Dinerstein, Yannelis, Chen, Debt Moratoria: Evidence from Student Loan Forbearance (NBER w31247; AER: Insights 6(2) 2024, 2023-05)
Summary
NBER Working Paper 31247, "Debt Moratoria: Evidence from Student Loan Forbearance," dated May 2023, by Michael Dinerstein, Constantine Yannelis, and Ching-Tse Chen. The abstract evaluates the effects of the 2020 student debt moratorium that paused payments for student loan borrowers. It states that the payment pause led to a sharp drop in student loan payments and delinquencies for borrowers subject to it, as well as an increase in credit scores. It states that borrowers used the new liquidity to increase borrowing on credit cards, mortgages, and auto loans rather than avoid delinquencies. The paper describes student loans as the second largest source of household debt in the United States in 2020, with an approximate $1.7 trillion outstanding.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“Dinerstein, Yannelis, Chen, Debt Moratoria: Evidence from Student Loan Forbearance (NBER w31247; AER: Insights 6(2) 2024, 2023-05).”
Original at www.nber.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/dinerstein-yannelis-chen-debt-moratoria-evidence-from-student-loan-for-e53b38b121/
, SHA-256 e17807816c1742868eb866c65fc7aa6403875cfe5e4b67b339a7cb26b4e98454.Full text
NBER WORKING PAPER SERIES
DEBT MORATORIA: EVIDENCE FROM STUDENT LOAN FORBEARANCE
Michael Dinerstein
Constantine Yannelis
Ching-Tse Chen
Working Paper 31247
http://www.nber.org/papers/w31247
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
May 2023
We are grateful to numerous colleagues for helpful comments. Samuel Earnest provided superb
research assistance. Constantine Yannelis is grateful to the Becker Friedman Institute and Fama
Miller Center for generous financial support. All tables and figures that list TransUnion as a
source have statistics calculated (or derived) based on credit data provided by TransUnion, a
global information solutions company, through a relationship with the Kilts Center for Marketing
at the University of Chicago Booth School of Business. The views expressed herein are those of
the authors and do not necessarily reflect the views of the National Bureau of Economic
Research.
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.
© 2023 by Michael Dinerstein, Constantine Yannelis, and Ching-Tse Chen. All rights reserved.
Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission
provided that full credit, including © notice, is given to the source.
Debt Moratoria: Evidence from Student Loan Forbearance
Michael Dinerstein, Constantine Yannelis, and Ching-Tse Chen
NBER Working Paper No. 31247
May 2023
JEL No. G51,I22
ABSTRACT
We evaluate the effects of the 2020 student debt moratorium that paused payments for student loan
borrowers. Using administrative credit panel data, we show that the payment pause led to a sharp drop
in student loan payments and delinquencies for borrowers subject to the debt moratorium, as well as
an increase in credit scores. We find a large stimulus effect, as borrowers substitute increased private
debt for paused public debt. Comparing borrowers whose loans were frozen with borrowers whose
loans were not frozen due to differences in whether the government owned the loans, we show that
borrowers used the new liquidity to increase borrowing on credit cards, mortgages, and auto loans
rather than avoid delinquencies. The effects are concentrated among borrowers without prior delinquencies,
who saw no change in credit scores, and we see little effects following student loan forgiveness announcements.
The results highlight an important complementarity between liquidity and credit, as liquidity increases
the demand for credit even as the supply of credit is fixed.
Michael Dinerstein Ching-Tse Chen
Department of Economics University of Chicago
University of Chicago chingtse.chen@chicagobooth.edu
1126 East 59th Street
Chicago, IL 60637
and NBER
mdinerstein@uchicago.edu
Constantine Yannelis
Booth School of Business
University of Chicago
5807 S. Woodlawn Avenue
Chicago, IL 60637
and NBER
constantine.yannelis@chicagobooth.edu
1 Introduction
During economic disruptions, governments may seek to stimulate the economy and smooth
consumption by making transfers to households. Economists have had a longstanding debate
regarding whether and how to make such transfers. If the disruption is transitory, then house-
holds may primarily require liquidity to raise short-term consumption or insure against adverse
shocks.1 Policymakers may then turn to debt moratoria, which inject immediate liquidity with-
out large long-term fiscal costs for the government. If the disruption has permanent effects,
however, then households may require compensation for permanent income losses in addition
to liquidity. Policymakers may then turn to direct cash transfers, which address both liquidity
and permanent shocks, but at higher fiscal cost.
In this paper, we study the effects of debt moratoria, the lower cost option, on borrowing,
consumption, and loan repayment during times of economic distress. Our empirical context is
the 2020 student loan payment freeze in the U.S. that led to a complete stoppage of student
loan payments for most borrowers. Student loans were the second largest source of household
debt in the United States in 2020, with an approximate $1.7 trillion outstanding. As part of
relief during the 2020 coronavirus pandemic, the federal government ordered a temporary
pause in student loan payments, aimed at relieving households from debt burdens. This pause
was subsequently extended until the end of 2022. We study the effects of the payment pause
by using historical features of the federal student loan program: due to the ownership of loans,
a subset of borrowers saw no change in payments. We use administrative credit panel data and
compare borrowers who saw payment freezes to those who had to continue paying down their
loans. The payment freeze led to a sharp drop in student loan payments and delinquencies for
borrowers subject to the debt moratorium, as well as an increase in credit scores.
Our paper makes three central contributions. First, we evaluate whether debt moratoria
increase consumption during a period of economic distress. We estimate that the student debt
payment pause immediately increased consumption, as borrowers used the new liquidity to
increase borrowing on credit cards, mortgages, and auto loans rather than avoid delinquencies.
Second, we study the mechanisms explaining this effect and argue that the effects are driven
1
Liquidity may prevent large aggregate losses in the presence of externalities arising from household balance
sheets (Mian, Rao and Sufi, 2013) or debt overhang (Donaldson, Piacentino and Thakor, 2019).
1
by an interaction between liquidity and credit demand, rather than credit supply or balance
changes. The estimated effects are concentrated among borrowers without prior delinquencies
who likely saw no change in their credit supply. Further, we find no evidence that borrowers
reacted to expected changes in balances from a student loan forgiveness announcement. The
results thus highlight an important complementarity between liquidity and credit, as liquidity
increases the demand for credit. Third, we compare the relative policy effectiveness of debt
moratoria versus the promise of direct transfers. We estimate the student debt payment pause
has a larger effect on immediate consumption than the proposed balance discharge, which
suggests that less costly liquidity-targeting policies may be more effective than direct transfers.
Our empirical strategy uses variation in the payment stoppage driven by historical details
affecting legal loan ownership. Prior to 2010, the federal government operated two nearly
identical loan programs, the Direct Loan (DL) program and the Federal Family Education Loan
(FFEL) program. The programs were identical in all loan terms; the only difference was the
source of the funds and ultimate ownership of loans. Under the DL program, the source of
funds was the US Treasury, while for the FFEL program the source of funds was private banks,
which carried guarantees from the government. In 2020, the federal government was only
able to legally pause payments for federally owned loans, under the DL program, and could
not pause payments for FFEL loans under the guarantee program. This generated essentially
random variation for borrowers from cohorts which received loans under both the FFEL and
DL programs, with most borrowers likely being unaware of the source of funds for their federal
loans.
To examine the effects of the payment pause, we compare DL and FFEL borrowers in similar
cohorts who received different exposure to the payment pause. For DL borrowers their pay-
ments drop, on average, by $138 relative to FFEL borrowers following the policy announce-
ment. We use a standard difference-in-difference approach, and study borrowing and loan
delinquency behavior following the policy announcement. We use national and comprehen-
sive administrative data from a large credit bureau, TransUnion. The data comprise a panel
sample consisting of one in ten individuals with a credit history in the United States.
