Saving and Wealth Accumulation among Student Loan Borrowers: Implications for Retirement Preparedness

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Saving and Wealth Accumulation among Student Loan Borrowers: Implications for Retirement Preparedness Lisa Dettling, Sarena Goodman, Sarah Reber 2022-019 Please cite this paper as: Dettling, Lisa, Sarena Goodman, and Sarah Reber (2022). “Saving and Wealth Accumulation among Student Loan Borrowers: Implications for Retirement Preparedness,” Finance and Economics Discussion Series 2022-019. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2022.019. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Saving and Wealth Accumulation among Student Loan Borrowers: Implications for Retirement Preparedness* Lisa Dettling Federal Reserve Board Sarena Goodman Federal Reserve Board Sarah Reber Federal Reserve Board February 2022 Abstract Borrowing for education has increased rapidly in the past several decades, such that the majority of non-housing debt on US households’ balance sheets is now student loan debt. This chapter analyzes the implications of student loan borrowing for later-life economic well-being, with a focus on retirement preparation. We demonstrate that families holding student loan debt later in life have less savings than their similarly educated peers without such debt. However, these comparisons are misleading if the goal is to characterize the experience of the typical student borrower, as they fail to account for student borrowers who already paid off their debt. We develop strategies to locate families that ever financed their education with student loans in two large datasets which enables us to draw more meaningful comparisons. We find that student loan borrowers roughly follow the earnings, saving, and wealth trajectories of other college-educated families into late-career ages and are much better off financially than those that did not attend college. * Email: lisa.j.dettling@frb.gov, sarena.f.goodman@frb.gov, and sarah.j.reber@frb.gov. The views expressed herein are those of the authors and do not necessarily reflect the views of the Federal Reserve Board. We thank John Sabelhaus and Alice Volz for generously sharing estimates of defined benefit plan asset values. All errors are our own.

2 Introduction Over the past several decades, student debt has become a large component of US household balance sheets. The implications of this increase for lifecycle wealth accumulation and retirement preparation are ambiguous. Theoretically, the underlying borrowing should support valuable but costly educational investments that potential students otherwise lack the liquidity to make. Even as postsecondary education has increasingly become the norm, the college wage premium has remained substantial and, on average, college graduates enter retirement with more wealth. Still, there are channels through which student borrowers may be left worse off financially. First, there is considerable uncertainty in the labor market return that a given student will see, and the realized return may be insufficient to cover the cost of servicing the debt. Further, the presence of student debt on borrowers’ balance sheets could constrain other investments, such as purchasing a home or starting a business. This paper describes the savings of student borrowers over their working years to shed light on whether and where each of these mechanisms dominates, with emphasis on the measurement challenges that must be overcome. A nascent student loan literature examines early lifecycle outcomes and illustrates this tension in mechanisms. Increased borrowing that results from expanded loan access appears to improve human capital outcomes; for example, Black et al. (2020) find positive effects on educational attainment, earnings, and loan repayment, with little effect on other financial indicators, including homeownership.2 That said, increased borrowing to buffer rising college prices appears to generate some negative downstream effects, such as reduced graduate school enrollment (Chakrabarti et al. 2020) and homeownership (Mezza et al. 2020). 2 Expanded loan access also increases household formation (Goodman, Isen, and Yannelis 2021). A possible exception is borrowing to finance a for-profit education, which primarily serves nontraditional students but is highly debt financed and has notoriously low returns (e.g., Looney and Yannelis 2015; Cellini and Turner 2019).

3 Linking student borrowing to later-life savings and wealth accumulation is difficult, largely due to a lack of data well organized to do so. To date, the best evidence relies on a comparison between otherwise similar young adults around 30 years old based on their student loan borrowing history (Rutledge, Sanzenbacher, and Vitagliano 2021). The study finds that, for those that had borrowed, retirement plan participation is about equal to that of non-borrowers, but balances are lower. Still, early wealth differences need not extend through the lifecycle (e.g., Botazzi, Crossley, and Wakefield 2015); in particular, at younger ages, retirement contributions are typically low, annual earnings are not well-correlated with lifetime income, and borrowers are still paying down their education debt. This study begins to fill this gap by carefully tracing out student borrowers’ lifecycle income and wealth profiles. We analyze two well-known, nationally representative datasets, the Federal Reserve Board’s Survey of Consumer Finances (SCF) and the Federal Reserve Bank of New York Consumer Credit Panel/Equifax (CCP/Equifax). The SCF is a triennial survey of US families with rich detail on their current balance sheet, financial characteristics, and demographics. The CCP/Equifax is an administrative panel drawn from the universe of consumer credit records with rich detail on interactions with credit markets and borrowers’ age and geography. Both datasets amply cover the increased prevalence of student debt over the past couple decades: The SCF’s core questionnaire has been relatively stable since 1989, with the most recent survey conducted in 2019, and the CCP/Equifax covers student loan debt since the early 2000s. We begin our analysis by using the SCF to describe student debt’s lifecycle pattern (figure 1).3 About half of young families have such debt. This share declines gradually with age, starting 3 We primarily analyze the 2016 and 2019 surveys—which added several valuable questions—although the 1989- 2019 surveys produce similar student debt curves. These SCF ‘lifecycle’ pictures reflect age-specific averages across multiple cross-sections and thus may be influenced by cohort differences. In particular, the decrease in the

4 when families are in their mid-30s when standard repayment terms typically end. There is some leveling off among families in their 40s and 50s, but at typical retirement ages, a much smaller fraction—no more than 5 percent—hold debt. This general pattern holds when restricting attention to families that have attended college. Guided by these findings, we compare families with student debt to the general population in four phases of the lifecycle—25-34, 35-44, 45-59, and 60-70 years old. Relative to the population, 60-70-year-old families with student debt display much more disadvantage than other age groups on measures of education, financial literacy, and family background, and the discrepancy grows when restricting attention to families that have attended college. Thus, it appears that families that still have student loan debt at older ages appear to be negatively selected from the population of borrowers. Middle-aged families are not nearly as disadvantaged, but their student loans are much more likely to have financed their children’s education rather than their own. Indeed, removing families that are primarily holding others’ debt from the lifecycle picture steepens the decline in debt-holding among middle-aged families. These lifecycle patterns indicate that families with student loan debt may not be representative of the typical family that financed their education with loans, particularly when families have reached ages at which saving and wealth are more meaningful concepts. To illustrate the central issue, we construct ‘naïve’ lifecycle wealth profiles, differentiating families in each age group only by their education and whether they currently hold student debt (not whether they ever borrowed for education). The analysis indicates that the median wealth of fraction of families with student debt that coincides with age likely reflects a combination of paydowns over the lifecycle and reduced educational attainment and borrowing rates for earlier cohorts (Looney and Yannelis 2014). Government-backed student loans were first available to select students in 1958 (an 18-year-old then would be 79 years old in 2019) under the National Defense Education Act and available broadly in 1965 (an 18-year-old then would be 72 years old in 2019).

5 college-educated families with student debt rises anemically with age—well below the trajectory of college-educated families without debt and in lockstep with those that did not attend college.4 But, factoring into this pattern is relatively well-off families paying off their loans earlier in their careers and exiting the population of families with student debt. This compositional shift implies that we cannot reliably measure student borrowers’ economic outcomes beyond early-career ages from current student loan debt. That said, one can surmise that families that still have such debt at older ages appear to be no better off financially than those who never went to college. The remainder of the analysis seeks to understand the long-run economic position of the typical family that financed their education with loans, drawing upon complementary information in each of our datasets. Within the SCF, since we cannot observe whether families that do not have debt financed their education with loans, we develop a strategy to impute borrowing histories, which we calibrate to cohort-level borrowing rates in historical SCFs. In the CCP/Equifax, we draw on the panel nature of the data and follow individuals who took out loans in their early 20s as they age. Our findings indicate that, on average, student loan borrowers follow the earnings, saving, and wealth accumulation trajectories of other college attendees. They also are similarly likely to participate in a retirement plan, have similar levels of resources in those plans, and report feeling similarly prepared for retirement. If the alternative for these borrowers was no college, they appear to be much better off financially having borrowed. Background 4 Median wages of college-educated families with student debt track those of college-educated families without debt and remain well above families that did not attend college until their mid-to-late 30s; thereafter, they move toward the wages of those that did not attend college.

