Source: college_matching_market.md
Research on how US college admissions fits (or fails to fit) the stable matching framework from market design theory, drawing on the work of Gale-Shapley, Roth, and subsequent researchers.
College admissions is a canonical example of a two-sided matching market: students have preferences over colleges, and colleges have preferences over students. Prices alone do not clear the market -- you cannot simply purchase a spot at Harvard. Both sides must agree to a match, making this fundamentally different from commodity markets where price equilibrates supply and demand.
The formal theory begins with Gale and Shapley's 1962 paper "College Admissions and the Stability of Marriage," which introduced the deferred acceptance (DA) algorithm and the concept of stability in matching. A matching is stable if there is no student-college pair who would both prefer to be matched with each other over their current assignments. Gale and Shapley proved that a stable matching always exists and can be found by their algorithm.
While the Gale-Shapley framework was literally named after the college admissions problem, the actual US admissions system departs from the theoretical model in several critical ways:
Decentralized process: There is no centralized clearinghouse. Students apply directly to colleges, and colleges make independent admission decisions. This contrasts with the NRMP (National Resident Matching Program) for medical residencies, which uses a centralized DA algorithm.
Incomplete and asymmetric information: Colleges cannot perfectly observe student quality, and students cannot perfectly observe their probability of admission. The Gale-Shapley model assumes complete preference orderings on both sides.
Multiple rounds with commitment devices: The US system operates through sequential rounds (ED, EA/REA, EDII, RD, waitlist) rather than a single simultaneous match. Each round has different strategic implications.
Capacity uncertainty and yield management: Colleges do not know how many admitted students will enroll, so they admit more students than they have seats for. This "yield management" problem has no analog in the standard matching model.
Financial aid as a contract term: Matching is not simply student-to-college; it includes the financial terms of the match. Hatfield and Milgrom (2005) extended matching theory to "matching with contracts," which can accommodate financial aid as a dimension of the match, but this adds substantial complexity.
Non-strict preferences and holistic review: Colleges use "holistic" review with subjective components (essays, recommendations, institutional priorities), producing preferences that are not strictly ordinal and may be intransitive or context-dependent.
The National Resident Matching Program is the paradigmatic example of a centralized matching clearinghouse that successfully resolved market failures:
| Feature | NRMP (Medical Residency) | US College Admissions |
|---|---|---|
| Structure | Centralized clearinghouse | Decentralized, sequential rounds |
| Algorithm | Student-proposing DA (Roth-Peranson) | No algorithm; colleges decide independently |
| Strategy-proofness | Truthful reporting is dominant strategy for students | Strategic behavior is pervasive and rewarded |
| Stability | Produces stable matching by design | No guarantee of stability; blocking pairs likely exist |
| Information | Rank order lists submitted after interviews | Holistic review with subjective criteria |
| Timing | Single match date | ED (Nov) -> EA (Nov) -> EDII (Jan) -> RD (Mar) -> Decisions (Apr) -> Waitlist (May+) |
| Binding | Match is binding on both sides | Only ED is binding on students; colleges can rescind |
The NRMP was established in the 1950s precisely because the medical residency market had suffered from severe unraveling -- offers were being made progressively earlier, eventually two years before graduation. Roth (1984) showed that the NRMP algorithm produces a stable matching, and Roth (1991) demonstrated that markets with stable clearinghouses tend to succeed while those without them tend to fail.
The key question is: why has US college admissions not adopted a similar centralized mechanism? The answer involves institutional resistance, the value colleges place on "demonstrated interest" as a selection tool, the role of financial aid in the match, and political economy considerations (see "Proposed Centralized Mechanisms" below).
Roth and Xing's landmark 1994 paper "Jumping the Gun: Imperfections and Institutions Related to the Timing of Market Transactions" documented a pervasive phenomenon across dozens of markets: unraveling, in which transactions occur progressively earlier over time, to the point where matches are made with severely incomplete information.
The paper examined markets ranging from medical internships (where offers were made two years before graduation) to postseason college football bowls, and identified common patterns:
US college admissions exhibits classic unraveling dynamics:
Historical progression: Early Decision programs have proliferated and expanded since the mid-20th century. The proportion of selective college slots filled through early rounds has grown dramatically, with some elite schools now filling 40-60% of their class through ED/EA.
EDII as further unraveling: The introduction of Early Decision II (January deadline) represents a further step in the unraveling process -- a second early round added on top of the first.
REA as a response: Restrictive Early Action (used by Harvard, Yale, Princeton, Stanford) is an institutional attempt to slow unraveling by allowing early applications without binding commitment, while restricting students from applying early elsewhere.