Consistent with the policy, we find a sharp reduction in payments for borrowers subject to
the payment pause, which leads to a sharp rise in balances. By the end of 2021, borrowers
2
subject to the payment freeze had an additional $1,500 in outstanding student loan balances
relative to those that did not see a payment stoppage. Borrowers subject to the pause are also
0.8 percentage points less likely to be delinquent on their student loans, and have significantly
higher credit scores. We see little effect, however, on delinquencies for non-student loans.
Despite having higher cash on hand, borrowers do not use their additional liquidity to pay
down other debt. In fact, household leverage rises as borrowers make higher payments on
other loans, and mortgage, auto, and credit card debt rise. Overall, excluding student loans,
household leverage increases by $1,200 (3%) for households subject to the pause. Student
loan balances increase by a similar amount. The results suggest that the payment pause led to
higher durable and non-durable consumption in the short term, but higher overall leverage,
consistent with binding liquidity constraints.
We find that the effects – except for the increased credit scores – are concentrated among
borrowers who have not previously been delinquent on a loan. These borrowers likely had
no change in their supply of credit. Instead, their increased use of credit implies that their
demand for credit increases as they have more liquidity. For instance, these borrowers may
have required liquidity to make down payments or initial monthly payments after taking out
new loans. This results suggests a complementarity between liquidity and use of credit.
We then estimate the effects of a second policy change that targets loan balances. In August,
2022 Biden Administration issued executive an executive order to cancel between $10,000 and
$20,000 in student loan debt.2 Like the payment pause, this policy only applied to DL borrow-
ers, not FFEL borrowers. We estimate whether the policy announcement affected borrowers’
use of credit and find no impact.
Our results provide new evidence on debt moratoria, a commonly used policy during eco-
nomic crises.3 First, we estimate large short-term effects. That households increase their con-
sumption in response to the liquidity shock indicates that policies targeting liquidity may be
particularly effective in helping households maintain consumption levels and stimulating the
2
As of December 2022, this proposal is undergoing legal challenges and is blocked by courts.
3
While we focus on a particular application, temporary forbearance programs pausing payments during crises
have existed since at the least the Great Depression (Rose, 2011), and have continued during recent downturns.
Along with the national student loan payment pause we study, the CARES act included provisions to pause mort-
gage payments. Several states such as California also paused mortgage payments in 2020. Cherry et al. (2021)
provide an overview of pandemic relief policies. The 2008 financial crisis also saw several state and federal
programs pausing loan payments, such as HAMP.
3
economy during periods of economic distress. Second, we find that debt moratoria have long-
run effects through households’ increase credit use. This effect may be positive if households
enjoy higher future consumption from making productive investments or durable purchases;
the effect may be negative if households face effects from future debt overhang.
The results also relate to the ongoing policy debate over how to provide relief to student
loan borrowers. The effects of student loan burdens have received considerable policy atten-
tion, with policy proposals ranging from full forgiveness to more modest capped proposals.
Much of the ongoing policy debate centers around the effects of student loan payments on
other credit outcomes, and whether these effects are driven by the need for liquidity, or due
to the effects of high debt burdens hindering borrowing through damaging credit scores or
otherwise lessening the ability to borrow. For example the August 2022 executive order can-
celling student debt balances noted high monthly payments and ballooning balances that make
it harder for them to build wealth, like buying homes, putting away money for retirement, and
starting small businesses. The announcement further noted that high student debt burdens and
delinquencies were lowering credit scores, with adverse impacts. On the one hand, if students
have short-term liquidity needs, policies effectively extending maturity terms, such as Income-
Driven Repayment, can alleviate burdens. On the other hand, if students have a longer term
structural inability to pay, then discharging debt, a costlier option from a fiscal perspective,
may be important. Our results are consistent with decreased liquidity as the key constraint on
student loan borrowers, such that the less costly student debt policies may still be effective.
This paper primarily joins a finance literature on household leverage that studies the im-
pacts of liquidity and related government policies on debt and consumption outcomes. A large
body of work focuses on times of crisis and studies the response of consumption and debt to
government crisis policies, such as loan relief (Agarwal, Amromin, Ben-David, Chomsiseng-
phet, Piskorski and Seru, 2017), household leverage (Mian and Sufi, 2011), monetary policy
(Di Maggio, Kermani, Keys, Piskorski, Ramcharan, Seru and Yao, 2017), and quantitative eas-
ing (Di Maggio, Kermani and Palmer, 2020). Several papers have also studied how debt and
consumption respond to income and wealth shocks (Agarwal, Liu and Souleles, 2007; Mian
and Sufi, 2009; Agarwal and Qian, 2014; Baker, 2018; Aydin, 2022). Aydin (2021) studies the
effect of loan forbearance on delinquency in a debt relief experiment, Fiorin, Hall and Kanz
4
(2023) study how debt moratoria increase interactions with the same lender, and Ganong
and Noel (2020) study how liquidity versus balance relief affect loan repayment. Indarte
(forthcoming) also finds that liquidity plays an important role in household bankruptcy de-
cisions. Other work studies consumption, debt, and employment responses to wealth shocks
(Chodorow-Reich, Nenov and Simsek, 2021) and firm responses (Giroud and Mueller, 2017).
This paper contributes by providing new evidence on the effects of temporary loan payment
pauses, a common policy pursued during recessions. The fact that we find large increases in
borrowing suggests that liquidity constraints are important as opposed to wealth effects. The
absence of results regarding loan delinquency in other credit markets also suggests that liq-
uidity is not an important driven of loan default, at least in the period studied. Finally, our
results caution that temporary debt relief may lead to greater household leverage in the future
and associated concerns with financial stability and debt overhang (Donaldson, Piacentino and
Thakor, 2019; Mian, Rao and Sufi, 2013).
This paper also joins a rapidly growing literature on student loans, the second largest source
of household debt in the United States.4 Specifically, this study analyzes a major policy inter-
vention in the student loan market. Previous work has focused on the effects of loan forgive-
ness (Catherine and Yannelis, 2022; Di Maggio, Kalda and Yao, 2019), loan limits (Black et
al., 2020; Goodman et al., 2021), alternative repayment plans (Mueller and Yannelis, 2019,
2022; Herbst and Hendren, 2021), or maturity extension (Boutros, Clara and Gomes, 2022).
Amromin and Eberly (2016), Lochner and Monge-Naranjo (2016), and Yannelis and Tracey
(2022) provide reviews of the literature. Several recent papers look at the same payment
pause, with a focus on distributional effects (Briones, Powell and Turner, 2023), financially
distressed borrowers (Chava, Tookes and Zhang, 2023), and the flypaper effect (Katz, 2023).
This paper contributes to this area in two ways. First, we provide real evidence on the effects of
student loan payments on consumption and debt outcomes.5 Second, our results speak to the
ongoing debate on student debt forgiveness. Liquidity from extra payments has large effects
on outcomes even in the absence of debt forgiveness, and lower payments may be offset by
4
In a narrow sense, this paper also joins a literature on guaranteed versus direct lending, which largely focuses
on smaller business rather than household lending (Erel and Liebersohn, 2022; Granja, Leuz and Rajan, 2022a;
Granja, Makridis, Yannelis and Zwick, 2022b).
5
Chakrabarti, Gorton and Lovenheim (2020) study the effects of state appropriations for higher education on
similar outcomes.