6 Key facts about student loans. At $1.6T, federal student loan debt is the largest source of nonhousing debt on the household balance sheet. A large majority of these loans were originated under one of two federal lending programs established under Title IV of the Higher Education Act of 1965—the Federal Direct Loan (DL) Program and the (now-defunct) Federal Family Education Loan (FFEL) Program—which grew markedly over time.5 Adjusting for inflation, the Federal programs lent $7.5B in 1970-1971 and $94.1B in 2018-2019 (Ma and Pender 2021). About one-half of US undergraduates rely on loans to help finance their education.6 Undergraduate Stafford Loans, the main loan type offered through the DL and FFEL programs, feature standardized terms and a congressionally set interest rate.7 Unlike other forms of credit, Stafford Loans can be made to any student who meets the basic eligibility criteria for federal financial aid, even those with thin or adverse credit histories.8 Federal student loan programs also exist for graduate and parent borrowers, with the principal differences being they are not subsidized, they are not subject to the statutory limits, and credit history is relevant for eligibility. These programs have grown in importance over time, with each accounting for around 5 percent of 2021 outstanding federal student debt.9 Our analysis attempts to distinguish borrowing under the parent program. Much of the increase in graduate debt, which 5 There is also a small private student loan market, representing less than 10 percent of outstanding student debt. Private student loans do not entail standardized terms and rates and often require an established credit record or a cosigner. Our analysis will not distinguish private student loan borrowing, but the vast majority of students that finance their education with private loans also use federal loans (Department of Education, Digest of Education Statistics 2020: Table 331.60). Thus, our discussion here will focus on federal loans. 6 See https://nces.ed.gov/programs/digest/d19/tables/dt19_331.20.asp 7 Most student borrowers receive a more favorable interest rate than the market would offer them. Stafford Loans come in two varieties: subsidized loans, which are need-based, and unsubsidized loans, which are not. For subsidized loans, interest that accrues early in the life of the loan is paid by the government. Annual Stafford Loan borrowing is subject to a statutory limit that varies with academic level and student dependency status. 8 To receive financial assistance through these programs, students must complete the Free Application for Federal Student Aid (FAFSA), which collects demographic, asset, and income information for students and their households for the calendar year prior to enrollment and computes a student’s Expected Family Contribution (EFC) for college, a key factor in addition to the statutory limit and the cost of college that determines the amount of subsidized loan a student may receive. Continuing students must reapply each year. 9 See https://studentaid.gov/sites/default/files/fsawg/datacenter/library/PortfoliobyLoanType.xls.

7 we cannot distinguish, is held by students taking out both graduate and undergraduate loans (Looney and Yannelis 2015). The standard repayment period for a federal student loan is 10 years, beginning in the year after a borrower graduates or leaves college. Thus, a traditional student that begins college immediately after high school and earns a degree in 4 (6) years is expected to fully repay their loan at around 33 (35) years old. Federal student loan borrowers may enter into alternative repayment plans under certain conditions, in which case the repayment period can be longer. Reflecting the fact that human capital cannot serve as collateral, nonpayment of student loan debt entails different consequences than other types of consumer debt. After 270 days of nonpayment, a federal student loan is in default, and overdue payments can be withheld from tax refunds, federal benefits, and/or wages. It is also relatively difficult to discharge federal student loans in bankruptcy and requires a separate proceeding with more stringent hardship rules. Figure 2 plots average college attendance and borrowing rates by birth year in the 1989-2019 SCFs. Although borrowing for college has become more prevalent over time, college students in older cohorts often took student loans as well. Indeed, 17 percent of all families born between 1958-1962 borrowed, compared with 43 percent of those born from 1991-1995. A large share of the increase—30 percent—reflects increased college attendance; nonetheless, the majority is due to increased reliance on loans to pay for college. Individual saving and the US retirement system. Over the past 40 years the US retirement system has shifted markedly, with the burden and risk of retirement saving moving from the employer to the individual.10 Following the introduction of the 401(k) plan in 1978, employers 10 This section omits a discussion of an important component of the US retirement system – government-sponsored

8 have increasingly replaced traditional defined benefit (DB) pensions with defined contribution (DC) plans (Jacobs et al. 2020). In 1975, 11.2 million private sector workers actively participated in DC plans and 27.2 million participated in DB plan; by 2019, there were over 85 million DC plan participants, compared to just 12.6 million DB plan participants (Myers and Topoleski 2021). DC plans allow the employee to accumulate tax-advantaged retirement savings in an individual account, which is typically funded by a combination of employee contributions and employer contributions.11 The employee has control over how much to contribute and how the funds are invested, and the benefits available to the employee in retirement are a function of the size of contributions and investment returns. In other words, DC plans do not provide guaranteed income, and it is the employee’s responsibility to choose the level of contributions and the investment mix which will determine the level of benefits the employee will be able to access in retirement and how long those benefits will last. By contrast, DB plans provide a steady stream of guaranteed benefits in retirement, and it is the employer’s responsibility to ensure the plan is sufficiently funded to pay the employee some promised benefit.12 In addition to DC plans, individuals can also access tax-preferred individual retirement accounts (IRAs).13 IRAs were initially introduced in 1974 to encourage retirement savings among individuals without pensions, but eligibility was expanded in 1981. They are generally not connected to a job like a DC plan, but most IRAs are funded by assets rolled over from DC plans when individuals change jobs or retire. As with a DC plan, the individual is responsible for social security retirement benefits. We will generally omit social security from our analysis because expected benefits are difficult to estimate (especially at younger ages) and because on average social security only covers about 40 percent of pre-retirement income so that most families will also need to save to plan for retirement. 11 Examples of DC plans include 401(k) plans, 403(b) plans, 457 plans, and Employee Stock Ownership Plans (ESOPs). 12 Some DB plans allow the participant to take the benefits as a lump sum amount at retirement. 13 Keoghs, SEP-IRAs, and other qualified retirement plans are similar to IRAs, except they are used by selfemployed individuals as an alternative to employer-sponsored DC plans.

9 determining contribution levels and how funds are invested. In 2019, about 25 percent of families had an IRA.14 In addition to savings held in tax-preferred, quasi-liquid retirement accounts (e.g., DC plans, IRAs), many families also plan to support their retirement consumption with other types of assets. Financial assets include those that are highly liquid, such as checking and savings accounts, as well as those held in investment accounts, directly-held stocks, bonds, and mutual funds. Nonfinancial assets—such as real estate or closely-held business—are generally illiquid but may still generate income or can otherwise be sold or borrowed against. Of note, the most commonly held nonfinancial asset and largest component of most families’ net worth is a primary residence. In addition to direct channels through which owning a house might support retirement consumption—for instance, via its sale (if a family downsizes) or a reverse mortgage—it also enables families to divert spending that would otherwise go toward housing to other areas. Literature Much of the public discussion of student debt has centered on its potential harmful impacts on borrowers’ future economic well-being, due to imperfect information in the higher education market, the ease at which the loans are obtained, and the relative financial naivete of young adults. For example, in 2013 and 2014, respectively, the New York Times featured articles titled ‘Heavy Load of Student Loan Debt Is Weighing on the Economy, Too’ and ‘Ripple effects from rising student debt.’ This conversation is rarely framed in terms of how student borrowers might have fared in a counterfactual state where they could not borrow or borrow as much as they did. 14 Authors calculation from 2019 SCF.

