Inverse Market Design: Auditing Hidden Objectives with Differentiable Economics

Nik-Khah and Mirowski (2019) argue that market design is not neutral—designers encode client objectives under the guise of unified theory. We can make this operational. Using ideas from differentiable economics (Bichler & Parkes, 2025) and preference-encoding mechanisms like PreferenceNet (Peri et al., 2021), this project builds an inverse problem: given observed rules and outcomes (prices, allocations, fairness/diversity patterns, error rates), infer the latent weights a designer placed on revenue, welfare, fairness, risk, or compliance. Apply this to historical sponsored search designs (Yudina, 2024) and contemporary BECCS procurement pilots (Fridahl et al., 2024) to test for implicit objectives, and compare them against stated policy goals. The novelty is methodological: rather than asserting bias qualitatively, we recover objective weights algorithmically using differentiable simulators of strategic behavior and equilibrium (extending auction-learning frameworks used by Ravindranath et al., 2023). A validated audit tool would let regulators and stakeholders test whether mechanisms align with public aims, turning a conceptual critique into a measurable property of market design.

References:

1. Data Market Design through Deep Learning. S. Ravindranath, Yanchen Jiang, David C. Parkes (2023). Neural Information Processing Systems.
2. On going the market one better: economic market design and the contradictions of building markets for public purposes. Edward Nik-Khah, Philip Mirowski (2019). Economy and Society.
3. The Genesis of Auction Design in Online Advertising. O. Yudina (2024). Scientific Research and Development Economics.
4. Potential and goal conflicts in reverse auction design for bioenergy with carbon capture and storage (BECCS). Mathias Fridahl, Kenneth Möllersten, Liv Lundberg, Wilfried Rickels (2024). Environmental Sciences Europe.
5. PreferenceNet: Encoding Human Preferences in Auction Design with Deep Learning. Neehar Peri, Michael J. Curry, Samuel Dooley, John P. Dickerson (2021). Neural Information Processing Systems.
6. Differentiable Economics: Strategic Behavior, Mechanisms, and Machine Learning. Martin Bichler, David C. Parkes (2025). Communications of the ACM.