Roth and Xing identified multiple causes that apply to college admissions:
Markets that have successfully resolved unraveling (NRMP, some law clerkship markets) did so by establishing centralized clearinghouses. College admissions has not done this because:
Colleges benefit from unraveling: ED gives colleges predictable yield and reduces uncertainty. Highly selective colleges can lock in committed students early.
No single authority: Unlike medical residencies (governed by professional organizations), there is no body with authority to impose a centralized mechanism.
Heterogeneous preferences over mechanism design: Elite colleges prefer the current system (which favors them), while less selective colleges would benefit from centralization.
Avery, Fairbanks, and Zeckhauser's 2003 book The Early Admissions Game: Joining the Elite (Harvard University Press) provided the definitive empirical analysis of early admissions as strategic behavior. Their study of over 500,000 applications to 14 elite colleges found that:
Applying ED provides an admissions advantage equivalent to ~100 SAT points. This is not fully explained by self-selection (stronger students applying early); there is a genuine boost from the commitment signal.
ED functions as a signaling device: Colleges want to admit students who will actually enroll. A binding ED application is a credible signal of strong preference ("demonstrated interest") that colleges reward.
The game is not fair: Students with access to sophisticated counseling (private schools, wealthy families) are far more likely to understand and exploit early admissions strategically.
From a matching theory perspective, Early Decision serves as a decentralized commitment mechanism that partially substitutes for a centralized clearinghouse:
Reducing instability: By having students commit early, ED reduces the number of "blocking pairs" (student-college pairs who would prefer each other to their actual match) that would otherwise exist.
Yield certainty as a substitute for stability: In a DA algorithm, colleges know exactly how many students will enroll because the algorithm produces a final matching. ED provides partial yield certainty by guaranteeing enrollment from ED admits.
Sequential mechanism as approximate DA: The round structure (ED -> EA -> RD) can be viewed as an approximation of sequential proposals in the DA algorithm, but with the crucial difference that ED makes proposals binding on one side, which the DA algorithm does not require during the proposal phase.
Preference revelation: ED applications reveal cardinal preference intensity (not just ordinal rankings), which the standard Gale-Shapley framework does not capture. This is information-theoretically valuable but creates strategic asymmetries.
Avery and Levin published "Early Admissions at Selective Colleges" in the American Economic Review (2010), providing a game-theoretic model that rationalizes the observed patterns:
Colleges value "enthusiasm" (probability of enrollment) alongside academic quality.
Early applications allow students to signal enthusiasm credibly because applying ED involves a real cost (forgoing other options).
In equilibrium, students sort into early vs. regular based on their relative preference intensity: students who strongly prefer one school apply ED there; students with more diffuse preferences apply RD to multiple schools.
The model predicts the observed admissions advantage for early applicants: colleges rationally give a boost to early applicants because those applicants are more likely to enroll and be satisfied (reducing "mismatch").
From Roth's perspective, ED is an exploding offer in reverse: instead of the college making an offer the student must accept immediately, the student makes a binding commitment before receiving the offer. This is even more extreme than a traditional exploding offer because:
The student commits before knowing the outcome (admission) or the terms (financial aid).
The commitment is one-sided: the college is not bound to admit the student.
The student gives up optionality across all other schools.
This creates a strategic asymmetry where colleges capture most of the surplus from the early matching.
The matching theory lens reveals distributional consequences:
Information advantage: Students who understand the strategic value of ED (typically affluent, well-counseled students) capture the signaling benefit, while uninformed students do not. Research shows ED applicants are three times more likely to be white.
Financial aid comparison: Low-income students rationally avoid ED because they need to compare financial aid packages, which the binding commitment prevents. This is a market failure: the mechanism designed to improve match quality systematically excludes students who most need financial flexibility.
Thick vs. thin market participation: ED makes the RD round "thinner" (fewer top students, fewer seats), degrading match quality for students who cannot participate in early rounds.
M. Bumin Yenmez proposed a centralized clearinghouse for college admissions that preserves the desirable features of ED (signaling, yield management) while eliminating undesirable aspects (unfairness, unraveling). Key features:
Students submit rank-order lists and can designate their top choice as a binding commitment (analogous to ED).
Financial aid terms are included in the match contracts -- students can specify minimum financial aid requirements.
A modified DA algorithm produces the matching, with the commitment signal factored into college preferences.
Students can be matched with multiple colleges simultaneously (for non-binding preferences), then choose.
This approach extends Hatfield and Milgrom's (2005) "matching with contracts" framework to accommodate the specific institutional features of college admissions.