10 The academic literature has made headway on these questions, and the myriad findings highlight that the circumstances under which borrowing occurs are relevant for its effect on student outcomes, both within and beyond school. In contexts where borrowing is driven by increased access to loans, students generally acquire valuable human capital, which would generally imply improvements in measures of future economic well-being, including savings. The best evidence to date for four-year students, who hold the majority of debt, indicates such borrowing leads to increased college completion and higher earnings, with little, if any, negative effects on financial well-being later in life (Black et al. 2020).15 Student borrowers at community colleges experience analogous benefits, although most of the literature has only examined shorter-run educational outcomes (Marx and Turner 2019; Barr, Bird, and Castleman 2019; Dunlop 2013; Wiederspan 2016; Denning 2019).16 Further, when students that have already completed most of college experience increased loan access, they apply the funds toward graduating earlier, housing, and other important lifecycle investments (Denning 2019; Goodman, Isen, and Yannelis 2021). Nevertheless, student loan borrowing has been associated with worse outcomes in some contexts, which suggests it could weigh on economic well-being. For one, borrowing to buffer price increases appears to negatively affect job choice (Rothstein and Rouse 2011), graduate education (Chakrabarti et al. 2020), entrepreneurship (Krishnan and Wang 2019), and homeownership (Mezza et al. 2020), suggestive of repayment difficulties or credit constraints that might affect saving and investment. Further, students induced to attend for-profit colleges— 15 Gervais and Ziebarth (2019) also provide suggestive evidence that borrowing leads to higher wages. In related work, Solis (2017) finds access to education loans in Chile improves academic outcomes. 16 Even though this sector is large, it is relatively inexpensive and students tend to be less reliant on loans for financing. Calculations that include all students who began their borrowing careers at community colleges—many of whom continue on to a four-year college—imply the sector accounts for only a little over one-tenth of outstanding debt (Looney and Yannelis 2015).

11 which tend to serve nontraditional students and to offer notoriously low labor market returns (e.g., Cellini and Turner 2018; Darolia et al. 2016; Deming et al. 2016; Deming, Goldin, and Katz 2012)—borrow more and see marked increases in default (Armona, Chakrabati, and Lovenheim 2020; Cellini, Darolia, and Turner 2020; Goodman and Volz 2020).17 A comprehensive read of the student loan literature leaves considerable uncertainty about the implications of the rise in aggregate student debt for longer-term savings and preparation for retirement, with only a handful of studies endeavoring to draw a direct connection. The deficit largely reflects a lack of data with broad coverage of the population and student loan borrowing histories that extend beyond young adulthood. To a lesser extent, it also reflects the issue that those that borrowed in the past several decades are still relatively young. Studies that have approached this question find evidence of reduced outcomes (e.g., Rutledge, Sanzenbacher, and Vitagliano 2021; Gicheva and Thompson 2013; Cooper and Wang 2014), but such studies usually suffer from at least one of two measurement issues. The first is that they cannot measure outcomes late enough in a person’s life, when the temporary pressure of student debt payments on spending and saving would likely have resolved and annual earned income is a better predictor of lifetime earnings. The second is that when analyzing older ages, they condition on the presence of student debt, thereby excluding the experiences of those that have already repaid their student loans. Our analysis attempts to overcome these two issues. There is also a small but growing literature on the nexus between consumer debt and retirement preparation. Bleemer et al. (2020) document that the likelihood older families hold debt has been increasing in recent decades, suggesting understanding this connection could grow in importance. Lusardi, Mitchel and Oggero (2018) find evidence that holding debt at older ages 17 This sector constitutes a small fraction of enrollment but one-fifth of outstanding student debt (Looney and Yannelis 2015).

12 is related to a lack of financial literacy and behavioral biases, which they argue is likely to be correlated with inadequate retirement preparation. Still, Butrica and Karamcheva (2013 and 2018) examine how holding debt at older ages affects the timing of retirement and benefitclaiming behavior and find that those with debt work longer and claim social security later, which could offset any negative effects of debt on income during retirement. To our knowledge, this literature does not directly engage on whether debt-financed investments—including, of course, education but also home, business, or vehicle purchases—which could help build wealth or improve labor market opportunities ultimately aids or detracts from retirement preparation.18 Data and Measurement The Survey of Consumer Finances. Our primary data source is the SCF. The SCF is a nationally representative, cross-sectional survey of US families produced triennially by the Federal Reserve Board that collects detailed information on their balance sheets, pensions, income, and demographic characteristics.19 The SCF provides highly reliable statistics on debt and assets broadly distributed in the population, as well as those concentrated in the high end of the wealth distribution. To accomplish this, it employs a two-part sample design—a geographically based random sample and an oversample of higher wealth families—with weights that combine the two samples to describe the full population (Bhutta et al. 2020). Consistent questionnaire design and methodology from 1989-2019 allows meaningful analysis over time. 18 See for example, Baum (2009) and Herbet, McCue, and Sanchez-Moyano (2013) on the importance of a vehicle or home for labor market advancement and wealth accumulation. 19 Detailed information is collected for the household’s primary economic unit (PEU)—the economically dominant single person or couple and all other persons in the household who are financially interdependent with that economically dominant person or couple—and limited information is collected for financially independent adults in the household.

13 Central to our analysis, the SCF asks respondents if their family has student loan debt and their educational attainment. Regarding their student loan debt, the SCF collects information on balance, age, and repayment status (on schedule, behind schedule, forbearance, etc.), among other details. The 2016 and 2019 waves of the SCF include several key questions. Principally, these surveys newly probed for whose education the debt was obtained, allowing for a distinction between debt that financed one’s own or one’s partner’s education (‘own student debt’) and debt that financed someone else’s education, including children’s. They also added questions on parental education, which we use as a proxy for childhood circumstances in our imputation. Due to the importance of these questions, the analysis largely draws from the 2016 and 2019 surveys. Where possible, we supplement the analysis with the larger history and greater statistical power afforded by the 1989-2019 surveys. All dollar values are expressed in 2019 dollars. The SCF contains a number of quantitative and qualitative measures of saving behavior, wealth accumulation and overall financial well-being which allows for a broad perspective on retirement preparation. To measure wealth, we use SCF’s measure of net worth—the difference between families’ gross assets and their liabilities. Due to their importance for retirement, we separately examine access to and participation in employer-sponsored retirement plans, including both DB and DC plans. We also examine other measures of financial well-being and saving behavior, including wage and salary earnings, and whether families reported saving or borrowing, on net, in the previous year. To study savings in assets that are earmarked for retirement, we examine the balances of quasi-liquid retirement accounts, including IRAs and employer-sponsored defined contribution (DC) retirement accounts (e.g., 401(k)-type

14 accounts).20 We also consider other asset classes of interest, such as stock market and housing equity. Finally, we examine families’ expectations for retirement, such as the year they expect to retire and whether they expect their retirement income to be adequate.21 While these features of the SCF make it well-suited to analyze how the presence of student loan debt interacts with retirement preparedness, the survey is not without shortcomings. Most importantly for this analysis, the SCF does not measure student loan borrowing history for families that do not currently have student loan debt, which we will show is an important impediment to understanding whether the later-life economic well-being of families that borrowed for education differs from those that did not. The SCF also does not capture the entirety of aggregate student loan balances (Dettling et al. 2015; appendix table 1).22 Because its sampling frame is household-based, it omits student loan debt held by those in institutional settings (e.g., dormitories). In addition, the SCF does not collect detailed balance sheet information for roommates or live-in relatives, including children, that are financially independent of the respondent, so any student loan debt held by such individuals will be omitted. As a result, our analysis will understate the prevalence of student loan debt, especially among younger families, and likely also somewhat overstate the financial well-being of those that borrowed for college.23 20 The SCF collects the asset value of DC plans but does not collect an analogous measure for traditional defined benefit (DB) pensions, as their market value is less salient to the respondent. For a more comprehensive analysis, we consider an alternative measure of retirement assets that adds imputed DB plan values based on earnings and employment histories from Sabelhaus and Volz (forthcoming). These calculations require stronger assumptions for younger families who will have typically completed a smaller portion of their employment and earnings history. 21 Our paper will not attempt to quantify whether an individual’s retirement income is adequate. For a discussion of retirement adequacy, see, for example, Jacobs et al. 2020. 22 Student debt growth over time in the SCF matches other aggregates (Bricker et al. 2015). 23 Young adults not captured in the data tend to be less well-off (Dettling and Hsu 2014), and student borrowers are more likely to live with their parents and thus will be omitted from the sample of SCF borrowers (Dettling and Hsu 2018, Bleemer et al. 2017).

