Research map / Learning theory and algorithms / Prediction and decision losses
Monoculture in Matching Markets
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Builds a continuum two-sided matching model of algorithmic monoculture (shared vs. independent noisy firm evaluations) and proves polyculture yields a wisdom-of-the-crowds firm welfare gain while monoculture gives applicants better top-choice matching, with simulations.
Problem
How algorithmic monoculture (many firms using a shared evaluation algorithm) affects firm and applicant welfare in two-sided matching markets with many participants, including differential application access.
Approach
Models firms' noisy estimated values in a continuum stable-matching (cutoff) model, with monoculture as a single shared noisy estimate and polyculture as independent per-firm estimates; uses maximum order statistics and maximum-concentrating noise to show polyculture approaches firm-optimal matching as the number of firms grows.
Key results
- Under polyculture with maximum-concentrating noise, the probability of matching approaches a step function at the firm-optimal threshold as the number of firms grows, while monoculture's matching probability is constant in m.
- Every applicant is at least as likely to match to their top choice under monoculture as under polyculture, and all matched applicants under monoculture get their top choice.
- Under polyculture applicants who apply to more firms are more likely to match, whereas under monoculture application count gives no advantage.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use independent firm-specific estimates instead of a shared estimate | matching markets with many firms · matching market | average value of matched applicants | Helped |
| use independent firm-specific estimates instead of a shared estimate | matching markets with many firms · matching market | average applicant match rank | Hurt |
| use a shared estimate across firms instead of independent firm-specific estimates | matching markets · matching market | probability of matching to the top choice | Helped |
| use independent firm-specific estimates instead of a shared estimate | matching markets with many firms and maximum-concentrating noise · matching market | firm welfare | Helped |
| use independent firm-specific estimates instead of a shared estimate | individual applicants in matching markets · matching market | probability of matching | Mixed |
| increase the number of applications submitted | polyculture matching markets · differential application access | probability of matching | Helped |
| increase the number of applications submitted | monoculture matching markets · differential application access | probability of matching | No clear effect |
| introduce differential application access instead of uniform application access | polyculture matching markets · differential application access | firm welfare | Hurt |
| introduce differential application access instead of uniform application access | monoculture matching markets · differential application access | firm welfare | No clear effect |
| use independent firm-specific estimates instead of a shared estimate | matching markets with correlated applicant preferences · simulated matching markets | average value percentile of matched applicants | Helped |
| use independent firm-specific estimates instead of a shared estimate | matching markets with correlated applicant preferences · simulated matching markets | average applicant match rank | Hurt |
| increase the number of applications submitted | polyculture matching markets · simulated markets with correlated preferences | probability of matching | Helped |
Surprising findings
- Monoculture can make some applicants less likely to match at all even while more likely to match to their top choice, so neither dominates stochastically.
- Firms can be better off using independent noisy evaluations than a shared, more accurate one when there are many firms.
Limitations
- Continuum model with symmetric uniformly random applicant preferences; correlated preferences only tested in simulation.
- Firms' preference approximation only; applicants' true preferences are known and fixed; capacity fully filled by assumption.
- Small-market stochasticity is not captured by the continuum model.
Experimental setup
Metrics: probability of matching by value average percentile value of matched applicants average rank of matches difference in matching probability by application count
Baselines: monoculture polyculture
Builds on: continuum model of stable matching (Azevedo and Leshno) algorithmic monoculture model (Kleinberg and Raghavan) single vs. multiple tie-breaking in school choice
Closest papers by meaning
- Matchings Under Biased and Correlated Evaluations (NeurIPS 2025 · similarity 0.86)
- Putting Gale & Shapley to Work: Guaranteeing Stability Through Learning (NeurIPS 2024 · similarity 0.86)
- The Subjectivity of Monoculture (NeurIPS 2026 · similarity 0.85)
- MATWA: A Web Toolkit for Matching Under Preferences (AAAI 2025 · similarity 0.84)
- Promoting Fairness Among Dynamic Agents in Online-Matching Markets under Known Stationary Arrival Distributions (NeurIPS 2024 · similarity 0.83)
- Equal Merit Does Not Imply Equality: Discrimination at Equilibrium in a Hiring Market with Symmetric Agents (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.