5
further borrowing.6
The remainder of this paper is organized as follows. Section 2 discusses institutional details
regarding student loans in the United States, and our empirical strategy to capture the effects
of the payment pause. Section 3 introduces the data used in the paper. Section 4 presents the
main results and discusses their implications. Section 5 concludes.
2 Institutional Details and Empirical Strategy
2.1 Student Loans
In January 2020, there were approximately 45 million student loan borrowers, with an out-
standing balance of over $1.7 trillion. This was the second largest source of household debt in
the United States, after mortgages, and unlike other consumer loans, the federal government
directly disburses or guarantees the vast majority of student loans. The average student loan
borrower holds $37,693, conditional on borrowing, and repayment typically begins six months
after separating from school. Prior to 2009, the vast majority of federal student loans were in a
ten-year fixed payment plan, called the Standard Plan. Since 2009, students have increasingly
enrolled in income-driven repayment (IDR) plans. By 2017, one quarter of students and half
of balances were in IDR (Karamcheva, Perry and Yannelis, 2020).
Modern federal student loan programs were established with the Higher Education Act in
1965. Congress sets interest rates and limits, and students take loans through financial aid
offices at their respective institutions. Initially, private banks provided capital for students
loans, with parameters set by the government under the FFEL Program. These loans were
owned by lenders and guaranteed by the government; in the case of default, lenders were
reimbursed. In 1992, the DL program was created under which loans were directly made
by the US Treasury. For many years, both the FFEL and DL programs existed side by side, and
colleges could choose the source of funds. Despite the different source of funds and ownership,
FFEL and DL loans were otherwise identical in terms of limits, interest rates, and repayment
6
Chava, Tookes and Zhang (2023) also look at consumption and debt outcomes, but with a focus on financially
distressed borrowers for whom the credit supply shock is largest. Katz (2023) compares moratoria to direct
transfers and finds the flypaper effect from the moratorium lowers the marginal propensity to consume.
6
plans available, and many students likely were unaware of the different source of funds when
taking loans through their schools’ financial aid offices. In 2010, following analysis and a
bribery scandal of college administrators, the Health Care and Education Reconciliation Act of
2010 eliminated new loans under the FFEL program.7 Since 2010, all federal student loans
have been made under the DL program.
2.2 2020 Payment Pause
The CARES act enacted a payment pause on March 27, 2020, as part of relief during the
2020 COVID-19 pandemic. The payment pause included a suspension of loan payments, the
forgiveness of interest, and the ending of collections on defaulted loans. Initially set to last
three months, the payment pause was subsequently extended via executive action several times
through June 30, 2023.
The pause applied to the majority of borrowers, but many borrowers were ineligible as
a result of the ownership status of loans through the history of the DL and FFEL programs.
Due to legal authority, only loans owned by the Department of Education were eligible for
the payment pause. All loans made under the DL program were owned by the Department of
Education, and thus eligible, along with a smaller number of FFEL program loans which were
bought by the Department of Education.8 FFEL program loans, being owned by banks, were
ineligible for the payment pause despite the fact that these loans were made under federal
programs and had identical terms to loans made under the DL program. Approximately 12
million FFEL loan borrowers in repayment saw no pause in payments, while 25 million DL
borrowers in repayment saw a pause in payments due to the student loan forbearance relief.
2.3 2022 Loan Forgiveness Announcement
While our focus is on the payment pause, a second policy targeted DL borrowers. In August
2022, the White House announced substantial loan forgiveness for many borrowers through an
7
See the New American Foundation for a discussion of the elimination of the FFEL program.
8
In the first quarter of 2020, there was approximately $250 billion outstanding of lender held FFEL loans,
while $88 billion FFEL loans were held by the Department of Education. The remaining volume of federal loans
are predominantly in the DL program. Federal Student Aid provides more information on loan status, and The
Department of Education provides further information on eligibility for the payment pause.
7
executive order. Most student loan borrowers where scheduled to receive $10,000 in forgive-
ness, while borrowers who received Pell grants were promised $20,000 in forgiveness. Like the
earlier forbearance, this action only applied to DL borrowers, as White House lawyers deemed
that the President did not have the legal authority to cancel FFEL loans due to their ownership
structure. The announcement was widely publicized and reported in the media. By the end of
2022, the policy was not yet implemented following successful legal challenges.
2.4 Empirical Strategy
The payment pause’s eligibility rules suggest a difference-in-difference empirical strategy. We
compare how outcomes differentially evolve for borrowers eligible for the payment pause (DL
borrowers) versus those ineligible (FFEL). Let i index borrowers and t index month-year. We
model outcome yi t as:
yi t = β Eli gi blei Post t + µi + τ t + εi t , (1)
where El i gi blei is an indicator for whether i is a DL borrower, Post t is an indicator for whether
the month is March, 2020 or later, and µi and τ t are individual and time fixed effects, respec-
tively. In all specifications, we cluster our standard errors at the level of the borrowing cohort –
the year when the borrower took out her last student loan – as average eligibility varies across
borrowing cohorts. When clustering at the individual level, standard errors are predictably
smaller.
Our identification assumption is that in the absence of the payment pause, outcomes for
the DL and FFEL borrowers would have followed parallel trends. To assess the existence of pre-
trends and to trace out how the treatment effects vary over time, we also estimate a dynamic
difference-in-difference specification. Our estimating equation is:
X
yi t = β t Eli gi blei + µi + τ t + εi t , (2)
t
where we include 12 months prior to the policy change and 18 months after the policy change.
We normalize the coefficient for December, 2019 to zero such that all treatment effects are
relative to levels a few months before the COVID pandemic reached high levels in the US.9
9
A recent literature has found that two-way fixed effect estimators of difference-in-difference specifications can
8
Our specifications’ additive structure may be a poor fit for our context for two reasons.
First, because the FFEL program was eliminated in 2010 while the DL program has continued,
eligible borrowers are likely to be at different points in the lifecycle from ineligible borrow-
ers. If lifecycle effects are nonlinear, then a parallel trends assumption is unlikely to hold.
Further, many FFEL borrowers are old enough that any current student loan balance implies
non-standard repayment plans or past delinquency or forbearance. We address these concerns
by restricting our sample of eligible (DL) borrowers to those whose most recent student loan
was opened in 2010 or earlier. This restriction places our eligible and ineligible groups at
similar parts of the lifecycle and with similar past events related to student loan repayment.
Second, the structure of loans means additivity for one outcome is unlikely to mean ad-
ditivity for another. As an example, if our eligible and ineligible groups have pre-treatment
differences in loan balance, then the nonlinearity of repayment plans (because of compound-
ing interest) would imply differential trends in minimum payment due, even in the absence
of treatment. We address this second concern by investigating the robustness of our results to
varying fixed effects. Specifically, instead of month fixed effects, we use fixed effects for each
payment group-month. Payment group is the borrower’s decile in the distribution of minimum
payments as of December, 2019. These fixed effects soak up initial variation in balances. We
interact these with time to capture nonlinearities.10
In a final analysis, we examine whether the loan forgiveness announcement changed bor-
rowers’ outcomes. This analysis serves two purposes. First, it provides direct evidence on the
effects of the loan forgiveness announcement. Second, it allows us to test whether outcomes
are more responsive to a liquidity shock or a balance reduction. The challenge is that the
loan forgiveness announcement occurs after the payment pause such that changes over time
could reflect responses to the new announcement or dynamic treatment effects from the pay-
ment pause. We thus parameterize the effect of the payment pause to be linear and test for
lead to negative weights on treatment effects (Roth et al., 2022). Our analysis is robust to these considerations, as
the treatment occurs at the same point for all treated units and random assignment is not conditional on further
covariates.