15 Our analytical choices strive for consistency in the unit of observation, but several key questions—such as race, ethnicity, and financial literacy—are asked only of the respondent, while others—such as employment, pensions, and educational attainment—are asked only of the respondent and their spouse or partner. Further, for simplicity, our analysis differentiates families by age and educational attainment using characteristics of the family reference person.24 Any family in which the reference person attended college, even if they did not obtain a degree, is considered college-educated. Finally, we note one issue with respect to interpretation. Our analysis relies on implied lifecycle profiles for various indicators from the SCF, constructed by deriving either a mean or median value within the relevant age group. We generally interpret these pictures as the evolution of a given indicator, such as student loan debt, over one’s life. In reality, these pictures are not derived from panel data, and they will reflect both the indicator’s lifecycle pattern as well as any differences between cohorts, including with respect to educational, borrowing, and retirement regimes. Restricting attention to families with a college education helps minimize the extent to which increased borrowing rates among more recent cohorts reflect broader access to college over time. Notably, the analysis does not disentangle increased borrowing rates among attendees or borrowing volumes among borrowers, which could influence the pictures as well. Description of CCP/Equifax data. We supplement the SCF analysis with analysis of consumer credit records from the Federal Reserve Bank of New York Consumer Credit Panel/Equifax (CCP/Equifax). The CCP/Equifax is an individual-level panel dataset of anonymized consumer credit reports obtained from Equifax—one of the three main credit bureaus in the United 24 The reference person is the male in a mixed-sex couple or the older person in a same-sex couple, who need not be the survey respondent.

16 States—formed from a 5 percent random sample of all US consumers with credit histories. The data are reported quarterly and include detailed information drawn from credit reports, including loan balances and payment status on mortgages, credit cards, student loans, auto loans, and other miscellaneous debt, as well as year of birth.25 We exploit the panel nature of the CCP/Equifax to follow student loan borrowers and nonborrowers over time. We treat June 2003-June 2007 as the origin and start by restricting the sample to individuals who established a credit record in that period and were between 17 and 20 years old.26 Our final sample includes over 180,000 borrowers. We classify these borrowers according to their educational borrowing. In particular, we distinguish between individuals who originated a student loan around the time they established their record and those who did not.27 We refer to the former group as student borrowers, and the latter as non-student borrowers.28 Because these data do not have information on educational attainment or enrollment, our non-borrower sample will include those that did not attend college and those that attended college but did not borrow. Thus, we further partition the student borrower sample into two groups based on initial borrowing, specifically, whether a student borrower’s first academic year of student borrowing was above or below the median for that year. Following analogous logic to our choice to compare college attendees who did and did not 25 For more information on the CCP/Equifax, see Lee and van der Klaauw (2010). 26 Following our work in Black et al. (2020), we restrict the set of student loan borrowers to individuals whose initial student loans did not exceed the statutory first-year federal borrowing limit to try to isolate traditional students that began borrowing at entry. We also restrict the sample to those that maintained a credit report for at least 10 years. This process removes any incorrectly duplicated records, which can appear (typically for a limited period) when new accounts are opened and have not yet been linked to an existing credit record. That said, because the CCP/Equifax sampling is based on Social Security numbers, it will also remove individuals who pass away over this period or whose credit file is too thin to populate. We focus on cohorts that began borrowing in mid-2003, because that is the first cohort for whom we can reliably observe first-year borrowing in the CCP/Equifax. 27 The CCP/Equifax tends to slightly undercount younger individuals because it takes time to establish a credit record and enter the sample. Lee and van der Klauww (2010) find that by age 25, population estimates from the CCP/Equifax are similar to those from other sources. 28 We exclude individuals that first take a student loan in subsequent years.

17 borrow in the SCF, this focus helps us draw comparisons between similar types of people, save for their level of borrowing. Within-cohort differences in borrowing primarily reflect unmet financial need to cover educational expenses (accounting for grants and own resources), the cost of the school, and individual preferences over borrowing and working during school. We follow these cohorts for up to 16 years from when they are first observed in the data— when they would be between 33 and 36 years old—and analyze credit attributes that measure financial well-being, including student loan balances, combined credit card limit (a proxy for income) and presence of a mortgage (a proxy for homeownership).29 Analysis General trends. Figure 1 plots student debt’s implied lifecycle profile, both for all families (unconditional) and restricting to those that have attended college. The figure includes averages computed from the 2016 and 2019 SCFs, as well as from those spanning the 1989-2019 period. There is a clear pattern. Among all households, younger families are much more likely to have student debt, and there’s a relatively steady decline as families age, such that only a small fraction of families—less than 5 percent—have student debt in retirement ages. This pattern remains upon restricting to families that have attended college, but unsurprisingly, such families are more likely to have debt. That said, differences based on education disappear when families are in their 60s, suggesting that the debt held by older families largely finances someone else’s education. 29 Credit card companies determine credit card limits partly based on borrowers’ incomes and credit scores, which reflect their overall financial well-being. Few young people purchase their homes outright, making the presence of a mortgage a good proxy for homeownership.

18 Comparing statistics generated from the more recent surveys to those from the full series produces several insights into the time series of student loan borrowing. One, zooming in on the youngest families, the propensity to have student debt is quite similar whether one considers the shorter or longer horizon, especially among those that attended college, suggesting conditional borrowing rates are somewhat stable over the full 30-year time series. Second, when leveraging the longer time series, the decline in the propensity to have debt begins at younger ages—which likely reflects several interacting phenomena occurring in more recent years (e.g., longer repayment terms owing to expanded availability of and enrollment in alternative repayment programs; increased amount of debt, conditional on borrowing; increased borrowing by graduate students; and increased repayment difficulties)—and there is a more obvious plateau over the 40s and 50s. The wedge between the two different time horizons becomes almost negligible once families are in their mid-50s and is, on average, smaller for families that have attended college. Finally, the biggest picture takeaway is that the reduced propensity to have debt with age is not exclusively an artifact of cohort differences, as the pattern holds over the full time series. A considerable share of student borrowers pays off their loans by their mid-40s, and most do by their mid-50s. Driven by these findings, we describe characteristics of families with student debt over different phases of the lifecycle using the 2016 and 2019 surveys. We group families into 4 age groups, early-career (25-34 year olds), mid-career (35-44 year olds), late-career (45-59 year olds), and near- (or at) retirement (60-70 year olds). Table 1 describes the prevalence of student debt and college attendance among families in these groups and illustrates some expected patterns. About one-half of early-career families have student debt, and families that went to

19 college are more likely to have student debt. Both college attendance and student debt rates are lower for older segments of the sample, but the decline in student debt is much more dramatic. The left panel of Table 2 characterizes families with student debt in each of these age groups. About one-fifth of families with student debt are Black, and just under one-third received or expect an inheritance. These figures are fairly stable across phases of the lifecycle. Older families with debt are more likely than younger families to have served in the military and to come from a less-educated family. They are less likely to have a graduate degree, but a greater share are financially literate.30 But, these comparisons mask cohort differences and/or natural lifecycle progressions. For example, financial literacy improves with experience. To adjust for these factors, the middle panel characterizes the different age groups in the general population, and the right panel characterizes the differences between the middle and left panels (indebted families). Thus, the right panel can tell us the degree to which families with debt in each age group are differentially selected from the population and whether any selection varies by age. Through this lens, we see financial literacy improves much more over the lifecycle for the population than among families with student debt, leading to a gap in financial literacy between the indebted population and the general population that’s wider for the near-retirement families than early-career families, and even more so when restricting on college attendance. We see similar differences for each of the characteristics we consider, with near-retirement families with debt appearing to be a particularly disadvantaged group with respect to family background, education, and financial literacy. These findings echo those in Lusardi, Mitchell, and Oggero, (2018), which found that those experiencing financial distress near retirement have similar traits. The implication is relevant for 30 Following the literature, a financially literate respondent answers all of the ‘big three’ financial literacy questions correctly (Lusardi and Mitchell 2011).

