Loading today’s stories
Science1 distinct publisher2 min readPublished
A forthcoming Management Science paper models pay-later as a liquidity lever rather than a payments feature, and finds it moves list prices and stocking levels for every buyer.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
The stocking half of the result has the cleanest mechanism, and it is arithmetic rather than psychology. Klarna, Affirm and Afterpay pay the retailer at checkout and keep a portion of each transaction [7]. A stockout costs a retailer the margin it would have earned on the unit that was not there. When more of those units come back as the sticker minus a provider's cut [7], the loss avoided by carrying one more unit is smaller, and the case for safety stock weakens with it. Nothing about the shopper's behaviour has changed. The payoff table on the retailer's side has.
Worth putting the scale in one place. The Richmond Fed's $70 billion estimate, spread across LendingTree's 91.5 million users, works out to roughly $765 per user for the year, about $64 a month [18]. If that $70 billion is close to 1 percent of US card spending, the card base is on the order of $7 trillion [19]. The financed channel is small. What makes it reach past its own users is that the variable it moves is the posted price, and the posted price is quoted to everybody who walks in.
The household side has drifted off discretionary goods. Roughly a third of users report using installments for groceries, and a growing share for rent and bills [14]. About 41 percent missed at least one payment in the previous year, according to data cited in the paper [15]. Because these loans sat largely outside credit reporting, a borrower could hold several at once with no single lender seeing the total, which the authors label phantom debt [16].
This is a model, circulating as an SSRN preprint ahead of the journal version [2], and what it yields are directions rather than measured elasticities. That is still enough to move where the argument sits. Read as a credit product, pay-later is a story about the borrower's own balance sheet [16]. Read as a pricing input, it adds a party to the ledger who never opened an account and cannot decline the terms [9].
Ranked by verification strength, evidence, and original report placement.
The paper "Buy Now, Pay Later: The Hidden Effects of Consumer Liquidity on Retail Prices and Inventories," by Naveed Chehrazi, Panos Kouvelis and Wenhui Zhao of Olin Business School at Washington University in St. Louis, is forthcoming in the journal Management Science and examines how pay-later financing changes consumer and retailer behaviour.
The paper is currently available on the SSRN preprint server.
The Federal Reserve Bank of Richmond estimated that pay-later purchases totalled about $70 billion in 2025, roughly 1% of US credit-card spending.
The store effectively serves full-price customers and financed customers from the same inventory; to offset the financing company's fee the retailer may raise its sticker price, meaning cash-paying customers can end up subsidising customers who use pay-later.
Across millions of simulated scenarios, the researchers found that accepting pay-later could lead retailers to stock less inventory because a lost sale becomes less costly to bear.
About 41% of pay-later users missed at least one payment in the previous year, according to data cited in the report.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One press summary of a simulation-based preprint
Every claim traces to a single publisher relaying a university research summary. The core pricing and inventory findings are explicitly simulation outputs from a paper that is still forthcoming, with no disclosed parameters, no retailer sample, and no observed market prices. Market-size and behaviour figures are attributed to named third parties (Richmond Fed, LendingTree) but only secondhand, and the 41% missed-payment statistic has no named dataset. The FICO scoring change is the single crisply verifiable institutional fact.
Mass consumer usage, thin share of credit
Adoption of the underlying practice is broad and quantified: an estimated 91.5 million US users in 2025, about $70 billion in volume growing roughly 20% annually, and a spread into groceries, rent and bills. It is nonetheless roughly 1% of credit-card spending and a small slice of consumer credit, and the strongest institutional adoption signal is FICO folding pay-later history into lender scores. Adoption of the paper's specific retailer conclusions is unmeasured — no retailer is shown changing pricing or stocking policy in response.
Modestly overstated relative to model-only evidence
The framing — pay-later repricing the shelf so cash buyers pay more and find less on it — is a causal market claim resting on simulated scenarios and a conditional pass-through mechanism, with no observed price or stock-level data. The same source undercuts the alarm level by noting pay-later remains about 1% of card spending and that the Richmond Fed saw no clear evidence of broader financial stress as of early 2026. The gap is real but bounded: the market statistics and the FICO change are concrete, and the researchers themselves are quoted resisting both miracle and crisis readings.
Institutional research promotion, one voice
The material originates as a business-school research communication promoting its own faculty's forthcoming paper and named endowed chair and research centre, then republished by an aggregator; that structure rewards a sharp, counterintuitive takeaway. The commercially interested parties — the pay-later providers whose fees are the mechanism — are named as subjects but given no voice, and no independent expert is quoted. Mitigating the read: the source names external estimators and includes findings that soften its own thesis.
Low: single publisher, unpublished paper
Confidence is limited by structure rather than plausibility. One publisher, one press-summary source, a paper not yet in its version of record, mechanism claims from undisclosed simulations, and one key statistic with no named dataset. The market-scale figures and the FICO scoring change are the parts most likely to hold; the pricing and inventory conclusions should be treated as testable hypotheses pending the published paper.
invest
Klarna's $1bn Quarter: Margin Dollars Up 42%, Revenue Up 27%, And That Gap Is The Story2 distinct publishers
product
Apple's Upgrade Program is a Klarna lease, and the phone is not yours at month 241 distinct publisher
invest
Your Landed Cost Is Being Litigated By Companies With $306,000 Problems1 distinct publisher
leadership
Gartner says agents aren't ready; 60% of companies plan to deploy them anyway1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
phys.org
1 article · August 25, 2026