College Opportunity Requires Better Underwriting, Testing, and Information

Caroline Hoxby frames college finance around what can reasonably be known at different stages of a student’s education. At the outset, a student may not know whether college is a good fit or whether they will succeed. Grants preserve what economists call “option value”: the value of being able to try an investment whose payoff and personal fit remain uncertain. A student who discovers early that college is not working should be able to leave without carrying a large debt burden.

That logic implies that grants should be front-loaded. By a student’s fourth year, Hoxby argues, both academic progress and likely repayment capacity are clearer. There is less remaining value in subsidizing an experiment, and more basis for lending against an educational investment.

Loans, in her view, also give students “skin in the game.” She frames this as a moral-hazard problem: students who receive money without a repayment obligation may have less reason to put in effort, choose courses carefully, or treat an expensive education as a serious investment. The analogy is venture capital, where investors can fund an entrepreneur’s early and uncertain experiment, then seek an equity stake or financial return once the project becomes more legible.

Human capital is not ordinary collateral, however. A lender cannot repossess a person’s education; Hoxby says that would be slavery. Government therefore has a role that it would not have in mortgage lending. She supports mechanisms that acknowledge this constraint: earnings garnishment to enforce repayment, income-based repayment that concentrates repayment in years of higher earnings, forgiveness for incapacitated borrowers, and forgiveness programs for public-service work such as teaching or public defense.

The issue is not whether borrowers need those protections. It is whether the system should continue to ignore information that becomes available after students have begun college.

Caroline Hoxby defines underwriting as estimating the probability that a borrower can repay a particular loan on its terms. Mortgage lenders routinely consider income, job stability, household circumstances, and the asset being purchased. Student lending, she argues, does not apply an equivalent discipline.

Instead, borrowers face a common federal rate despite sharply different probabilities of default. Hoxby’s presentation sorts institutions from highly selective nonprofit baccalaureate colleges to for-profit certificate programs. In her reading of the chart, default probability is close to zero among students at the most selective colleges and exceeds 30% among students at the least selective institutions.

The chart’s central point is not simply that students have unequal outcomes. A single loan price redistributes across risks. Hoxby says students at selective institutions pay interest rates above the actuarially fair rate—the rate that would generate equal expected profit across loans—while students at institutions with higher repayment risk pay less than that rate.

Hoxby says the information for better underwriting already exists. In a study of roughly a quarter-million U.S. students, she observed secondary-school grades and admissions-test scores, high schools, the colleges to which students applied and were admitted, their enrollment decisions, and their progress through college. Using regression and machine-learning methods, she says, the analysis reached predictive accuracy of about 70% in identifying who would not repay a student loan—better, in her comparison, than predicting mortgage repayment.

This is why she treats blanket loan forgiveness as poorly targeted. The largest balances, she says, often belong to people who attended highly selective colleges or pursued graduate degrees such as MBAs and law degrees. Those borrowers are also least likely to default in the displayed selectivity data. Broad forgiveness therefore reaches not only low-income students who struggled after attending less selective institutions, but also borrowers Hoxby expects to earn enough over their lifetimes to repay.

The accountability she proposes extends to institutions. Hoxby points to Brazil, where she says colleges can be held responsible when they consistently produce students who do not find employment, do not earn after graduation, or remain non-employed. That arrangement gives schools reason to ensure students learn something with labor-market value and to advertise outcomes more candidly. She says it has worked very well in Brazil.

Her proposed repair combines underwriting, government oversight, institutional risk-sharing, and potentially better loan servicing by private banks. The throughline is that public support should cover uncertainty that students cannot reasonably manage alone, while later evidence should shape the obligations of students, lenders, and colleges alike.

Caroline Hoxby makes a parallel argument about college admissions. Tests and information matter, she says, because they reveal ability that might otherwise remain obscured by family background, school prestige, or familiarity with the application process.

The usual rationale for test-optional admissions or test bans is to improve socioeconomic diversity and reduce barriers for students from disadvantaged families. Hoxby argues that the effect can run the other way. Students from ordinary schools need a portable way to show they are academically strong. When tests are removed or de-emphasized, applicants compete more heavily through credentials that affluent families can more readily shape: coached essays, outside help with writing, and sophisticated letters of recommendation.