20 policy and research: families still holding onto student debt as they approach retirement are measurably worse off along observable dimensions than the typical family that relied on student loans to finance their education. Older families with student debt may benefit from targeted policies, but their experience is not generalizable to the entirety of student borrowers, many of whom will no longer have debt at retirement. Table 1 also reveals that the gap in the likelihood that a family’s student debt was for their own education widens with age. Indeed, there is perfect overlap between whether a family has student debt and whether a family has student debt from their own education among the youngcareer families, with the two concepts dovetailing notably beginning at late-career ages. Figure 3 plots the lifecycle profile of ‘own’ student debt and of any student debt for families that have attended college. The decline in student debt with age is steeper when restricting attention to one’s own debt, indicating that in Figure 1, new parent borrowing masks families continuing to steadily pay off their own student loans, particularly at late-career ages. The remainder of the analysis aims to compare saving and student loan debt between those that financed their own education with loans and those that did not. It is common to encounter unqualified descriptions of the economic positions of those with student debt at various segments of the lifecycle, including older ages. While such descriptions can be informative, they do not necessarily belong in the broader debate about whether financing education with loans is ultimately helpful or harmful. To illustrate the central issue, Figure 4a and 4b plot median income and median wealth, respectively, by age for families with student debt that went to college, families without student debt that went to college, and families that did not go to college. Annual earned income for the typical college-educated family that has student debt track that for the typical college-educated family that does not until families are in their late

21 30s, precisely the same age where we start to see a drop-off in the share of families with student debt. Without accounting for the notable share of families that have paid off their debt, we might erroneously conclude that at early-saving ages (i.e., mid-30s), wages of student loan borrowers are not keeping up with those of their peers. This gap grows with age, as more families pay off their debt and those that do not are increasingly negatively selected. By late-career ages, the wages of the typical college-educated family that has student debt essentially merge with those of families that did not go to college. The picture is even more striking when considering wealth, as families that went to college and do not have debt have increasingly more wealth with age than families with debt and families that did not go to college. In fact, the naïve wealth profile is essentially the same for those with debt and those that did not go to college. A simplistic interpretation of this picture, without considering the dynamics behind it, is that families that borrow for college accumulate wealth anemically relative to their college-educated peers that did not borrow, and from a wealth perspective, they are no better off than those that never went to college. What generates this pattern? We already have insight from the companion discussion of earnings that student loan borrowers paying off their loans distort comparisons beyond midcareer ages, but what is happening at early-career ages when the share of families with loans is stable? First, wealth, by definition, nets any student debt the family has, so all else equal, earlycareer student debtors will mechanically have less wealth than those that attended college without borrowing, even if they have similar earnings (as they do). Indeed, the dotted line on the figure plots median wealth for those with debt, excluding student loan debt, and the line better tracks college-educated families without debt through the early-career phase. Further, families with student loan debt must spend some of their earned income servicing their debt and are thus

22 less able to save, which likely exacerbates early differences in wealth. It is quite possible that once the student loan debt is paid off, those that had borrowed are able to save more of their earnings, and their wealth accumulation will rise; in this picture, such families would be included in the calculation of median wealth for those that went to college and do not have debt. This exercise tells us two things. First, college attendees who still hold student loan debt around retirement ages are no better off than their peers who did not go to college. The small share of families in this category appear not to experience the typical wage boost associated with a college education that would enable them to repay their debt and save, leaving them relatively ill-prepared for retirement. Second, it illustrates the importance of properly accounting for the many borrowers that successfully pay off their student loans in understanding the implications of student loan borrowing for economic well-being. The attrition of these borrowers from the set of families with student loan debt, coupled with the increased negative selection with age evidenced in Table 2, will distort comparisons that rely on the observation of current debt alone. Imputing student loan borrowing history from the SCF. Key indicators of economic wellbeing and retirement preparation, such as wealth, lifetime income, and saving behavior, are best measured at ages when many families that borrowed to finance their education would have already paid off their loans. Thus, to be able to examine how the typical family that took student loans fares on these dimensions within the SCF, we need to locate families that paid off their loans. To do this systematically, we develop an algorithm that imputes families’ student loan borrowing histories from their observable characteristics. We first predict the likelihood that families in the 2016 and 2019 SCFs borrowed for college, , using the observed relationship 𝜌𝜌�

23 between own student loan debt and family characteristics among 24-30-year-old families. Families in this age range are unlikely to have paid off their debt or to have acquired debt for others—their student debt rates are stable, high, and coincide with own student debt rates. The family characteristics in the prediction equation are either fully predetermined or at least minimally likely to change beyond 24 years old.31 Families with above some threshold, , are classified as student loan borrowers and those below it as non-bor𝜌𝜌�r owers. 𝜌𝜌̅ The threshold, , is calibrated to cohort borrowing rates derived from the longer 1989-2019 SCF time series, fo𝜌𝜌c̅using on college-educated families for maximum comparability. These ‘true’ rates are obtained by stacking 24-30-year-old, college-educated families from these 11 surveys and calculating the share with student debt by birth year. The oldest cohort in the calibration is 57 years old in the 2019 SCF; thus, the imputation cannot accommodate older age groups.32 Beginning with 0.500, we increment by 0.025 and select the threshold that minimizes the mean squared difference between the imputed rates and truth – which is 0.550. Figure 5 graphs the correspondence between the imputed cohort borrowing rates and those implied by the 1989-2019 SCFs, expressed in 2019 age. For comparison, the figure also includes imputed rates using neighboring thresholds. The imputation does reasonably well approximating truth, especially for younger cohorts. That said, while both series trend downward with age, the imputation does so more gradually and, as a result, modestly overstates borrowing rates for older cohorts. Importantly, the steep decline with age that emerged when relying exclusively on the observation of debt is not visible in the imputation. 31 Specifically, the equation includes indicators for whether a family receives or expects an inheritance, whether at least one of the respondent’s parents completed college, whether the respondent’s mother completed high school, whether the respondent believes borrowing for education is okay, and whether the respondent has served in the military, as well as a set of indicators for the respondent’s race or ethnicity and the reference person’s educational attainment (whereby potential categories are less than high school, high school, some college, or college degree). 32 For a cohort to be included in the calibration, its borrowing rate must derive from more than 150 observations.

24 While the calibration reduces concerns that our imputation is far from the truth, we note two less-desirable properties. First, our prediction equation does not explain very much of the variation in own student loan debt among 24–30-year-old families – just over 13 percent. Second, the imputation assumes that the relationship between a family’s observed characteristics and borrowing history across cohorts is roughly stable. This assumption may not hold if, for example, our prediction equation is missing important independently predictive characteristics (e.g., parental wealth) or if families, conditional on these characteristics, have become more reliant on student loans over time, which based on the calibration appears likely to be at least somewhat true. Comparisons between student loan borrowers and non-borrowers. Figures 6a-6r plot various economic indicators of interest for college-educated families based on their imputed borrowing histories, grouped in 3-year age bins. Annual income from wages and salaries for the typical student loan borrower roughly coincides with that for the typical non-student loan borrower for all ages and is modestly higher in most instances (Figure 6a). Importantly, there is no meaningful divergence with age. What we see for wealth reflects the similar patterns for income, especially once the wealth measure accounts for differences in student loan debt (Figure 6b-6c). In general, wealth accumulates at similar rates for the typical borrower and the typical non-borrower. At younger ages, median wealth among borrowers is slightly lower than for non-borrowers. The gap narrows upon exclusion of student debt from the calculation of wealth, but some remains, potentially owing to income diverted from savings toward servicing the debt. These differences accumulate

25 somewhat over the lifecycle, with the gap between borrowers and non-borrowers growing over mid-career ages; nonetheless, the gap disappears by late-career ages. Including an estimate for the value of DB pension assets raises wealth and steepens implied accumulation over the lifecycle for both groups (Figure 6d-6e). That said, the gap at mid-career ages widens, which is consistent with findings that student borrowers are less likely to enter public sector jobs (Rothstein and Rouse 2011), where DB plans are more commonly offered, and indicates retirement consequences of student loan borrowing can emerge through features of jobs beyond earnings. Focusing on retirement plans, there is little distinction between borrower and non-borrower families in the likelihood they participate in an employer-sponsored retirement plan (Figure 6f), nor whether their employer offers such plans (Figure 6g). Thus, the findings from Rutledge, Sanzenbacher, and Vitagliano (2021) appear to extend to later ages. There is some evidence that borrowers are more likely to own a quasi-liquid retirement account like a 401(k) or IRA (Figure 6h), and have more money saved in such accounts (Figure 6i), but the differences are tiny. Furthermore, once we consider the combination of estimated DB pension assets and quasi-liquid retirement account balances, non-borrowers have higher balances, on average (Figure 6j). Turning to other types of assets and portfolio allocation decisions, stock ownership–which includes stocks that are either directly-held or held in a retirement or investment account—is somewhat higher for borrowers (Figure 6k), as are balances (Figure 6l), which could suggest a preference among borrowers toward holding their wealth in riskier assets. Indeed, while homeownership rates are similar between borrowers and non-borrowers (Figure 6m), median home equity is, on balance, a bit higher for non-borrowers (Figure 6n). Similar homeownership between borrowers and non-borrowers is consistent with findings in Black et al. (2020) and from

