Research map / Multi-agent systems and games / Games and equilibria
Learning in reverse causal strategic environments with ramifications on two sided 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.
Introduces reverse causal strategic classification where agents change outcomes that propagate to features, and shows in Coate-Loury labor market models that performative (optimal) employers raise employer reward and labor skill, with mixed effects on equity and worker welfare.
Problem
Strategic classification usually assumes agents manipulate features or features-then-outcomes causally; here agents manipulate outcomes that causally shift features (reverse causal), as in labor markets where employers' hiring policies anticipate workers becoming skilled. The question is how anticip
Approach
Formulates reverse causal strategic learning where the learner minimizes post-strategic-response loss, proves performative hiring thresholds beat stable ones for employers and raise worker skill, and gives an SGD-with-REINFORCE algorithm for the performative risk when the feature mechanism is known.
Key results
- Theorems show that in high-wage markets optimal (performative) hiring yields higher skilled fraction and employer utility at least (1+delta_1) times stable utility, under stated conditions.
- In the general equilibrium simulations, performative employers increase labor force skill levels but reduce aggregate worker welfare.
- Fairness benefits of performative hiring are brittle: in the non-linear utility/non-flat wage experiments, optimal policies no longer enforce fairness.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use a performatively optimal hiring policy instead of a stable hiring policy | Coate-Loury labor market · Coate-Loury model | employer reward | Helped |
| use a performatively optimal hiring policy instead of a stable hiring policy | Coate-Loury labor market · Coate-Loury model | fraction of skilled workers | Helped |
| use a performatively optimal hiring policy instead of a stable hiring policy | two-group Coate-Loury labor market · Coate-Loury model | qualification disparity between groups | Helped |
| use a performatively optimal hiring policy instead of a stable hiring policy | continuous-skill labor market · Linear utility and flat wages | fraction of qualified workers | Helped |
| use a performatively optimal hiring policy instead of a stable hiring policy | continuous-skill labor market · Linear utility and flat wages | aggregate worker welfare | Hurt |
| use a performatively optimal hiring policy instead of a stable hiring policy | continuous-skill labor market · Linear utility and flat wages | employer utility | Helped |
| use a performatively optimal hiring policy instead of a stable hiring policy | continuous-skill labor market · Non-linear utility and non-flat wages | fraction of qualified workers | Mixed |
| use a performatively optimal hiring policy instead of a stable hiring policy | continuous-skill labor market · Non-linear utility and non-flat wages | average worker welfare | Hurt |
| use a performatively optimal hiring policy instead of a stable hiring policy | continuous-skill labor market · Non-linear utility and non-flat wages | discrimination between groups | No clear effect |
| use a performatively optimal hiring policy instead of a stable hiring policy | general-equilibrium labor market · Continuous skill levels | aggregate worker welfare | Hurt |
| use a performatively optimal hiring policy instead of a stable hiring policy | general-equilibrium labor market · Continuous skill levels | fraction of skilled workers | Helped |
| use a performatively optimal hiring policy instead of a stable hiring policy | general-equilibrium labor market · Continuous skill levels | employer reward | Helped |
Surprising findings
- Performative employers harm aggregate worker welfare even as they raise skill levels, creating a trade-off between worker skill and worker welfare.
- Performative employers do not reliably prevent discrimination; fairness benefits disappear under non-linear utility and non-flat wages.
Limitations
- Theorems require strong market assumptions (wage bounds, likelihood ratio conditions, strong concavity, Lipschitz inverse best response).
- Algorithm assumes the learner knows the reverse causal map and feature mechanism phi(x|y).
- Experiments use simple synthetic markets with specific parametrizations rather than real labor data.
Experimental setup
Metrics: employer utility proportion of qualified workers worker welfare discrimination gap between groups
Baselines: stable (reactive) employer policy repeated risk minimization
Builds on: performative prediction strategic classification Coate-Loury model repeated risk minimization stateful performative prediction
Closest papers by meaning
- Equal Merit Does Not Imply Equality: Discrimination at Equilibrium in a Hiring Market with Symmetric Agents (AAAI 2025 · similarity 0.87)
- Discrimination in Online Markets: Effects of Social Bias on Learning from Reviews and Policy Design (NeurIPS 2019 · similarity 0.86)
- Strategic Causal Policy Learning: Welfare, Safety, and Fairness Thresholds (NeurIPS 2026 · similarity 0.86)
- Causal Strategic Learning with Competitive Selection (AAAI 2024 · similarity 0.85)
- Strategic Costs of Perceived Bias in Fair Selection (NeurIPS 2025 · similarity 0.84)
- Understanding Model Selection for Learning in Strategic Environments (NeurIPS 2024 · 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.