Several factors explain the persistence of the decentralized system:
Elite college resistance: Top colleges benefit from the current system's strategic asymmetries. A centralized mechanism would be strategy-proof for students (truthful reporting as dominant strategy), which reduces colleges' ability to exploit demonstrated interest and yield management tactics.
Financial aid complexity: Unlike medical residencies (where salary is roughly standardized), college financial aid packages vary enormously and are a key dimension of the match. Incorporating financial aid into a centralized mechanism is technically feasible (via matching with contracts) but institutionally complex.
Holistic review: Colleges value the discretion of holistic admissions (essays, interviews, institutional priorities). A centralized mechanism requires colleges to submit explicit rank orderings, which conflicts with the narrative of "holistic" review. In practice, colleges do have implicit rankings, but formalizing them raises political and legal issues.
Common Application as partial centralization: The Common Application already centralizes the application process but not the matching process. Moving to centralized matching would require colleges to cede decision-making authority to an algorithm, which no US institution has been willing to do.
Political economy: No entity has the authority or incentive to impose centralization. The Common Application organization, NACAC, and individual colleges all have different interests.
Several countries use centralized college matching:
Turkey: Uses a constrained DA mechanism for university placement. After a 1999 reform to post-exam preference submission, the system produced stronger assortative matching and less under-capacity at lower-ranked programs.
Brazil: Introduced a centralized clearinghouse (SISU) using DA with scores from the ENEM national exam. Research found that centralization increased stratification by quality and attracted higher-scoring students from other regions.
Chile, Germany, Taiwan: Also use centralized assignment systems for university admissions.
These international examples demonstrate that centralized college matching is technically feasible and can improve match quality, but they also reveal trade-offs: increased stratification, potential reduction in diversity, and the need for standardized assessment metrics.
The foundational equilibrium concept for matching markets. A matching is stable if:
Every student is matched to an acceptable college (or unmatched).
Every college fills at most its capacity.
There is no blocking pair: a student s and college c such that s prefers c to their current match and c prefers s to at least one of their current admits (or has unfilled capacity).
Applicability to US admissions: The decentralized system almost certainly produces unstable matchings. Students frequently end up at schools they prefer less than schools that would have preferred them (because they did not apply, applied in the wrong round, or were waitlisted). The sequential round structure creates artificial constraints that prevent stable outcomes.
When students must choose where to apply (with limits on number of applications and strategic round selection), the relevant concept is Nash equilibrium in the application game:
Each student's strategy is a portfolio of applications (which schools, in which rounds).
Colleges' strategies are admission policies (thresholds, ED boosts, yield protection).
A Nash equilibrium is a profile of strategies where no student or college can improve their outcome by unilaterally changing their strategy.
Research shows that Nash equilibria in truncated preference lists (where students don't apply everywhere) produce at most one matching, and any stable matching in equilibrium must be unique. However, multiple Nash equilibria may exist, and they need not be stable in the Gale-Shapley sense.
Kelso and Crawford (1982) showed that under a substitutability condition (adding a student to a college's choice set never causes the college to reject a previously accepted student), stable matchings correspond to competitive equilibria with personalized prices. This connection is powerful because:
It links matching theory to classical price theory.
It provides welfare theorems: the student-optimal stable matching maximizes student welfare among all stable matchings.
It breaks down when colleges have complementary preferences (e.g., wanting to balance gender, achieve geographic diversity, or pair athletes with non-athletes), which is precisely what "holistic review" entails.
The core of the college admissions market is the set of all stable matchings. In the standard Gale-Shapley model with substitutable preferences, the core is always non-empty and has a lattice structure (with student-optimal and college-optimal extremes).
However, Yenmez and others have shown that when financial aid is included (the "need versus merit" problem), the core can be very large -- many different stable matchings exist with very different distributional properties. This "large core" result means that the choice of mechanism (student-proposing DA vs. college-proposing DA vs. some other stable mechanism) has significant distributional consequences, even though all mechanisms produce stable outcomes.
Hatfield and Milgrom (2005) generalized matching theory to include contract terms (not just binary match/no-match). Their framework encompasses:
College admissions with financial aid as a contract term.
The Kelso-Crawford labor market model with wages.
Ascending package auctions.
Under a substitutes condition (and a "law of aggregate demand"), the DA algorithm generalizes to matching with contracts and retains its desirable properties (stability, strategy-proofness for the proposing side). This is the most promising theoretical framework for a centralized college matching mechanism that incorporates financial aid.
Recent work on "protocol-free equilibrium" shows that an outcome is a subgame perfect equilibrium robust to the choice of bargaining protocol if and only if it corresponds to a stable matching. This result applies to college admissions problems, matching with contracts, and assignment games. It provides a game-theoretic foundation for stability that does not depend on the specific mechanism used.