26 the CCP/Equifax (shown in the next section). Finally, median total financial assets are extremely similar between the two groups (Figure 6o). Overall, it seems that average portfolios of borrower families are slightly more balanced toward riskier assets than non-borrower families. The question probing expectations for retirement income yields a noisy picture, although closer to retirement age, it seems borrower families, on average, feel more financially prepared (Figure 6p). That said, the average borrower family plans to retire later than the average nonborrower family (Figure 6q), which is consistent with Butrica and Karamcheva (2013 and 2018) and may be a factor in this belief. Finally, differences in wealth accumulation are well reflected in savings behavior reported by borrower and non-borrower families (Figure 6r). Non-borrower families are much more likely to report saving through mid-career ages, on average, which is a phase over which the wealth gap widens. Borrower families are more likely to do so in latecareer ages, when the gap disappears. All in all, the imputed lifecycle comparisons imply income and wealth follow similar trajectories for college-educated borrowers and non-borrowers, and differences in any given indicator are quite minimal. Together with evidence from Figure 4, the analysis suggests borrowers would be financially worse off forgoing the education that borrowing afforded them. Interestingly, borrowers appear to balance their portfolio with modestly more risk—they hold slightly more of their wealth in stocks and account-based retirement plans and less in housing and DB plans. On more subjective metrics, our comparisons indicate that as they approach retirement age, the saving rate of borrower families is higher than that of similarly educated nonborrower families and they feel better prepared for retirement.

27 Evidence from the CCP/Equifax. An important limitation in using the SCF to study lifecycle outcomes is that the time series is formed from repeated cross-sections rather than observing the same individuals over time. To try to disentangle cohort differences from true lifecycle differences, we turn to the CCP/Equifax, which offers a shorter time series but allows us to corroborate our earlier findings for several key outcomes. Figure 7 graphs a) average student loan balances, b) average credit card limits, and c) proportion with a mortgage by years since college entry.33 These outcomes are plotted for three groups based on their initial level of student loan borrowing: those with above-median borrowing, those with below-median borrowing, and those with no borrowing. Figure 7a plots average student loan balances.34 By construction, the non-borrowers have an average balance of zero throughout, while above-median borrowers initially have a higher average balance than below-median borrowers. Throughout the analysis horizon, these initial differences persist, but the slopes change over time. Average balances among above-median borrowers grow quickly over the first decade and decline thereafter (around 27-30 years old), whereas average balances among below-median borrowers do not grow as quickly over the first decade and stabilize thereafter. In other words, above-median borrowers eventually start to pay off their debt at a faster rate than below-median borrowers, suggestive that increased borrowing is associated with higher returns to education that better permit borrowers to pay down their principals. This interpretation is consistent with Black et al. (2020). 33 Technically, we use the year the credit record was established, which is a proxy for college entry within our sample of individuals whose credit record was established at traditional college entry ages. 34 We also examined the propensity to have a positive student loan balance. The results are similar to the mean balances: the higher balance sample has more debt throughout, but the two groups appear to be converging and by 16 years out, are about equally as likely to have paid off their loans completely.

28 Figure 7b plots average credit card limits, a proxy for income. Initially, average limits exhibit similar levels and growth rates for all three groups, reflecting lengthening credit records. Over time, average limits among above-median borrowers exhibit more growth than the other two groups, so that by 15 years out, their average limits are well above those of the other two groups, which remain fairly similar to one another. Consistent with panel a, this pattern suggests that above-median borrowers see increased returns. Finally, Figure 7c plots the share of each group that has a mortgage, a proxy for homeownership. Homeownership rates grow over time for all groups. However, initially, nonborrowers are more likely to be homeowners than borrowers, and below-median borrowers are more likely to be homeowners than above-median borrowers. Then, about 8 years out (around 25-28 years old), the pattern reverses and above-median borrowers become more likely to own homes. Through the remainder of the analysis horizon, their homeownership rates grow faster than the other two groups. This suggests that above-median borrowers are less likely to purchase a home as they complete their schooling and begin to service their debt, but they quickly catch up and ultimately surpass their peers that borrowed less, likely reflecting the higher labor market returns suggested by our other metrics. Overall, our analysis of the CCP/Equifax corroborates the pattern of results from the SCF and provides evidence that borrowing for education is associated with better financial positions. Conclusion The accumulation of wealth over the lifecycle feeds into the retirement experience along several important dimensions, including the resources one has in retirement, the age one stops working, and health. In the United States, the retirement system currently places much of the

29 burden and risk of saving and planning for retirement on the individual, a shift from an earlier era in which the government and employers bore most of this responsibility. Against this backdrop, this chapter has examined whether the rise in student loan borrowing over the past several decades may have impinged on families’ ability to save and accumulate wealth over the lifecycle and to prepare for retirement. Our analysis delivers several findings which may be useful for setting policy in this area. We find that student loan borrowers tend to follow the lifecycle earnings and savings trajectories of other similarly educated families into late-career ages. Both groups are wealthier entering retirement than those that didn’t attend college at all. Thus, if the appropriate counterfactual is that in the absence of the student loan program student loan borrowers would have lacked the resources to attend college, they are better off having borrowed on nearly all dimensions. Overall, student loans appear to, on average, be financing valuable human capital investments that translate into long-run improvements in families’ economic well-being, suggesting that the program is largely having its intended effect. Given this, policies aimed at improving retirement preparation that target resources toward reducing student loan balances or increasing saving among student borrowers are likely to be regressive. Still, we demonstrate that the small subset of student borrowers who carry student debt into near-retirement ages are, on average, observably worse off than the full population of student loan borrowers and much less prepared for retirement than their peers. Such families likely have other factors contributing to lower saving and wealth accumulation and are ultimately not comparable to their peers with similar education that have repaid their student loan debt. Highly targeted support for this small population—for example, loan forgiveness at a certain age threshold—could be a cost-effective policy to improve retirement security for this small group.

30 From a methodological perspective, our analysis is instructive on the pitfalls of statistics that derive from observed student debt beyond early-career ages. We demonstrate that such statistics will not generalize to the experience of the typical student loan borrower and could lead to misleading conclusions about the role of student debt in financial well-being. Finally, we note two caveats. First, our statements concerning retirement preparedness and economic well-being more generally reflect student loan borrowers’ relative positions rather than their absolute position. It is possible that the different groups that drive our comparisons, even on average, have different expectations for their retirement consumption and/or realization of consumption needs in retirement. Further, all of our findings are predicated on conditions today that may change as younger cohorts who borrowed in different conditions age toward retirement. References Armona, L., R. Chakrabarti, and M. Lovenheim (2020). ‘Student Debt and Default: The Role of For-Profit Colleges.’ FRB of NY Staff Report No. 811, Rev. Oct. 2021. New York, NY: Federal Reserve Bank of New York. Barr, A., K. Bird, B. Castleman, and W. Skimmyhorn (2019). ‘A comparison of postsecondary outcomes for Army service members, Veterans, and civilians.’ EdWorking Paper 19-50. Providence, RI: Brown University. Baum, C. L. (2009). ‘The effects of vehicle ownership on employment’ Journal of Urban Economics 66 (3):151–63. Black, S.E., J. T. Denning, L. J. Dettling, S. Goodman, and L. J. Turner (2020). ‘Taking it to the limit: Effects of increased student loan availability on attainment, earnings, and financial

