Research · Job market candidate 2026–27 · Household finance · Consumer credit

Who gets credit,
and what happens next.

I study household and consumer behavior in credit markets: how people search for loans, how lenders decide, and what happens to the people they turn down. My method is causal inference on register data.

PhD expected Spring 2027Dissertation in pre-examinationSupervisors: Camelia M. Kuhnen, Anders Löflund

01 · Job market paper

Solo-authoredNovember 2026Causal inferenceConsumer creditJEL D14 · D18 · G21 · G28 · G51 · C26

Protective Rejections in Consumer Credit

Paying out the loan raises 18-month default by 20.3 percentage points for the applicants whom the weekend delay keeps from borrowing.

+20.3 pp Higher 18-month default when the loan is paid out. Local average treatment effect, 95% CI 1.88 to 48.84 pp.
Question

If a rejected applicant doesn't default, was the rejection a mistake, or did it protect them?

Data and method

758,622 applications on a Finnish loan-comparison platform, linked to credit-register defaults for every applicant, funded or not. Among applicants who already have an offer, those who apply Friday to Sunday wait longer for the payout, and fewer of them borrow. That weekend delay is the instrument.

Why it matters

Without the loan these applicants default at 12.65%, below the 14.3% rate at which a loan breaks even, so they look like false rejections. With the loan they default at 32.9%: rejecting a hundred of them avoids twenty defaults. One default offsets the profit on six repaid loans, so each avoided default lets lenders approve more applicants who would repay.

Abstract
A rejected applicant who does not default is often counted as someone who should have received the loan, so lenders seem to supply too little credit. I find that some rejections are protective, because the loan itself would cause the default. I link applications on a Finnish loan-comparison platform to credit-register defaults for every applicant, funded or not. Using the slower processing of applications made Friday to Sunday as an instrument, I estimate that receiving the loan raises the 18-month default rate by 20.3 percentage points for the applicants whom this delay keeps from borrowing. The estimate corrects for the lower default rate of weekend applicants that is unrelated to the loan. If funded, the applicants whom this delay keeps from borrowing would default above the lender's break-even rate, yet they do not stand out on credit score, income, debt-to-income or other characteristics. A rejected applicant who does not default is therefore not evidence of a mistaken rejection, and rejecting these applicants protects both them and the lender.

02 · Working papers

NBER WP 35024April 2026With Camelia M. Kuhnen

Learning How To Borrow in a Fintech World: Consumer Behavior When Search Costs Are (Near) Zero

Data on 730,000 applications, 750,000 offers and 200,000 individuals from a Finnish online loan marketplace, merged with credit-registry records. Four facts:

  1. 1Offered terms vary widely for the same applicant, so search is valuable.
  2. 2Marketplace nudges reduce choice complexity.
  3. 3Applicants search in ways consistent with their creditworthiness.
  4. 4Lenders penalize repeat applicants, so adverse selection limits search.
Presentations (14)
  • Colorado Finance Summit (selected) · 2026
  • Stockholm School of Economics · 2026
  • NHH Bergen · 2026
  • Rice University · 2026
  • HEC Lausanne and EPFL · 2025
  • University of Geneva · 2025
  • American University · 2025
  • NYU Stern · 2025
  • University of Michigan · 2024
  • FDIC · 2024
  • Erasmus University Rotterdam · 2024
  • HEC Montréal · 2024
  • University of South Carolina · 2024
  • Federal Reserve Bank of Philadelphia · 2024

Presented by co-author Camelia M. Kuhnen.

Working paperSolo-authoredNovember 2026Credit scoringPrediction

The Path, Not the Position: What Credit Records See That Credit Scores Miss

A credit score shows where an applicant stands today, not how they got there. I study 82,794 repeat applicants on a Finnish lending marketplace, with register defaults observed for every applicant, funded or not. At the same current score, applicants whose score improved since their previous application default 18.86 percentage points more than applicants whose score is unchanged, and those whose score worsened 27.01 points more. Application timing accounts for most of this: applicants returning within 24 hours default at 2.7%, those returning after 30 to 90 days at 36.6%. A clean record that hides a deleted paid default mark carries 6.8 points more default risk. Adding earlier scores, marks and application dates to the current file raises out-of-sample AUC from 0.584 to 0.788. At the volume lenders funded, applicants selected using this history default at 1.56%, against 19.12% using the current file alone.

03 · In progress

In developmentWith Camelia M. Kuhnen

Data as a Strategic Asset

Do the loan outcomes a lender accumulates make it better at predicting default? Every application reaches all lenders at once, so lenders with very different amounts of data compete for the same applicants. The test is whether a lender's edge grows with its own record, which separates lenders that learn from lenders that only grow.

Early stageWith Camelia M. Kuhnen and Yunzhi Hu

Auction and pricing project

How lenders price when each application works as an auction among competing offers. Across about 80,500 competitive auctions, the winning rate sits on average 1.73 percentage points below the next-best offer.

Research agenda

Online loan marketplaces look as if they solve the classic information problems of consumer credit: shopping is nearly free and lenders see rich data. My research asks which of these problems actually go away. Cheap search, default observed without a loan and a current credit score each look like they close an information gap, and in this market none of them does. Next I move from the borrower to the lender: what lenders learn from the data they hold, how they price when every application is an auction, and how data-sharing rules shape competition when some lenders hold far more data than others.

04 · Education and affiliations

2021 – 2027 · Expected Spring 2027

PhD in Finance, Hanken School of Economics

Graduate School of Finance (GSF), Finland's national doctoral program in finance. Supervisors Camelia M. Kuhnen (UNC Kenan‑Flagler) and Anders Löflund (Hanken).

2025 – 2026

Visiting Scholar, UNC Kenan‑Flagler

Hosted by Camelia M. Kuhnen.

2022 – 2026

Research funding

16 competitive research grants and fellowships.

2005 – 2009

M.Sc. and B.Sc., Hanken School of Economics

Economics and Business Administration.

05 · References

Camelia M. Kuhnen

Professor of Finance, UNC Kenan‑Flagler Business School

Yunzhi Hu

Associate Professor of Finance, UNC Kenan‑Flagler Business School

Anders Löflund

Professor of Finance, Hanken School of Economics

On the 2026–27 job market, open now.
Academic and industry research roles.

Download academic CV