Paper Cuts: Quick Findings and Ideas from Stanford GSB Insights

Welcome to the first edition of Paper Cuts, a roundup of recent stories highlighting innovative, impactful, and just plain interesting research from Stanford Graduate School of Business faculty. To read many more stories like these, head over to GSB Insights.

What If We’re Looking at the National Debt All Wrong?

The ratio of outstanding government debt to annual gross domestic product has become the primary measure of a country’s indebtedness. Yet Jonathan Berk, a professor of finance at Stanford GSB, has a bone to pick with the debt-to-GDP ratio.

In a recent article, Berk and Jules van Binsbergen of Wharton argue that the measure is fundamentally flawed because it compares a “stock” (debt) to a “flow” (GDP). “There really is no theory that says this is the correct measure,” Berk says. For example, Argentina defaulted at 40% debt-to-GDP while Japan’s 250% debt-to- GDP raises no default concerns. And while U.S. debt-to-GDP has tripled from 40% to 120% over recent decades — an all-time high —the U.S. federal government still borrows at low rates and the stock market remains strong.

Berk proposes two alternative measures: debt-to-equity (comparing government debt to total wealth) and interest-to-GDP (comparing annual interest expense to GDP).

When applied to the U.S. and 19 other large economies over 125 years, these measures show flat trajectories — nothing like debt-to-GDP’s steady climb since 1980. When looking at the sustainability of the United States’ debt regime, Berk suggests asking a different question. “I think the focus should be on the budget deficit,” he says. “Are we collecting more than we are spending?” — Alexander Gelfand

The Best Way to Sell a Rotisserie Chicken

When a major retailer found it was discarding 9% of its freshly prepared food items, it went looking for solutions. It consulted Erica Plambeck, professor of operations, information, and technology; Dan Iancu, an associate professor of operations, information, and technology; and Jae-Hyuck Park, PhD ’21, from the New Jersey Institute of Technology.

The researchers built a model to find the most efficient, least wasteful way to sell prepared foods. The results challenged conventional wisdom. While supermarkets typically use a first-in, first-out (FIFO) inventory rule, selling the oldest items first, the team found that last-in, first-out (LIFO) works better. “It’s true that some portion on the back of the shelf may expire,” Plambeck says. “But LIFO increases the average quality of items sold. That leads to increased demand and, paradoxically, less gets thrown out.”

The model revealed another surprise: Removing time stamps from prepared foods reduces waste. When items are time-stamped, customers cherry-pick the newest ones, leaving older products unsold. The combination of LIFO plus hidden time stamps is crucial, Plambeck explains: “If you do one without the other, it doesn’t work.”

Plambeck says the project gave her a new appreciation of what it takes to run a grocery store. “Next time you’re in a supermarket, stop and look around. It’s a complex operation, and the details really matter for the cost and quality of the food we eat — and for what we don’t eat.” — Lee Simmons

What It Would Cost to End Extreme Poverty

Around 700 million people live on less than $2.15 per person per day. How much would it cost to lift them out of extreme poverty? Paul Niehaus, the co-founder of GiveDirectly, believed the answer lies in direct cash transfers — simply giving people enough to cross the poverty threshold. But calculating the cost of such an endeavor proved remarkably complicated.

Niehaus discussed the problem with Susan Athey, PhD ’95, a professor of economics at Stanford GSB. She recommended Stefan Wager, an associate professor of operations, information, and technology, who in turn took the problem to Stanford computer science doctoral student Roshni Sahoo. And, Wager says, “Roshni solved it.”

Sahoo’s model built on a statistical method developed by Athey and Wager that incorporates observational data and real-world constraints to determine who should benefit from a certain policy. According to her calculations, reducing global poverty to 1% would cost $318 billion annually — just 0.3% of global GDP. In contrast, providing universal basic income at the poverty line would cost $895 billion per year.

“I’ve been working in the space of largescale data-driven policy learning for a while,” Wager says. “Sometimes you get the impression that the main application area for these methods is ad targeting. But our results here highlight how, given access to high enough quality data, policy learning methods can also help make a difference in public-interest settings.” — Aimee Levitt

How Technology Can Evolve Without Breaking Our Brains

When Boeing’s new B-17 crashed during a 1935 test flight, the problem wasn’t mechanical failure or poor training — the aircraft was simply “too much airplane for one man to fly.” The solution? A simple safety checklist that enabled pilots to fly the bomber 1.8 million miles without accident. Now, checklists are a routine part of aviation as well as fields such as surgery and construction.

The Boeing crash is a prime example of how a technology can fail when it requires more brainpower than a single person can muster. “Tech can get too cumbersome to be used,” explains Helena Miton, an assistant professor of organizational behavior. “This is a huge constraint on whether tech systems can evolve or not. It doesn’t matter how great the tech is. If it’s too complicated to operate, it won’t work.”

In a recent paper, Miton explores how “cognitive load distribution” enables people to manage increasingly complex technologies. With her coauthor, she traces cognitive distribution from 19th-century looms that used punch cards to modern project management software. These tools distribute mental labor across people and systems, though careful coordination remains essential.

Another strategy for cognitive distribution requires limiting or filtering how much information users have access to. For example, simplified controls enable almost anyone to operate a car, and user-friendly interfaces like Claude and ChatGPT make generative AI accessible to people who do not understand the algorithms working behind the scenes. Yet these interfaces also widen the gap between being able to use a technology and knowing how it works. “It’s a tradeoff,” Miton says, but a worthwhile one. — Sara Harrison

Why Even Big-Name Brands Can’t Stop Advertising

Why do household-name auto insurance companies like Geico and Progressive spend over $1.5 billion annually on advertising?

Navdeep Sahni, a professor of marketing, and doctoral candidate Yifan Yang go a long way toward answering this question in a recent working paper that shows the dramatic benefit of advertising. They found that ads not only boost visits to an advertiser’s website but also interfere with consumers’ recall of alternative brands, dislodging the competition from their minds. The findings help explain why spending on repetitive campaigns, even by well known brands, is essential to remaining in people’s memories.

The researchers showed banner ads for an auto insurer to over 325,000 website visitors in a single day. (A control group saw unrelated public service announcements.) The results were dramatic: Visits to the insurer’s website jumped 300% on the day of the experiment. While the advertiser’s traffic surged, competitors experienced an 11% decrease in visits the following day, suggesting that its ad had displaced rival brands.

The research also showed that competitive advertising creates a combative cycle. When competitors advertised heavily beforehand, the banner ad’s effect was actually stronger—the new ad reclaimed mental space previously captured by rivals. “This creates an incentive to advertise more, which is one of the key implications of our work,” Sahni explains. “This is a combative environment.” — Dylan Walsh