31 well-being.’ Working Paper No. w27658. Cambridge, MA: National Bureau of Economic Research. Bleemer, Z., M. Brown, D. Lee, and W. van der Klaauw (2017) ‘Tuition, Jobs, or Housing: What’s Keeping Millenials at Home?.’ Federal Reserve Bank of New York Staff Reports No. 700. New York, NY: Federal Reserve Bank of New York. Bleemer,Z.,M. Brown, D. Lee, and W. van der Klaauw (2020). ‘The Graying of American Debt’ In O.S. Mitchell and A. Lusardi, eds., Remaking Retirement: Debt in an Aging Economy. Oxford, UK: Oxford University Press. Bhutta, N., J. Bricker, A. C. Chang, L. J. Dettling, S. F. Goodman, A. H. Volz, J. W. Hsu, K. B. Moore, S. Reber, and R. Windle (2020). ‘Changes in US family finances from 2016 to 2019: Evidence from the Survey of Consumer Finances.’ Federal Reserve Bulletin 106, no. 5. Washington, DC: Board of Governors of the Federal Reserve System. Bottazzi, R., T. F. Crossley, and M. Wakefield (2015). ‘First‐time house buying and catch‐up: a cohort study.’ Economica 82: 1021-1047. Bricker, J., M. Brown, S. Hannon, and K. M. Pence (2015). ‘How much student debt is out there?.’ FEDS Notes 2015-08. Washington, DC: Board of Governors of the Federal Reserve System. Butrica, B. A., and N. S. Karamcheva (2013). ‘Does household debt influence the labor supply and benefit claiming decisions of older Americans?.’ CRR WP 2013-22. Chestnut Hill, MA: Center for Retirement Research at Boston College. Butrica, B. A., and N. S. Karamcheva (2018). ‘In debt and approaching retirement: Claim social security or work longer?.’ In AEA Papers and Proceedings, vol. 108, pp. 401-06.

32 Cellini, S. R., R. Darolia, and L. J. Turner (2020). ‘Where do students go when for-profit colleges lose federal aid?.’ American Economic Journal: Economic Policy 12(2): 46-83. Cellini, S. R., and N. Turner (2019). ‘Gainfully employed? Assessing the employment and earnings of for-profit college students using administrative data.’ Journal of Human Resources, 54(2): 342-370. Chakrabarti, R., V. Fos, A. Liberman, and C. Yannelis. (2020) ‘Tuition, debt, and human capital.’ FRB of New York Staff Report 912. New York, NY: Federal Reserve Bank of New York. Cooper D. H, and J. C. Wang (2014). ‘Student Loan Debt and Economic Outcomes.’ Current Policy Perspectives No. 14-7. Boston, MA: Federal Reserve Bank of Boston. Darolia, R., C. Koedel, P. Martorell, K. Wilson, and F. Perez-Arce (2016). ‘Race and gender effects on employer interest in job applicants: new evidence from a resume field experiment.’ Applied Economics Letters, 23(12): 853-856. Deming, D. J., C. Goldin, and L. F. Katz (2012). ‘The for-profit postsecondary school sector: Nimble critters or agile predators?.’ Journal of Economic Perspectives, 26(1): 139-64. Deming, D. J., N. Yuchtman, A. Abulafi, C. Goldin, and L. F. Katz (2016). ‘The value of postsecondary credentials in the labor market: An experimental study.’ American Economic Review, 106(3), 778-806. Denning, J. T. (2019). ‘Born under a lucky star financial aid, college completion, labor supply, and credit constraints.’ Journal of Human Resources, 54(3), 760-784. Dettling, L. J., and J. W. Hsu (2014). ‘The State of Young Adults' Balance Sheets: Evidence from the Survey of Consumer Finances.’ Federal Reserve Bank of St. Louis Review, 96(4). St. Louis, MO: Federal Reserve Bank of St. Louis.

33 Dettling, L. J., and J. W. Hsu (2018). ‘Returning to the Nest: Debt and Parental Co-Residence Among Young Adults.’ Labour Economics, 54: pp. 225-236. Dettling, L. J., S. Devlin-Foltz, J. Krimmel, S. Pack, and J. P. Thompson (2015). ‘Comparing micro and macro sources for household accounts in the United States: evidence from the Survey of Consumer Finances.’ FEDS Working Paper No. 2015-86. Washington, DC: Board of Governors of the Federal Reserve System. Dunlop, E. (2013). ‘What do Stafford Loans actually buy you? The effect of Stafford Loan access on community college students.’ National Center for Analysis of Longitudinal Data in Education Research Working Paper 94. Washington DC: CALDER, American Institutes for Research. Gervais, M., and N. L. Ziebarth (2019). ‘Life after debt: Postgraduation consequences of federal student loans.’ Economic Inquiry 57(3): 1342-1366. Gicheva, D., and J. Thompson (2015). ‘The effects of student loans on long-term household financial stability.’ In B. Hershbein and K. M. Hollenbeck, eds., Student loans and the dynamics of debt. Kalamazoo, MI: W. E. Upjohn Institute for Employment Research, pp. 287-316. Goodman, S., A. Isen, and C. Yannelis (2021). ‘A day late and a dollar short: Liquidity and household formation among student borrowers.’ Journal of Financial Economics 142(3): 1301-1323. Goodman, S. and A. H. Volz (2020). ‘Attendance spillovers between public and for-profit colleges: Evidence from statewide variation in appropriations for higher education.’ Education Finance and Policy, 15(3): 428-456.

34 Herbert, C. E., D. T. McCue, and R. Sanchez-Moyano (2013). ‘Is homeownership still an effective means of building wealth for low-income and minority households?(Was it ever?).’ Homeownership Built to Last 10(2): 5-59. Jacobs, L., E. Llanes, K. Moore, J. P. Thompson, and A. H. Volz (2020). ‘Wealth distribution and retirement preparation among early savers.’ Working Papers No. 20-4. Boston, MA: Federal Reserve Bank of Boston. Korkki, P. (2014) ‘The Ripple Effects of Rising Student Debt.’ New York Times. May 25: BU6. Krishnan, K., and P. Wang (2019). ‘The cost of financing education: can student debt hinder entrepreneurship?.’ Management Science 65(10): 4522-4554. Lee, D. and W. van der Klaauw (2010). ‘An Introduction to the New York Fed Consumer Credit Panel’ Federal Reserve Bank of New York Staff Reports 479. New York NY: Federal Reserve Bank of New York. Looney, A., and C. Yannelis. (2015) ‘A crisis in student loans?: How changes in the characteristics of borrowers and in the institutions they attended contributed to rising loan defaults.’ Brookings Papers on Economic Activity no. 2. Washingto, DC: Brookings Institution. Looney, A., and C. Yannelis (2019). ‘How useful are default rates? Borrowers with large balances and student loan repayment.’ Economics of Education Review 71: 135-145. Lowrey, A. (2013) ‘Heavy Load of Student Debt is Weighing on the Economicy, Too.’ New York Times May 11: B1. Lusardi, A and O. Mitchell (2011) ‘Financial literacy and planning: Implications for retirement wellbeing,’ NBER Working Paper No. 17078. Cambridge, MA: National Bureau of Economic Research.

35 Lusardi, A., O. S. Mitchell, and N. Oggero (2018). ‘The changing face of debt and financial fragility at older ages.’ AEA Papers and Proceedings 108: pp. 407-11. Ma, J. and M. Pender (2021). ‘Trends in College Pricing and Student Aid.’ New York, NY: College Board. Marx, B. M., and L. J. Turner (2019). ‘Student loan nudges: Experimental evidence on borrowing and educational attainment.’ American Economic Journal: Economic Policy 11(2): 108-41. Mezza, A., D. Ringo, S. Sherlund, and K. Sommer (2020). ‘Student loans and homeownership.’ Journal of Labor Economics 38(1): 215-260. Myers, E. and J. Topoleski. (2021). ‘A visual depiction of the shift from defined benefit (DB) to defined contribution (DC) pension plans in the private sector.’ In Focus. Washington, DC: Congressional Research Service. Rothstein, J., and C. E. Rouse (2011). ‘Constrained after college: Student loans and early-career occupational choices.’ Journal of Public Economics 95(1-2): 149-163. Ruttledge, M., G. Sanzenbacher, and F. Vitagliano (2016). ‘How Does Student Debt Affect Early- Career Retirement Saving?’ CRR WP 2016-9, Rev. 2018. Chestnut Hill, MA: Center for Retirement Research at Boston College. Sabelhaus, J, and A. H. Volz (forthcoming). ‘Social Security Wealth, Inequality, and Lifecycle Saving.’ In Measuring and Understanding the Distribution and Intra/Inter-Generational Mobility of Income and Wealth. NBER Book Series Studies in Income and Wealth. Chicago, IL: The University of Chicago Press. Solis, Alex (2017). ‘Credit access and college enrollment.’ Journal of Political Economy 125(2): 562-622.