10
We also show estimates from a specification with state-month fixed effects, where state is the borrower’s state
of residence as of December, 2019. These fixed effects control for geographic-time variation that may reflect
borrowers of different types having sorted to different labor markets since their studies.
9
deviations in outcomes from this linear trend:
yi t = β El i gi blei Post t +γEli gi blei M onthsPost t +πEli gi blei Post For giveness t +µi +τ t +εi t ,
(3)
where M onthsPost t is the number of months since the start of the payment pause, and Post For giveness t
is an indicator for being after the student loan forgiveness announcement.
3 Data
We use data from the Booth TransUnion Consumer Credit Panel. The data are an anonymized
10% panel sample of all TransUnion credit records from 2000 to 2022. All individuals who
were initially in the sample in 2000 have their data continuously updated, and each year 10%
of new first time individuals in TransUnion data are added.11 We further restrict to accounts
that have been updated after 2018 and have been making positive payments prior to March
2020 and drop duplicates, joint accounts, cosigned loans, and individuals missing birthdates,
as some of these may be parent borrowers. Scaling to match aggregates, there are 48 million
such individuals, which is slightly higher than the 45 million federal student loan borrowers
in the United States.
To compare relatively similar cohorts that had both FFEL and DL options available, we
include only borrowers whose most recent loan was opened in 2010 or earlier. We classify DL
borrowers as those who see more than a 50% decline in scheduled payments after March 2020,
and classify the rest as FFEL borrowers.12 The fact that we use scheduled rather than actual
payments to construct the groups avoids concerns that we may be misclassifying delinquent
FFEL borrowers as DL borrowers. We have 299,637 DL borrowers and 354,680 FFEL borrowers
in our final sample.
Table A.1 presents summary statistics. The table shows balances, payments, and delin-
11
All tables and figures that list TransUnion as a source have statistics calculated (or derived) based on credit
data provided by TransUnion, a global information solutions company, through a relationship with the Kilts Center
for Marketing at the University of Chicago Booth School of Business.
12
70.3% of DL borrowers see their scheduled payments drop to zero. The remainder have scheduled payments
drop by less than the full amount, likely due to having both FFEL and DL program loans. The fact that we drop
cosigned loans makes it unlikely that many of loans classified as FFEL are non-government private loans, as the
vast majority of private student loans in the United States require a parent or other cosigner.
10
quencies for various loan types, as well as credit scores and the share of DL borrowers. Table
A.2 further breaks down summary statistics by each group. The FFEL borrowers tend to have
opened loans slightly earlier, as the program existed for a longer time and was larger prior to
its end in 2010. Generally, the two groups are similar across credit outcomes, although FFEL
borrowers have higher mortgage balances.
4 Main Results
4.1 Results
Table 1 shows estimates of the coefficient β from equation (1), varying fixed effects. The first
column includes time period (month) and individual fixed effects, the second column adds state
by month fixed effects, and the final column adds pre-treatment payment group (in deciles) by
time period fixed effects. The top panel shows student loan outcomes, the second panel shows
mortgage loan outcomes, the third panel shows auto loans, the fourth panel shows revolving
credit, and the final panel shows total loan balances excluding student loans.
We first examine the drop in student loan balances. Figure 1 shows the drop in payments
and evolution of loan balances. Specifically, the figure shows estimates of the coefficients β t
from equation (2), along with a pointwise 95% confidence interval. Consistent with the policy,
we see a $138 drop in payments following the payment moratorium for DL borrowers, relative
to FFEL borrowers who did not see a drop in payments. This leads to an approximate $1,500
in additional student loan balances in the following year, as the payment pause continued.
The top panel of table 1 presents regression estimates from the corresponding difference-in-
difference model (equation 1).
There is a sharp drop in student loan delinquencies following the payment moratorium,
shown in figure 2.13 The drop in student loan delinquencies is in part mechanical, as existing
defaults were cured and individuals with no payments due are unable to become delinquent.
For other loan types, however, which did not see payment pauses (mortgage, auto, and revolv-
ing debt), we see no change in delinquencies. Mortgage delinquencies show a slight increase
13
We measure delinquencies as a flow. That is, our primary outcome is an indicator of whether a borrower is
at least 90-days delinquent on a particular type of loan, for the first time in a given month.
11
of 0.0002, but this is economically tiny. This suggests that the additional liquidity from not
making student loan payments did not have effects on the ability to pay other pre-existing
loans during the two years of payment pauses. Thus, additional liquidity was not important in
preserving financial stability arising from loan defaults.
The drop in delinquency is associated with a sharp rise in credit scores, shown in the bottom
panel of figure 2. By the end of the sample period, credit scores increase by approximately 8
points.
Figure 3 shows auto, mortgage, credit card, and total loan payments and balances following
the student loan payment pause. For all types of credit, payments and balances increase. By
the end of the sample period individuals with a student loan payment pause owe an additional
$1,800 in other debt, and pay an additional $20 monthly. Most of the increase in debt is
driven by mortgage payments, which tend to be longer maturity, and revolving debt.14 The
third through fifth panels of table 1 present regression estimates corresponding to figures 2
and 3, presenting the coefficient β in order to quantify effects over the full sample period.
The estimates show that the payment pause had effects on use of credit (through increased
payments and balances) and creditworthiness (through credit score). To understand whether
the creditworthiness may have led to more use of credit, we split the sample of borrowers
based on an important factor in baseline creditworthiness – whether the borrower had ever
been delinquent on a loan prior to the payment pause. We repeat our difference-in-difference
analysis separately for these two borrower groups and report the results in Table 2.
We start with the student loan outcomes, in the top panel. We see similar decreases across
these two groups in student loan payments from the payment pause. Because the policy ap-
plies to both groups’ loans, this is unsurprising. We see differences, though, when we look
at effects on student loan delinquency rates. The payment pause does not affect the proba-
bility of becoming newly delinquent, as not-yet-delinquent borrowers see no treatment effect.
But borrowers who have been delinquent at least once enjoy a 5 percentage point decrease in
the probability of being delinquent. These heterogeneous effects suggest that the effect of the
14
Note that taking an additional $1,000 in mortgage debt at a 7% interest rates would lead to an approximately
$7 increase in monthly payments under a standard 30-year fixed mortgage. Similarly, taking an additional $100
in auto loan debt at a 7% interest rate would lead to a payment of $2 under a 5-year payment plan, which is the
typical maturity of auto loan contracts.
12
pause was curing existing defaults rather than avoiding new ones.
We next turn to the effects on credit scores and other borrowing on non-student loans. We
estimate that the effect on credit scores is concentrated among borrowers who have previously
been delinquent on a loan. The payment pause, by curing existing defaults, led to a partic-
ularly large change in (perceived) creditworthiness: an increase in credit score of 28 points.
Borrowers who had not previously been delinquent see only a modest increase in credit score
of just 2 points. Interestingly, we see that the effects on balances and payments for non-student
debt are concentrated among borrowers who have not previously been delinquent. For exam-
ple, we estimate increased mortgage balances of $917 for never delinquent borrowers and a
decrease of $120 for ever delinquent borrowers. In terms of non-student total debt, the effect
on balances is large and positive for never delinquent borrowers while small and negative for
ever delinquent borrowers. The effect on payments is 2.2 times as large for never delinquent
borrowers.
