Research

Publications

Working Papers

  • Kei Ikegami,
    Policy Learning by a Magician
    (New draft: October 29, 2026)
    Replication code

    We study eligibility learning for sequential programs under an exact capacity constraint, which limits treatments to available slots, and arrival neutrality, which gives eligible applicants a common treatment probability across arrival orders. We propose magician-adjusted empirical welfare maximization, weighting estimated gains by treatment probabilities from the conservative magician of Alaei (2012). Per-applicant regret relative to the best rule under this magician decreases at the inverse-square-root rate in experimental sample size. Under certain conditions, we show that the best magician implementation retains nearly all of the welfare attainable under an expected-capacity constraint as capacity grows.

  • Kei Ikegami and Hikaru Kawarazaki,
    Remembering the Shogun Era: Long-Run Effects of a Catastrophe on Policymaking over 160 years
    (New draft: September 14, 2026)
    Supplemental Appendix

    We examine whether disaster experience shapes political attention and preparedness across generations. Lords’ predetermined residence schedules provide a proxy for exposure to Japan’s unexpected 1855 Ansei Edo earthquake, which we link to modern municipalities through historical village holdings. Greater exposure is associated with larger discussion responses to new earthquakes, including risk, evacuation, and communication, especially in consistently observed councils. Exposure also predicts preparations for outside assistance and shelter supplies. Differences in specifying tasks for outside disaster-response personnel emerge after 2018. Seismic-retrofit support shows no detectable differential change. Preserving experience may therefore contribute to public preparedness across generations.

  • Kei Ikegami,
    Who Is Shown to Whom? Sorting, Screening, and Exposure Design on a Two-Sided Matching Platform
    (New draft: September 27, 2026)
    Accepted for WINE 2026

    Two-sided platforms combine decentralized search with recommendations, so recommendation design affects both who meets and how users subsequently search. I study a Japanese medical-staffing platform where doctors apply to short-term posts and platform agents recommend doctors to institutions. Using route-identified contacts and bilateral acceptance decisions, I estimate an equilibrium model of contact, acceptance, and the value of remaining unmatched. The counterfactuals yield three design implications. First, platforms should retain doctor search alongside algorithmic recommendations. Second, a long-run objective that accounts for how continuation values adjust when the recommendation policy changes performs robustly across the platform’s multiple objectives. Third, increasing recommendation volume alone produces modest gains; reallocating recommendations across doctors and posts is substantially more consequential.

    • APIOC 2026, SSCW 2026, IIOC 2026, ERATO weekly meeting

  • Kei Ikegami, Atsushi Iwasaki, Akira Matsushita, and Kyohei Okumura,
    Caps vs. Prices: Regulating Matching Markets with Distributional Disparities
    (New draft: August 24, 2026)
    Accepted for EC 2025 (Exemplary Paper Award, Empirics Track)

    Many matching markets rely on quantity-based instruments, such as caps, to achieve distributional objectives, but little is known about their efficiency relative to price-based instruments. We develop a matching model with transferable utility that incorporates regional constraints and characterize the taxes and subsidies that maximize surplus subject to these constraints. We then embed the model in an aggregate matching framework and show how the optimal policy can be constructed from aggregate match data. We apply the framework to the Japan Residency Matching Program, which limits urban residency positions to encourage training in rural areas. Relative to a benchmark without cap tightening, the policy increases rural placements but reduces social surplus by 3.23 billion yen per month. A targeted subsidy policy achieves the same placement gains in each rural prefecture while reducing social surplus by only 0.15 billion yen per month, recovering 95% of the loss under the current caps. Our results suggest that most of the efficiency loss comes from the choice of instrument rather than from the distributional objective itself.

    • MATCH-UP 2026
    • SITE 2025
    • EC 2025
    • IIOC 2025
    • APIOC 2024
    • NYU seminar
    • JEA annual meeting 2024

  • Kei Ikegami, Ken Onishi, and Naoki Wakamori,
    Joint Venture Formation in Procurement Auctions
    (New draft: October 23, 2024)

    We propose a model for joint venture formation in the context of procurement auctions. This model enables us to identify the formation mechanism, which is necessary for simulating counterfactual auction settings. We estimate the model using newly collected Japanese procurement auction data. Our estimates reveal the presence of cost synergies: joint ventures are more likely to be cost-effective compared to standard bidders. Despite this pro-competitive effect, our simulation indicates that excessive encouragement of joint ventures hinders cost-effective procurement by reducing competition. This anti-competitive effect arises from the diminished incentive to enter the auction due to the possibility to compete with joint ventures.

    • AEA 2024
    • Otaru University of Commerce
    • Yokohama National University
    • NYU seminar
    • APIOC 2022

Work in Progress

  • Kei Ikegami, Shuhei Otani, and Yuta Nakamura, The Signal Magician: Online Allocation with Imperfect Screening

  • Kei Ikegami, Bargaining over Leasing Contracts
    (New draft coming soon)

    This paper investigates how kinked revenue-sharing contracts are negotiated and how they embed bargaining power. Using proprietary panel data on tenant contracts and sales in shopping malls, I estimate a structural model in which a risk-averse tenant and mall renegotiate contracts based on realized sales. The results show that the mall-to-tenant power ratio shapes contract terms through two channels: (i) risk posture—powerful malls can insist on low-variance, high-fixed-rent contracts; and (ii) sales expectations—pessimistic forecasts cap fixed rent and shift weight to commissions. Because these forces offset, contract form alone seldom signals power imbalance. Counterfactual simulations imposing fairer bargaining weights indicate that, after risk adjustment, total rent can triple even when the mall is relatively weak.

    • ESWC 2025
    • IIOC 2025
    • APIOC 2024
    • UTMD Rising Star in Market Design
    • NYU seminar

  • Kei Ikegami and Kan Kuno, Wage Stagnation in Daycare Industry: A Two-Sided Market Perspective

  • Kei Ikegami, Recommendation and Long-run Matching Outcomes, presented at SWET 2026

  • Kei Ikegami, Procuring Hunting Effort