unique core allocation (Feng, 2023). Subsequent work has recast TTC for school choice, multi-center allocation, multiple-type housing markets, networked exchange, randomized assignment, and post-deviation reassignment, while also sharpening its axiomatic, informational, and computational structure (Cheng et al., 26 Sep 2025).

1. Classical model and cycle algorithm

In the standard housing-market formulation, there is a finite set of agents and a finite set of indivisible objects, with each agent initially endowed with exactly one object and each agent holding a strict preference ordering over all objects. An allocation is a bijection from agents to objects. TTC proceeds iteratively on the remaining agents and remaining objects. At each round, every remaining agent points to the most-preferred remaining object, and every remaining object points to its current owner. Because the graph is finite and every vertex has out-degree one, at least one directed cycle exists. Every agent in a cycle receives the object she points to, and all agents and objects in those cycles are removed. The algorithm terminates after finitely many rounds (Feng, 2023).

This cycle semantics is the mechanism’s defining feature. Each executed cycle identifies a self-contained exchange in which every participating agent receives her current top feasible object, given the remaining market. In the one-type housing market, this structure is equivalent to the standard Shapley–Scarf trading interpretation. In school choice, the same graph-theoretic logic is used after replacing ownership by school priorities; in multi-center allocation, objects inherit their center’s priority order and point to the highest-priority remaining member of that center (Ortega et al., 2022).

A useful way to distinguish TTC from other matching procedures is that TTC is fundamentally a trading mechanism rather than an acceptance-rejection mechanism. Deferred Acceptance (DA) is organized around tentative admissions and rejections, whereas TTC is organized around simultaneous execution of cycles. This distinction is structurally important for both welfare and axiomatic analysis (Ortega et al., 2022).

2. Core properties in housing markets

In the classical Shapley–Scarf housing market with strict preferences, TTC is individually rational: every agent weakly prefers the assigned object to her initial endowment. It is also Pareto efficient: no other feasible allocation can make every agent weakly better off and at least one agent strictly better off. Classical results further imply that TTC selects the unique core allocation, so no coalition can reallocate its endowed objects among its members to make all of them weakly better off and one strictly better off (Feng, 2023).

Strategy-proofness is equally central in the baseline model. Truthful reporting is a dominant strategy: no agent can benefit by misreporting her preference order. This property underlies TTC’s canonical status in one-sided matching with private endowments and strict preferences (Feng, 2023). A recurrent theme in later work is that once the environment is generalized—multiple endowments, multiple object types, priorities, capacities, or restricted visibility—this clean package of properties no longer transfers automatically (Phan et al., 2018).

A substantial axiomatic literature shows that TTC is not merely one mechanism among many satisfying these properties but, in important domains, the unique rule satisfying specified combinations of them. On the unrestricted housing-market domain, TTC is uniquely determined by individual rationality, pair efficiency, and strategy-proofness, where pair efficiency rules out a pair of agents who both strictly prefer the other’s assignment (Goel et al., 26 Jan 2025). On the same domain, TTC is also characterized by individual rationality, Pareto efficiency, and truncation-proofness, and by individual rationality, pair-efficiency, and truncation-invariance, where truncation-based axioms weaken full strategy-proofness by restricting the relevant manipulation class (Chen et al., 2021).

The distinction between pair efficiency and Pareto efficiency matters. Pareto efficiency implies pair efficiency, but the converse need not hold in general. The characterization results show that, in the relevant domains, the weaker pair-efficiency condition can still be sufficient once combined with appropriate incentive constraints (Goel et al., 26 Jan 2025).

3. Domain restrictions and axiomatic structure

Restricted preference domains play a major role in determining whether TTC remains uniquely characterized by familiar axioms. One recent criterion is the top-two condition: within any subset of objects, if two objects can each be most-preferred, then they can also appear as the top two most-preferred objects in both possible orders. On domains satisfying this condition, the characterization by individual rationality, pair efficiency, and strategy-proofness continues to identify TTC uniquely (Goel et al., 26 Jan 2025).

This criterion unifies several previously separate domain-specific results. The data explicitly states that the condition strengthens and unifies earlier findings on single-peaked and single-dipped domains. It also shows necessity in the special case of three objects: when the domain fails the top-two condition with three objects, TTC need not be uniquely characterized by those axioms (Goel et al., 26 Jan 2025).

The axiomatic picture becomes more intricate in environments with multiple objects per agent. In multiple-object reallocation problems with lexicographic preferences, TTC is uniquely determined by balancedness, Pareto efficiency, the weak endowment lower bound, and either truncation-proofness or drop strategy-proofness. On the more general responsive domain, TTC remains an important benchmark but is only uniquely characterized within the class of individual-good-based rules, and only under individual-good efficiency rather than full Pareto efficiency (Coreno et al., 2024). This indicates that the classical housing-market package cannot simply be transplanted wholesale to richer bundle environments.