4.2 Student Loan Forgiveness
Finally, we study effects of the loan forgiveness announcement. Figures 2 and 3 show little
relative change in outcomes for DL borrowers following the August, 2022 loan forgiveness
announcement. Payments, balances, and delinquencies evolve smoothly around the date of
announcement. This negligible change in borrowing or payments occurs despite the promise
of a significant boost in wealth of $10-20,000, which is 10-20% of median household wealth
in the United States. The fact that we do not see any changes in borrowing for DL borrowers,
who were affected by the forgiveness announcement, relative to FFEL borrowers, who were
unaffected by the announcement, suggest very small wealth effects arising from the debt for-
giveness announcement. The lack of response to the loan forgiveness announcement is also
reflected in the regression results. In Table A.4, we present the estimates of Equation 3. We
see the estimated deviations from trend are small and statistically insignificant. The estimates
are stable across fixed effect specifications.
The lack of changes in outcomes following the announcement of significant debt forgiveness
is again consistent with our finding that liquidity plays a larger role in debt and consumption
outcomes. The announcement led to no actual change in cash on hand; the announcement,
13
however, promised future payment cancellation. An alternative hypothesis is that student loan
borrowers did not actually believe that any forgiveness would occur, and hence treated the
announcement with skepticism. Our data do not allow us to rule out the possibility that the
lack of reaction is simply due to borrowers anticipating that courts would strike down potential
forgiveness.15
4.3 Discussion
Our results speak to several policy comparisons. First, we can compare the effects of a payment
pause relative to no policy change. We find that the payment pause had a large effect on im-
mediate consumption. Policymakers focused on boosting short-term consumption, especially
for stimulus effects, may then find debt moratoria to be effective policy tools at relatively low
long-term fiscal cost. At the same time, we find that the student debt payment moratorium
led to higher levels of overall leverage, not only through borrowers not paying down student
debt balances, but also through the accumulation of other types of household debt. By the
end of the sample period, student debt borrowers have about 5% more household debt, driven
roughly half by student and non-student debt. Perhaps paradoxically, temporary student debt
relief leads to higher overall household debt levels and larger future debt burdens. On the one
hand, this increased debt could be financing productive investments or durable purchases that
smooth consumption. On the other hand, this could be a potential concern to policymakers,
as both theoretical and empirical work has shown that higher levels of leverage can affect ag-
gregate consumption and the transmission of costs through household balance sheet or debt
overhang channels.
Second, we can compare the effects of a payment pause to an announced direct transfer. As
the former targets liquidity while the latter targets debt reduction, the optimal policy depends
crucially on the nature of constraints households face.16 The larger effects from the payment
15
It is also possible FFEL borrowers misunderstood the announcement, and thought that they would be eligible.
This appears to be less likely, however, as we see no obvious changes in raw means following the announcement,
which is shown in appendix figure A.1. For various outcomes, we do not see changes in either group following
the announcement.
16
Because the loan forgiveness has only been announced, it has not yet affected liquidity. While an implemented
forgiveness policy would also provide liquidity – and thus may have similar effects to a payment pause – it would
come at a higher fiscal cost.
14
pause indicate that it may be more effective in providing stimulus and consumption smoothing.
Our combination of results also highlights the potential mechanisms at play. That outcomes
only respond to the payment pause and not the loan forgiveness announcement suggests that
borrowers are more constrained by limited liquidity than high balances.17 Further, the finding
that the increased consumption (through borrowing) is concentrated among never delinquent
borrowers, whose creditworthiness hardly changes, highlights an important interaction be-
tween liquidity and credit.18 The policy serves as a liquidity shock that induces increased use
of credit. The effects do not appear to be driven by the subset of borrowers who saw an im-
provement in their credit scores. This raises the question of why such credit was not previously
being used. We speculate that using credit requires a certain level of liquidity for making down
payments or making the first due monthly payments. This interaction between liquidity and
credit is important for policy design and highlights that policymakers should consider them
jointly.
5 Concluding Remarks
We study the role of debt moratoria, a commonly used form of relief to distressed borrowers,
through the lens of student loan forbearance. We find that, relative to borrowers who had to
continue paying their loans, borrowers who had a pause in their payments sharply increased
mortgage, auto, and credit card borrowing. We find little effect on loan delinquencies. The
effects appear driven by increased credit demand and ability to service loans, as opposed to a
credit supply effect. The results indicate that debt payment pauses can increase consumption
in the short term, but that overall debt increases, as borrowers use increased liquidity to service
new debt.
17
A small set of borrowers subject to the payment pause may see a change in expected future payments if they
are enrolled in income-based repayment plans that offer forgiveness after a certain number of periods. If changes
in expected total future payments were the mechanism driving our results, we likely would see responses to the
loan forgiveness announcement.
18
Individuals may have increased their consumption without borrowing, as well. These actions are not captured
in our credit report data, so we simply note that some of the consumption increase came through borrowing.
15
References
Agarwal, Sumit and Wenlan Qian, “Consumption and debt response to unanticipated income
shocks: Evidence from a natural experiment in Singapore,” American Economic Review, 2014,
104 (12), 4205–30.
, Chunlin Liu, and Nicholas S Souleles, “The reaction of consumer spending and debt to
tax rebates—evidence from consumer credit data,” Journal of Political Economy, 2007, 115
(6), 986–1019.
, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, Tomasz Piskorski, and
Amit Seru, “Policy intervention in debt renegotiation: Evidence from the home affordable
modification program,” Journal of Political Economy, 2017, 125 (3), 654–712.
Amromin, Gene and Janice Eberly, “Education financing and student lending,” Annual Review
of Financial Economics, 2016, 8, 289–315.
Aydin, Deniz, “Forbearance, interest rates, and present-value effects in a randomized debt
relief experiment,” Available at SSRN 3982587, 2021.
, “Consumption response to credit expansions: Evidence from experimental assignment of
45,307 credit lines,” American Economic Review, 2022, 112 (1), 1–40.
Baker, Scott R, “Debt and the response to household income shocks: Validation and application
of linked financial account data,” Journal of Political Economy, 2018, 126 (4), 1504–1557.
Black, Sandra E, Jeffrey T Denning, Lisa J Dettling, Sarena Goodman, and Lesley J Turner,
“Taking it to the limit: Effects of increased student loan availability on attainment, earnings,
and financial well-being,” Technical Report, National Bureau of Economic Research 2020.
Boutros, Michael, Nuno Clara, and Francisco Gomes, “Borrow now, pay even later: A quan-
titative analysis of student debt payment plans,” Working Paper, 2022.
Briones, Diego, Eileen Powell, and Sarah Turner, “Student Loan Payment Pause Benefits
High-Income Households the Most,” Education Next, Winter 2023, 23 (1).
Catherine, Sylvain and Constantine Yannelis, “The distributional effects of student loan for-
givenesss,” Journal of Financial Economics, 2022.
Chakrabarti, Rajashri, Nicole Gorton, and Michael F Lovenheim, “State Investment in
Higher Education: Effects on Human Capital Formation, Student Debt, and Long-Term Fi-
nancial Outcomes of Students,” Technical Report, National Bureau of Economic Research
2020.
Chava, Sudheer, Heather Tookes, and Yafei Zhang, “Leaving Them Hanging: Student Loan
Forbearance, Distressed Borrowers, and Their Lenders,” 2023.
Cherry, Susan F, Erica Jiang Jiang, Gregor Matvos, Tomasz Piskorski, and Amit Seru, “Gov-
16
ernment and private household debt relief during Covid-19,” Technical Report Fall, National
Bureau of Economic Research 2021.