36 Topoleski, J. and E. Myers (2021). ‘Worker Participation in Employer-Sponsored Pension: Data in Brief.’ CRS Report R43439. Washington DC: Congressional Research Service. Wiederspan, M (2016). ‘Denying loan access: The student-level consequences when community colleges opt out of the Stafford loan program.’ Economics of Education Review 51: 79-96.

37 Figure 1: Share of SCF Families with Student Debt 1.0 0.8 0.6 0.4 0.2 0.0 19212325272931333537394143454749515355575961636567697173757779 unconditional (1989-2019) college-educated (1989-2019) unconditional (2016-19) college-educated (2016-19) Note: This figure plots the share of families with student loan debt and the share of collegeeducated families with student loan debt in the SCF by age of the household reference person. College-educated refers to families in which the reference person attended at least some college. The surveys used in the calculation are denoted in the legend.

38 Figure 2: College Attendance and Borrowing Rates by Cohort Average rates among 24-30 year olds in 1989-2019 SCFs 100% Borrowing Attendance 80% 60% 40% 20% 0% 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 Note: This figure plots the share of families that attended college and the share of families with student loan debt in the SCF by cohort, derived from families in the 1989-2019 surveys whose household reference person was between 24 and 30 years old.

39 Figure 3: Share of College-Educated Families with Student Debt by Whose Education It Financed 1.0 0.8 0.6 0.4 0.2 0.0 19212325272931333537394143454749515355575961636567697173757779 Any student debt Student debt for own education Note: This figure plots the share of college-educated families with student loan debt and the share of college-educated families with student loan debt whose largest student loan financed the respondent’s (or their partner’s) education in the 2016 and 2019 SCFs by age of the household reference person. College-educated refers to families in which the reference person attended at least some college.

40 Figure 4a: Median Wages by Student Debt 120,000 90,000 60,000 30,000 0 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 College-educated, no own student debt College-educated, own student debt No college, no own student debt Figure 4b: Median Net Worth by Student Debt 650,000 550,000 450,000 350,000 250,000 150,000 50,000 -50,000 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 College-educated, no own student debt College-educated, own student debt College-educated, own student debt (measured ex. student debt) No college, no own student debt Note: This figure plots (a) the median annual salary and wage income and (b) the median net worth of families in the 2016 and 2019 SCFs based on their education and own student loan debt status by age of the household reference person. College-educated refers to families in which the reference person attended at least some college, and own student debt refers to families with student loan debt whose largest student loan financed the respondent’s (or their partner’s) education. The dotted line in figure b excludes student loan debt from net worth.

41 Figure 5: Student Loan Borrowing by College-Educated Families Share of families that borrowed, imputed vs. truth 1.0 0.8 0.6 0.4 aging forward 24-30 year olds from 1989-2019 SCFs 0.2 imputed from 2016-2019 SCFs using 0.575 threshold imputed from 2016-2019 SCFs using 0.550 threshold imputed from 2016-2019 SCFs using 0.525 threshold 0.0 25 30 35 40 45 50 55 Age in 2019 Notes: This figure plots student loan borrowing rates for college-educated families in the SCF by 2019 age, derived by averaging borrowing rates by cohort among 24-30-year-old families in the 1989-2019 SCFs or by the imputation described in the text. For the imputation, the solid line denotes the best-fit threshold used in the analysis and the dashed and dotted lines denote the two next-best-fit thresholds. College-educated refers to families in which the reference person attended at least some college.

42 Figure 6

43

44

45 Notes: This figure plots various economic indicators for college-educated families in the 2016 and 2019 SCFs by imputed student loan borrowing history and age of the household reference person in 3-year age bins. College-educated refers to families in which the reference person attended at least some college.

46 Figure 7. Credit report characteristics by student borrower status Notes: Source is CCP/Equifax. Panels display means of each outcome across the three groups (above-median first year student borrowing, below-median first year student borrowing, and no student borrowing) by year since entry into the sample.

47 Table 1: Student debt and attendance, by age group Has student Has own Attended Age debt student debt college Unconditional 25-34 45% 45% 69% 35-44 34% 33% 66% 45-59 20% 13% 63% 60-70 17% 7% 61% Total 30% 26% 65% Attended college 25-34 55% 55% -- 35-44 42% 41% -- 45-59 25% 17% -- 60-70 20% 10% -- Total 38% 34% -- Source: 2016 and 2019 Survey of Consumer Finances.

48 puorg ega yb ,seilimaf fo noitisopmoC :2 elbaT noitalupop dna seilimaf detbedni neewteb ecnereffiD seilimaf llA tbed tneduts htiw seilimaF rehtoM ro ecnatirehnI rehtoM ecnatirehnI rehtoM ro ecnatirehnI detelpmoc etaudarG yllaicnaniF detcepxe detelpmoc etaudarG yllaicnaniF detcepxe ro detelpmoc etaudarG yllaicnaniF detcepxe loohcs hgih eerged etaretil etihW kcalB yratiliM ecnatirehni loohcs hgih eerged etaretil etihW kcalB yratiliM ecnatirehni loohcs hgih eerged etaretil etihW kcalB yratiliM ecnatirehni slevel noitacude llA %5 %1 %3 %4 %2 %1- %0 %48 %2 %73 %55 %71 %8 %62 %88 %4 %04 %95 %91 %7 %72 43-52 %7 %1 %2 %4 %5 %1 %4 %08 %4 %24 %65 %61 %9 %52 %68 %6 %44 %16 %12 %01 %92 44-53 %4 %0 %1 %5- %6 %4 %4- %47 %4 %74 %36 %51 %41 %03 %87 %4 %84 %85 %12 %71 %62 95-54 %5- %4- %1 %11- %6 %5 %31- %27 %5 %05 %96 %41 %31 %93 %76 %2 %15 %85 %02 %91 %62 07-06 %7 %1 %1 %0 %5 %0 %1- %87 %4 %34 %95 %61 %11 %82 %48 %4 %44 %95 %02 %11 %72 latoT egelloc dednettA %0 %1 %1- %1- %4 %2- %2- %09 %3 %54 %06 %51 %9 %92 %09 %4 %34 %95 %91 %7 %72 43-52 %0 %0 %4- %1- %6 %0 %1 %88 %7 %15 %26 %61 %11 %92 %98 %7 %74 %26 %22 %11 %03 44-53 %0 %2- %3- %8- %8 %3 %6- %28 %6 %75 %96 %41 %51 %63 %28 %5 %45 %16 %12 %81 %03 95-54 %01- %6- %8- %71- %31 %31 %22- %08 %8 %16 %47 %9 %31 %74 %07 %2 %35 %75 %22 %62 %42 07-06 %1 %0 %5- %4- %6 %1- %4- %68 %6 %25 %56 %51 %21 %23 %78 %5 %74 %06 %12 %21 %82 latoT .secnaniF remusnoC fo yevruS 9102 dna 6102 :ecruoS

49 Appendix Table 1: Student Loan Aggregates by Age ($T) 18-29 30-39 40-49 50-59 60-69 70+ TOTAL FRBNY Household Debt and Credit Report (2019:q1) 0.35 0.5 0.32 0.21 0.09 0.02 1.49 SCF (2019) 0.24 0.43 0.21 0.15 0.07 0.01 1.11 difference 0.11 0.07 0.11 0.06 0.02 0.01 0.38 <=24 25-34 35-49 50-61 62+ Nonreport All Federal Student Loan Portfolio (FY2019:Q2) 0.12 0.50 0.56 0.23 0.07 0.00 1.48 SCF (2019) 0.05 0.41 0.41 0.17 0.06 #N/A 1.11 difference 0.08 0.08 0.14 0.06 0.01 #N/A 0.37 Sources: 2019 Survey of Consumer Finances; page 21 of the supplemental data file from the FRBNY Household Debt and Credit Report 2019:Q1 publication; the US Department of Education Federal Student Loan ‘Portfolio by Age’ file accessible at: https://studentaid.gov/sites/default/files/fsawg/datacenter/library/Portfolio-by-Age.xls.