A related development concerns multiple-type housing markets. Full Pareto efficiency is incompatible with individual rationality and strategy-proofness when agents consume bundles across types. This incompatibility motivates weaker efficiency notions such as coordinatewise efficiency and pairwise efficiency. These, in turn, characterize two TTC extensions: coordinatewise TTC (cTTC) and bundle TTC (bTTC). The results can be read as a compatibility test: efficiency notions not satisfied by cTTC or bTTC may be incompatible with individual rationality and strategy-proofness in these environments (Feng, 2023).

Another line of work studies limited externalities. In housing markets where agents care both about the house they receive and about who receives their endowment, TTC continues to be characterized by individual rationality, pair efficiency, and strategy-proofness on the demand-lexicographic domain. But once the domain is expanded to include both demand-lexicographic and supply-lexicographic preferences, no rule satisfies individual rationality, pair efficiency, and strategy-proofness simultaneously (Klaus, 2024). This sharply marks the boundary of TTC’s classical logic when ownership externalities are introduced.

4. Priority-based TTC in school choice and multi-center allocation

In school choice, TTC is adapted by replacing ownership with school priorities. In the one-to-one priority-based formulation, each student points to her most-preferred remaining school, and each school points to its highest-priority remaining student; cycles are then executed and removed exactly as in housing markets. TTC is strategy-proof and Pareto-optimal for students, but unlike DA it is not stable in general and may generate justified envy (Ortega et al., 2022).

This instability is not a minor technicality. In school choice, DA is stable and yields no justified envy, whereas TTC trades away stability in exchange for Pareto optimality. The paper on the cost of strategy-proofness quantifies the resulting trade-offs. In random one-to-one markets, TTC has expected average rank asymptotically proportional to $\log n$, worst-off rank asymptotically larger than $0.5n$, and a positive asymptotic fraction of students with justified envy equal to $0.3863$. In Budapest admissions data, TTC allocates more first choices than DA, but its worst-off placement is much worse than that of the rank-minimizing mechanism and it still produces substantial justified envy (Ortega et al., 2022).

At the same time, TTC acquires additional strategic interpretations in school choice. The student-optimal TTC mechanism is characterized by MBG-quota-rationality and MBG-robust efficiency, where MBGs are mutual best groups defined through the interaction of favorite-school sets and top-priority students. This characterization provides a comparison point against student-optimal stable mechanisms, especially under Ergin-acyclic and Kesten-acyclic priority structures (Chen et al., 2021).

The school-choice literature also connects TTC to farsighted stability. With farsighted students and myopic schools, the matching produced by TTC forms a singleton farsighted stable set, whereas the DA outcome may fail to belong to any farsighted stable set. Moreover, looking forward three steps is already sufficient to stabilize the TTC matching (Atay et al., 2022). This does not overturn TTC’s classical lack of stability in the justified-envy sense; rather, it shows that under a different stability concept based on farsighted deviations, TTC has a distinct robustness advantage.

In multi-center allocation, TTC is extended to settings with multiple centers, each endowed with objects and strict within-center priorities but no cross-center priorities. Here TTC is defined by assigning each remaining object the priority order of its center and running the agent–object cycle algorithm accordingly. The resulting rule is strategy-proof and pair efficient, and it satisfies three fairness notions introduced for this environment: internal fairness, external fairness, and procedural fairness (Cheng et al., 26 Sep 2025).

Those fairness notions yield several sharp characterizations. Strategy-proofness, pair efficiency, internal fairness, and external fairness characterize TTC. Strategy-proofness together with procedural fairness alone also characterizes TTC. A further result replaces queuewise rationality by the weaker center lower bound and again recovers TTC when combined with strategy-proofness, pair efficiency, and internal fairness. Finally, TTC is the unique allocation in the ultimate core defined for the model (Cheng et al., 26 Sep 2025). This is a notable shift in emphasis: the mechanism is no longer justified only by efficiency and incentives, but also by carefully tailored fairness constraints under incomplete priorities.

5. Extensions beyond the classical one-object market

A concise way to view later TTC research is as a sequence of controlled relaxations of the classical environment.