Chodorow-Reich, Gabriel, Plamen T Nenov, and Alp Simsek, “Stock market wealth and the
real economy: A local labor market approach,” American Economic Review, 2021, 111 (5),
1613–57.
Donaldson, Jason Roderick, Giorgia Piacentino, and Anjan Thakor, “Household debt over-
hang and unemployment,” The Journal of Finance, 2019, 74 (3), 1473–1502.
Erel, Isil and Jack Liebersohn, “Can FinTech reduce disparities in access to finance? Evidence
from the Paycheck Protection Program,” Journal of Financial Economics, 2022, 146 (1), 90–
118.
Fiorin, Stefano, Joseph Hall, and Martin Kanz, “How do Borrowers Respond to a Debt Mora-
torium? Experimental Evidence from Consumer Loans in India,” 2023.
Ganong, Peter and Pascal Noel, “Liquidity versus wealth in household debt obligations: Ev-
idence from housing policy in the great recession,” American Economic Review, 2020, 110
(10), 3100–3138.
Giroud, Xavier and Holger M Mueller, “Firm leverage, consumer demand, and employment
losses during the great recession,” The Quarterly Journal of Economics, 2017, 132 (1), 271–
316.
Goodman, Sarena, Adam Isen, and Constantine Yannelis, “A day late and a dollar short: Liq-
uidity and household formation among student borrowers,” Journal of Financial Economics,
2021, 142 (3), 1301–1323.
Granja, João, Christian Leuz, and Raghuram G Rajan, “Going the extra mile: Distant lending
and credit cycles,” The Journal of Finance, 2022, 77 (2), 1259–1324.
, Christos Makridis, Constantine Yannelis, and Eric Zwick, “Did the paycheck protection
program hit the target?,” Journal of Financial Economics, 2022, 145 (3), 725–761.
Herbst, Daniel and Nathaniel Hendren, “Opportunity unraveled: Private information and
the missing markets for financing human capital,” Technical Report, National Bureau of Eco-
nomic Research 2021.
Indarte, Sasha, “Moral hazard versus liquidity in household bankruptcy,” Journal of Finance,
forthcoming.
Karamcheva, Nadia, Jeffrey Perry, and Constantine Yannelis, “Income-driven repayment
plans for student koans,” April 2020, (56337).
Katz, Justin, “Savings and Consumption Responses to Student Loan Forbearance,” Available
at SSRN 4344262, 2023.
Lochner, Lance and Alexander Monge-Naranjo, “Student loans and repayment: Theory, ev-
17
idence, and policy,” in “Handbook of the Economics of Education,” Vol. 5, Elsevier, 2016,
pp. 397–478.
Maggio, Marco Di, Amir Kermani, and Christopher J Palmer, “How quantitative easing
works: Evidence on the refinancing channel,” The Review of Economic Studies, 2020, 87 (3),
1498–1528.
, , Benjamin J Keys, Tomasz Piskorski, Rodney Ramcharan, Amit Seru, and Vincent
Yao, “Interest rate pass-through: Mortgage rates, household consumption, and voluntary
deleveraging,” American Economic Review, 2017, 107 (11), 3550–88.
, Ankit Kalda, and Vincent Yao, “Second chance: Life without student debt,” Technical
Report, National Bureau of Economic Research 2019.
Mian, Atif and Amir Sufi, “The consequences of mortgage credit expansion: Evidence from the
US mortgage default crisis,” The Quarterly Journal of Economics, 2009, 124 (4), 1449–1496.
and , “House Prices, Home Equity-Based Borrowing and the US Household Leverage
Crisis,” American Economic Review, 2011, 101 (5), 2132–56.
, Kamalesh Rao, and Amir Sufi, “Household balance sheets, consumption and the economic
slump,” The Quarterly Journal of Economics, 2013, 128 (4), 1687–1726.
Mueller, Holger and Constantine Yannelis, “The rise in student loan defaults,” Journal of
Financial Economics, 2019, 131 (1), 1–19.
and , “Increasing enrollment in income-driven student loan repayment plans: Evidence
from the Navient field experiment,” The Journal of Finance, 2022, 77 (1), 367–402.
Rose, Jonathan D, “The incredible HOLC? Mortgage relief during the Great Depression,” Jour-
nal of Money, Credit and Banking, 2011, 43 (6), 1073–1107.
Roth, Jonathan, Pedro HC Sant’Anna, Alyssa Bilinski, and John Poe, “What’s Trending in
Difference-in-Differences? A Synthesis of the Recent Econometrics Literature,” arXiv preprint
arXiv:2201.01194, 2022.
Yannelis, Constantine and Greg Tracey, “Student loans and borrower outcomes,” Annual
Review of Financial Economics, 2022, 14.
18
Figure 1: Student Loan Balances and Payments
Payments Balances
0 2500
25
2000
50
75 1500
100 1000
125
500
150
175 0
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Notes: This figure shows the coefficients β t from the OLS regression yi t = t β t Eli gi blei + µi + τ t + εi t , where yi t
P
is student loan balances or payments, along with a 95% confidence interval. We include fixed effects for individual
and initial loan payment decile. The outcome is labelled above each panel. The solid line shows the month before the
payment moratorium announcement, while the dashed line shows the month before the forgiveness announcement.
Standard errors are doubled clustered at the repayment cohort level. Source: TransUnion
19
Figure 2: Loan Delinquency
Student Loan Delinquencies Auto Loan Delinquencies
0.0008
0.002
0.0006
0.000
0.002 0.0004
0.004 0.0002
0.006 0.0000
0.008 0.0002
0.010 0.0004
0.012 0.0006
0.014
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Mortgage Delinquencies Credit Card Delinquencies
0.0006
0.0010
0.0004
0.0005
0.0002
0.0000
0.0000
0.0005
0.0002 0.0010
0.0004 0.0015
2019-03 2019-09 -03 -09 -03 -09 -03
2020 2020 2021 2021 2022 2022
-09
2019
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2 2 2 2 2 2 2
Date Date
Credit Score
10
8
6
4
2
0
-03 -09 -03 -09 -03 -09 -03 -09
2019 2019 2020 2020 2021 2021 2022 2022
Date
Notes: This figure shows the coefficients β t from the OLS regression yi t = t β t Eli gi blei + µi + τ t + εi t , where yi t is
P
loan delinquency or credit scores, with a 95% confidence interval. We include fixed effects for individual and initial
loan payment decile. The outcome is labelled above each panel. The solid line shows the month before the payment