The current system is not a stable matching mechanism. The simulation should not model US admissions as a DA algorithm. Instead, it should model the decentralized, sequential process with its strategic complexities.
Round structure matters. The ED -> EA/REA -> EDII -> RD -> waitlist sequence is not just administrative; it is a mechanism that creates strategic incentives. The simulation should model each round separately with appropriate behavioral rules:
ED: Binding commitment with admissions boost (the "signaling premium"). Model as ~1.5-2x admission probability multiplier.
EA/REA: Non-binding signal of interest. Smaller boost than ED.
RD: No signaling premium. Largest applicant pool, most competitive.
Waitlist: Yield management tool. Colleges admit from waitlist to hit enrollment targets.
Student application portfolios are strategic. Students do not simply apply to their top-N choices. They construct portfolios balancing reach/match/safety schools and strategically allocate their ED application (if any). The simulation should model this portfolio construction.
Yield management is central. Colleges over-admit because they cannot predict yield. The simulation should model yield rates by round and adjust admission thresholds accordingly.
Hook multipliers as preference heterogeneity. Legacy, athletic, donor, and first-gen hooks represent heterogeneous college preferences that violate the simple "rank by quality" assumption. These are exactly the kind of complementarities that the Hatfield-Milgrom framework accommodates but that simple DA does not.
The simulation can measure instability. After producing a matching, the simulation can count the number of blocking pairs -- cases where a student ended up at school A but would have been admitted to (preferred) school B, which would have preferred them to a marginal admit. This is a direct measure of how far the decentralized outcome is from a stable matching.
Equity analysis through the matching lens. The simulation can compare outcomes for students with and without access to early rounds (a proxy for wealth/information access) to quantify the distributional effects documented by Avery et al.
Do not use Gale-Shapley DA as the core engine. The simulation correctly uses a round-based decentralized process. This is more realistic than DA for modeling the US system.
Model the ED signaling premium explicitly. The current simulation uses round multipliers (ED gets a boost); this is consistent with Avery and Levin's signaling model.
Consider adding a "counterfactual DA" comparison. Run the same students and colleges through a DA algorithm to compute what the stable matching would have been, and compare it to the decentralized outcome. This directly measures the welfare cost of decentralization.
Measure blocking pairs as a quality metric. After each simulation run, count how many student-college pairs are blocking pairs. This provides a theory-grounded measure of market efficiency.
Model financial aid constraints for low-income students. Students who cannot apply ED due to financial aid uncertainty represent a key market failure. The simulation could track which students "should have" applied ED but did not, and measure their outcome penalty.
| Parameter | Theoretical Basis | Suggested Range |
|---|---|---|
| ED admission multiplier | Signaling premium (Avery & Levin 2010) | 1.5x - 2.0x |
| EA admission multiplier | Weaker signal than ED | 1.1x - 1.3x |
| EDII admission multiplier | Late commitment signal | 1.3x - 1.5x |
| Yield rate by round | ED: ~100%, EA: 30-50%, RD: 20-40% | Calibrate to real data |
| Hook multipliers | Preference heterogeneity / complementarities | Legacy 2-3x, Athlete 3-4x, Donor 3-5x |
| Student portfolio size | Strategic portfolio construction | 8-15 applications |
| Information asymmetry | Well-counseled vs. poorly-counseled students | Varies by high school type |
Gale, D. and Shapley, L. (1962). "College Admissions and the Stability of Marriage." American Mathematical Monthly, 69(1), 9-15.
Roth, A.E. (1984). "The Evolution of the Labor Market for Medical Interns and Residents: A Case Study in Game Theory." Journal of Political Economy, 92(6), 991-1016.
Roth, A.E. and Xing, X. (1994). "Jumping the Gun: Imperfections and Institutions Related to the Timing of Market Transactions." American Economic Review, 84(4), 992-1044.
Kelso, A. and Crawford, V. (1982). "Job Matching, Coalition Formation, and Gross Substitutes." Econometrica, 50(6), 1483-1504.
Avery, C., Fairbanks, A., and Zeckhauser, R. (2003). The Early Admissions Game: Joining the Elite. Harvard University Press.
Hatfield, J.W. and Milgrom, P. (2005). "Matching with Contracts." American Economic Review, 95(4), 913-935.
Avery, C. and Levin, J. (2010). "Early Admissions at Selective Colleges." American Economic Review, 100(5), 2125-2156.
Yenmez, M.B. (2018). "College Admissions." Working paper, Boston College.
Niederle, M. and Roth, A.E. (2009). "Market Culture: How Rules Governing Exploding Offers Affect Market Performance." American Economic Journal: Microeconomics, 1(2), 199-219.