Setting	TTC modification	Salient implication
Multiple-type housing markets	cTTC and bTTC	Efficiency must be weakened to retain IR and SP (Feng, 2023)
Fractional/random allocation	FTTC	Preserves SD-efficiency and fairness on weak-preference domains (Yu et al., 2020)
Networked housing markets	TTCD or TTCI	Invitation structure changes incentive constraints (Zhang et al., 2023)
Shared CPS reassignment	ReACT-TTC	Handles capacities and unassigned resources after deviations (Satpathy et al., 31 Jan 2026)

In fractional assignment, classical TTC is inadequate for ex-ante fairness because deterministic cycles alone cannot incorporate the randomization needed in many allocation environments. Fractional TTC extends TTC to fractional endowment exchange and to the full preference domain with indifferences. The mechanism preserves individual rationality and SD-efficiency, and with appropriate parameter choices also satisfies equal treatment of equals, equal-endowment no envy, bounded envy, and, under equal endowments, envy-freeness (Yu et al., 2020). In house allocation, this yields an extension of Probabilistic Serial to weak preferences.

In networked housing markets, the issue is not randomization but participation. Classical TTC assumes a fixed market; in invitation-mediated markets, participants may strategically suppress invitations because invitees can become competitors. One paper proves that no mechanism can achieve individual rationality, strategy-proofness, and Pareto efficiency simultaneously in such networked markets without restricting the preference domain, and it identifies precise inviter–invitee conflict patterns that break TTC’s strategy-proofness (Zhang et al., 2023). A related paper proposes TTCI, which restricts each agent’s available set to neighbors and descendants in the generated invitation graph; under that restriction, truthful preferences and inviting all neighbors are dominant strategies, while the resulting allocation remains Pareto optimal (Zheng et al., 2020).

In cyber-physical systems, TTC has recently been repurposed as a post-deviation reassignment layer rather than a primary allocation rule. ReACT-TTC starts from any initial assignment, runs only after users refuse assigned resources, extends TTC to many-to-one capacities and unassigned resources via virtual owners, and proves termination, Pareto efficiency, individual rationality, and strategy-proofness under strict preferences (Satpathy et al., 31 Jan 2026). This suggests that TTC’s cycle logic is portable well beyond classical housing exchange, provided ownership and capacity are encoded carefully.

6. Computation, information, and open questions

The standard TTC algorithm is already computationally simple, but recent work examines whether its full output is always necessary. One paper studies the Core Identification Problem in one-sided matching and argues that identifying which agents keep their endowed object in the TTC core allocation is strictly easier than computing the full allocation. The paper proposes an $O(Ln)$ procedure based on a preference-derived Markov transition matrix, where $L$ is the maximum reported list length, and claims this strictly improves on the $O(n\log n)$ complexity cited for full TTC computation (Aldridge, 25 Apr 2026). This suggests a separation between full allocation computation and certain TTC-derived diagnostic tasks.

Another informational perspective compares TTC with DA, Immediate Acceptance, and Serial Dictatorship in school choice. Informational size measures how much preference and priority information is needed to secure a particular assignment in the worst case. In the heterogeneous-priority case, TTC requires less information than DA when there are at least four students, whereas DA can require less information in smaller markets. Under common priorities, TTC, DA, and priority-respecting Serial Dictatorship coincide in informational size (Feng et al., 2024). A plausible implication is that TTC’s transparency and strategic usability depend not only on its formal incentive properties but also on market size and priority heterogeneity.

Computation becomes more delicate in generalized exchange economies. In housing markets where agents hold multiple endowments, TTC is manipulable, and the complexity of finding a beneficial misreport depends sharply on the endowment-size parameter $k$. When $k$ is fixed, manipulation can be found in polynomial time; parameterized by $k$, however, the problem is $\mathcal{W}[1]$-hard, so fixed-parameter tractability is unlikely (Phan et al., 2018). This shows that TTC’s practical strategy-proofness does not survive naively once the environment moves beyond one object per agent.

Several papers also identify substantive open directions. In multi-center allocation, incomplete cross-center priorities complicate stability notions, and cycle-length constraints remain to be integrated formally into fairness-based TTC design (Cheng et al., 26 Sep 2025). In multiple-type markets, richer efficiency notions often become impossible once complementarities are admitted (Feng, 2023). In school choice with endogenous residential sorting, TTC can amplify long-run wealth segregation relative to DA and neighborhood assignment, because local priority rights become highly tradable through cycles (Artemov et al., 13 Nov 2025). And in the one-sided matching setting, enumerating the full Pareto frontier via inverse TTC methods reveals that focusing on a single TTC outcome can conceal significant fairness and welfare trade-offs among Pareto-optimal allocations (Dodda et al., 23 Apr 2026).

These developments do not displace the classical mechanism. Rather, they clarify its exact domain of validity. TTC remains the benchmark trading mechanism for one-sided reallocation with endowments and strict preferences, and a recurrent endpoint of axiomatic characterization. But the contemporary literature treats TTC less as a monolith than as a design schema: when ownership, priorities, capacities, domains, or participation constraints change, the question is no longer whether cycles exist, but which cycle system preserves the relevant combination of efficiency, fairness, robustness, and incentive compatibility.