moratorium announcement, while the dashed line shows the month before the forgiveness announcement. Standard
errors are doubled clustered at the repayment cohort level. Source: TransUnion
20
Figure 3: Student Loan Balances and Payments
Payments Balances
Auto Loans
250
6
200
4 150
100
2
50
0
0
50
2 100
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Mortgage Loans
12.5
1500
10.0
7.5 1000
5.0
500
2.5
0.0 0
2.5
500
5.0
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Credit Card
8 400
6 300
4 200
2 100
0 0
2 100
2019-03 2019-09 -03 -09 -03 -09 -03
2020 2020 2021 2021 2022 2022
-09
2019
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2 2 2 2 2 2 2
Date Date
All Non-Student Loans
2500
25
2000
20
1500
15
10 1000
5 500
0 0
5 500
10
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Notes: This figure shows the coefficients β t from the OLS regression yi t = t β t Eli gi blei + µi + τ t + εi t , where yi t is
P
loan balances or payments, along with a 95% confidence interval. We include fixed effects for individual and initial
loan payment decile. The outcome is labelled above each panel. The solid line shows the month before the payment
moratorium announcement, while the dashed line shows the month before the forgiveness announcement. Standard
errors are doubled clustered at the repayment cohort level. Source: TransUnion
Table 1: Main Results by Credit Line
(1) (2) (3)
Panel A: Student Loans
Balance 1501.564∗∗∗ 1509.361∗∗∗ 1197.205∗∗∗
(101.104) (99.947) (121.535)
Payments -138.390∗∗∗ -138.402∗∗∗ -149.067∗∗∗
(5.748) (5.811) (3.567)
Delinquent -0.008∗∗∗ -0.008∗∗∗ -0.007∗∗∗
(0.001) (0.001) (0.001)
Credit Score 5.793*** 5.730∗∗∗ 5.193∗∗∗
(0.4827) (0.466) (0.373)
Panel B: Mortgages
Balance 773.149∗∗ 712.731∗∗ 837.799∗∗
(227.770) (251.106) (237.115)
Payments 7.255∗∗ 7.046∗ 7.658∗∗
(2.628) (2.735) (2.719)
Delinquent 0.000 0.000 0.000∗∗
(0.000) (0.000) (0.000)
Panel C: Auto Loans
Balance 95.275∗ 101.409∗∗ 85.385∗
(36.215) (35.009) (39.198)
Payments 2.282∗ 2.429∗ 2.214∗
(0.994) (0.968) (1.013)
Delinquent 0.000 0.000 0.000
(0.000) (0.000) (0.000)
Panel D: Credit Cards
Balance 327.736∗∗∗ 331.501∗∗∗ 290.710∗∗∗
(43.122) (41.651) (45.847)
Payments 6.617∗∗∗ 6.734∗∗∗ 5.886∗∗∗
(0.833) (0.805) (0.928)
Delinquent 0.000 0.000 0.000∗
(0.000) (0.000) (0.000)
Panel E: Non-Student Debt
Balance 1192.812∗∗∗ 1142.312∗∗∗ 1214.139∗∗∗
(269.279) (292.501) (293.234)
Payments 15.671∗∗∗ 15.730∗∗∗ 15.401∗∗
(4.110) (4.157) (4.383)
Delinquent 0.000 0.000 0.001∗
(0.000) (0.000) (0.000)
Individual FE Ø Ø Ø
Time FE Ø Ø Ø
State × Time FE Ø Ø
Payment Group × Time FE Ø
Observations 28,789,948 28,789,948 28,789,948
Notes: This table shows the coefficients β t from the OLS regression yi t = β Eli gi blei Post t +µi +τ t +εi t . The outcome
is labelled above each panel, and the inclusion of fixed effects is noted at the bottom of the table. Standard errors are
in parentheses, and are clustered at the repayment cohort level. Source: TransUnion. ∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01.
22
Table 2: Results Split by Delinquency Status
Never Delinquent Ever Delinquent
(1) (2) (3) (1) (2) (3)
Panel A: Student Loans
Balance 1633.790∗∗∗ 1640.227∗∗∗ 1303.036∗∗∗ 904.610∗∗∗ 913.541∗∗∗ 801.676∗∗∗
(88.045) (87.335) (105.607) (138.349) (138.737) (162.136)
Payments -140.795∗∗∗ -140.642∗∗∗ -152.624∗∗∗ -126.021∗∗∗ -125.708∗∗∗ -133.197∗∗∗
(5.725) (5.797) (3.634) (5.019) (5.026) (3.795)
Delinquent 0.000 0.000 0.000 -0.048∗∗∗ -0.048∗∗∗ -0.048∗∗∗
(.) (.) (.) (0.000) (0.000) (0.000)
Credit Score 1.610∗∗∗ 1.554∗∗∗ 0.981∗∗ 27.840∗∗∗ 27.677∗∗∗ 27.655∗∗∗
(0.380) (0.369) (0.298) (0.377) (0.377) (0.373)
Panel B: Mortgages
Balance 917.130∗∗∗ 860.045∗∗∗ 958.247∗∗∗ -120.076 -142.825 -139.124
(209.264) (234.348) (227.464) (237.693) (238.173) (243.960)
Payments 7.749∗∗ 7.572∗∗ 8.143∗∗ 3.817 3.761 3.473
(2.571) (2.703) (2.702) (1.955) (1.936) (2.058)
Delinquent 0.000 0.000 0.000∗∗ 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Panel C: Auto Loans
Balance 103.056∗∗ 110.097∗∗ 86.740∗ 49.174 54.617 46.673
(36.405) (35.073) (39.993) (57.021) (56.473) (58.620)
Payments 2.581∗∗ 2.735∗∗ 2.389∗ 0.518 0.685 0.536
(0.901) (0.889) (0.956) (1.657) (1.602) (1.650)
Delinquent 0.000∗∗ 0.000∗ 0.000∗∗∗ -0.001∗ -0.001∗ -0.001∗
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Panel D: Credit Cards
Balance 381.447∗∗∗ 385.715∗∗∗ 339.698∗∗∗ 58.175∗ 58.878∗ 45.725
(47.048) (45.395) (49.319) (23.575) (22.493) (23.690)
Payments 7.127∗∗∗ 7.276∗∗∗ 6.355∗∗∗ 3.978∗∗∗ 3.963∗∗∗ 3.473∗∗∗
(0.926) (0.895) (1.024) (0.724) (0.681) (0.711)
Delinquent 0.000 0.000 0.000∗∗∗ -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Panel E: Non-Student Debt
Balance 1397.255∗∗∗ 1352.039∗∗∗ 1383.846∗∗∗ -13.474 -31.040 -47.776
(242.463) (268.406) (277.569) (273.878) (273.545) (281.277)
Payments 16.987∗∗∗ 17.124∗∗∗ 16.568∗∗∗ 7.689∗ 7.780∗ 6.878
(3.917) (3.986) (4.278) (3.420) (3.291) (3.524)
Delinquent 0.000 0.000 0.001∗∗∗ -0.001 -0.001 -0.001
(0.000) (0.000) (0.000) (0.001) (0.001) (0.001)
Individual FE Ø Ø Ø Ø Ø Ø
Time FE Ø Ø Ø Ø Ø Ø
State × Time FE Ø Ø Ø Ø
Payment Group × Time FE Ø Ø
Observations 24,114,244 24,114,244 24,114,244 4,675,704 4,675,704 4,675,704
Notes: This table shows the coefficients β t from the OLS regression yi t = β Eli gi blei Post t +µi +τ t +εi t for never delin-
quent and ever delinquent groups. An individual is classified as ever delinquent if she has one or more delinquencies
reported on her student loans between 2019 and 2021 and never delinquent if she has zero reported delinquencies.
The outcome is labelled above each panel, and the inclusion of fixed effects is noted at the bottom of the table.Standard
errors are in parentheses, and are clustered at the repayment cohort level. Source: TransUnion. ∗ p < .1, ∗∗ p < .05,
∗∗∗
p < .01.