Hoxby traces the argument to the postwar expansion of U.S. admissions testing. Before World War II, she says, college admissions used relatively little standardized testing. The broad adoption of tests afterward made admissions more meritocratic and less dependent on wealth, family privilege, or attendance at a prestigious secondary school. Colleges drew more students from poorer and rural counties, more students from public schools rather than selective private boarding schools, and a more socioeconomically diverse student body.

Family background still correlates with attendance at expensive and prestigious institutions, she says. But Hoxby argues that this relationship has declined since the postwar adoption of admissions testing, in substantial part because tests give students from less advantaged schools a legible signal of ability.

Test optionality can also create an asymmetric strategic response. Hoxby says students from more sophisticated, higher-income families may continue looking for ways to submit scores or otherwise strengthen applications, while lower-income students may take “optional” literally and decline to submit a score that could distinguish them. She cites experiments by other economists in which students changed their application behavior when told a school would be test optional or would ban tests; students from more sophisticated families became more strategic to a greater extent.

Tests can help applicants understand their own academic match as well. A student in the top 10% of the score distribution may learn that a nonselective college is not the best curricular or peer fit. For a high-achieving, low-income student, Hoxby adds, a highly selective institution can even cost less than a local community college because wealthy universities have greater endowment resources for aid.

Caroline Hoxby describes the Expanding College Opportunities study, conducted with University of Virginia economist Sarah Turner, as evidence that many qualified students lack information rather than academic capacity.

The randomized controlled trial included about 45,000 students and focused on low-income, high-achieving students. Participants received customized information about where they stood in the college-ability distribution and the kinds of institutions at which they would be academically typical rather than unusually strong or weak.

The intervention changed application behavior. Hoxby says the number of colleges to which participants applied rose by 31%. The probability of admission to a “peer college”—one where a student’s aptitude resembled that of other students—rose by 78%.

Participants subsequently attended colleges that spent more on their education and had higher graduation rates, Hoxby says. The researchers now also have post-college outcomes for the group. The study’s influence, she says, can be seen in state policy: Michigan has adopted a program intended to identify high-achieving students statewide and tell them they can attend the University of Michigan, with admission largely automatic unless they fail to complete the rest of the application.

Related work is underway for middle-achieving students across income levels. That group is larger and faces more genuine tradeoffs among institutions. For the highest-achieving low-income students, Hoxby’s conclusion is more straightforward: more selective institutions may offer a better academic match, stronger peers, and lower net cost.

The intervention reinforces her broader objection to test bans. College opportunity does not become fairer simply by withholding signals. Students need credible ways to demonstrate what they can do and practical information about the institutions for which they are qualified.

Caroline Hoxby does not present existing admissions tests as sufficient. Everyone dislikes aptitude tests, she says, partly because taking them is unpleasant. But she separates dislike of testing from an argument against testing itself.

Current U.S. admissions tests predict college outcomes better than other commonly used factors, including high-school grades, according to Hoxby. Their predictive power nevertheless remains limited. In a regression using tests alone, she says, explanatory power is about 15%, leaving 85% of students’ college outcomes unpredicted by test scores.

That limitation is a reason to improve testing, not to abandon it. Hoxby identifies computer-adaptive testing as one potential improvement: it could measure a broader range of college aptitude more precisely and make cheating harder, though not impossible. Test providers can also observe information they currently do not use, including how many practice tests students take and how often they sit for an exam.

She points to the University of Cambridge as a precedent. Dissatisfied with the United Kingdom’s secondary-school A-level examinations, Cambridge devised its own test, and other highly selective British universities adopted it. In Hoxby’s telling, admissions tests improve when universities and test designers work directly together on what institutions actually need to predict.

Her criticism of the SAT, ACT, and their governing organizations is that this collaboration has weakened. The tests remain more informative than many alternatives, she argues, but are no longer closely aligned with the outcomes colleges and universities want to assess.