23
Appendix
A Additional Tables and Figures
Figure A.1: Student Loan Balances and Payment
Payments Balances
28000
250
26000
200 24000
150 22000
20000
100
18000
50 16000
14000
0
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Notes: This figure shows mean student loan payments and balances for student loans, broken down by DL and FFEL
borrowers. The solid line shows the month before the payment moratorium announcement, while the dashed line
shows the month before the forgiveness announcement. Source: TransUnion
24
Figure A.2: Loan New Delinquencies
Student Loan Delinquencies Auto Loan Delinquencies
0.0050
0.012
0.0045
0.010
0.0040
0.008
0.0035
0.006
0.0030
0.004
0.0025
0.002
0.0020
0.000
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Mortgage Delinquencies Credit Card Delinquencies
0.00225
0.011
0.00200
0.010
0.00175
0.009
0.00150 0.008
0.00125 0.007
0.00100 0.006
0.00075 0.005
0.00050 0.004
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Credit Score
710
700
690
680
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2
DL FFEL
Notes: This figure shows mean delinquencies and credit scores, broken down by DL and FFEL borrowers. The solid
line shows the month before the payment moratorium announcement, while the dashed line shows the month before
the forgiveness announcement. Source: TransUnion
25
Figure A.3: Student Loan Balances and Payments
Payments Balances
Auto Loans
8500
200 8000
7500
150
7000
100 6500
6000
50 5500
5000
0
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Mortgage Loans
400 55000
50000
300 45000
40000
200 35000
30000
100 25000
20000
0
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Credit Card
9000
200
8000
150
7000
100
6000
50
5000
0
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Non-Student Debt
800 70000
60000
600
50000
400
40000
200
30000
0
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
DL FFEL DL FFEL
Notes: This figure shows mean payments and balances for non-student loans, broken down by DL and FFEL borrowers.
The solid line shows the month before the payment moratorium announcement, while the dashed line shows the month
before the forgiveness announcement. Source: TransUnion
26
Table A.1: Full Sample Summary Statistics
Mean St. Dev.
Panel A: Student Loans
Current Balance Amount 21332.29 30707.40
Payments 139.05 189.39
Delinquency 0.01 0.08
Credit Score 702.19 91.82
Direct Loan 0.46 0.50
Panel B: Mortgages
Current Balance Amount 37989.60 98938.03
Payments 281.10 706.27
Delinquency 0.00 0.04
Panel C: Auto Loans
Current Balance Amount 6692.26 12073.25
Payments 183.82 292.46
Delinquency 0.00 0.06
Panel D: Credit Cards
Current Balance Amount 6571.84 10019.11
Payments 157.18 231.29
Delinquency 0.01 0.09
Panel E: Non-Student Debt
Current Balance Amount 51650.16 104167.74
Payments 631.08 901.52
Delinquency 0.01 0.11
Observations 28,789,948
Notes: This table shows summary statistics for the main analysis variables. The data
are averaged on all observations from March, 2019 to December, 2021. Source:
TransUnion
27
Table A.2: Direct Loan vs FFEL Loan Summary Statistics
Direct Loans FFEL
Mean St. Dev. Mean St. Dev.
Panel A: Student Loans
Current Balance Amount 23462.71 31285.62 24210.63 31790.39
Payments 206.84 183.79 226.29 186.11
Delinquency 0.01 0.04 0.01 0.03
Credit Score 678.06 94.10 700.07 90.29
Open Year 2007 3.57 2005 2.90
Panel B: Mortgages
Current Balance Amount 27547.93 76776.47 39430.98 97164.26
Payments 215.95 576.44 311.09 728.99
Delinquency 0.00 0.02 0.00 0.02
Panel C: Auto Loans
Current Balance Amount 6166.54 10331.30 6589.07 11243.08
Payments 170.02 254.21 187.30 285.01
Delinquency 0.01 0.03 0.00 0.03
Panel D: Credit Cards
Current Balance Amount 6075.87 9057.24 7496.17 10440.39
Payments 148.93 211.84 174.55 238.43
Delinquency 0.01 0.04 0.01 0.04
Panel E: Non-Student Debt
Current Balance Amount 40031.24 81612.10 53923.32 102623.31
Payments 540.73 756.71 683.16 929.82
Delinquency 0.02 0.05 0.02 0.05
Number of Subjects 299,637 354,680
Notes: This table shows summary statistics for the Direct Loan and FFEL groups. The data are averaged on observations
prior to the policy (March, 2019 to March, 2020). Open years are rounded to the nearest whole year. Source:
TransUnion
28
Table A.3: Delinquency Status Summary Statistics
Ever Delinquent Never Delinquent
Mean St. Dev. Mean St. Dev.
Panel A: Student Loans
Current Balance Amount 26514.39 34048.98 20327.49 28984.16
Payments 127.88 146.14 141.22 149.53
Delinquency 0.04 0.02 0.00 0.00
Credit Score 591.49 66.08 723.65 70.24
Panel B: Mortgages
Current Balance Amount 19509.26 60374.71 41572.90 93035.50
Payments 147.26 438.20 307.06 663.28
Delinquency 0.00 0.01 0.00 0.01
Panel C: Auto Loans
Current Balance Amount 4141.66 6824.66 7043.05 9332.94
Payments 100.98 150.71 168.07 211.85
Delinquency 0.01 0.03 0.00 0.01
Panel D: Credit Cards
Current Balance Amount 4141.66 6824.66 7043.05 9332.94
Payments 100.98 150.71 168.07 211.85
Delinquency 0.02 0.03 0.01 0.02
Panel E: Non-Student Debt
Current Balance Amount 29775.12 64526.76 55891.69 98305.82
Payments 409.26 593.98 674.09 850.78
Delinquency 0.03 0.04 0.01 0.02
Number of Subjects 106,266 548,051
Notes: This table shows summary statistics for the delinquency groups. The data is averaged over all observations
(March of 2019 to October of 2022). Source: TransUnion
29
B Sample Construction
B.1 Main Sample
The main data set is constructed from the Booth TransUnion Consumer Credit Panel, which
is a 10% panel sample of the full US population with credit reports. We begin by considering
all individuals with open student loans in the year 2018. There are approximately 5.2 million
such individuals. We first remove individuals whose information has not been updated beyond
2018, duplicate accounts, and borrowers with loans originating outside the continental United
States. Of the remaining borrowers we filter out subjects with missing birthdays and borrowers
possessing student loans that have co-signers attached as both conditions indicate possible
parent borrowers, or private student loans which typically require a co-signer.
Because FFEL was discontinued in 2010 and we wish to compare relatively similar cohorts,
we do not include borrowers with student loans originating after 2010. We additionally restrict
to accounts that have made at least 3 positive payments between June, 2018 and March, 2020.
Lastly, we remove borrowers that have more than one month of $0 payments concurrent with a
positive loan balance. This eliminates any borrowers who may have been in forbearance prior
to the policy announcement. This results in a set of 654,317 borrowers.
To classify DL and FFEL we consider changes in scheduled student loan payments before
and after the forbearance policy for loans aggregated at the borrower-month level. We define
the pre-period to be June, 2018 to March, 2020 and the post-period to be March, 2020 to
January, 2022. Individuals with a sum total of $0 scheduled student loan payments in the
post-period are classified as direct loan borrowers. Additionally, individuals who experienced
greater than a 50% decrease in average monthly pre-period payments to average monthly post-
period payments are classified as direct loan borrowers. Those with less than a 50% decline in
pre-period to post-period payments are classified as FFEL borrowers.
For each individual in the sample of DL and FFEL borrowers we aggregate the balances,
payments, and delinquencies of their mortgages, credit cards, and auto loans at the month
level. We assume a balance, payment, and delinquency of zero for those with no reported
debt in a given credit line. Finally, we calculate total debt as the sum of mortgage, credit card,
and auto loan balances, payments, and delinquencies. We winsorize all balance and payment
outcomes at the 1% level.
B.2 Alternative Sample Construction
A potential concern with our construction is that it may induce measurement error, if payments
drop to zero due to discretionary forbearance or other mechanisms to defer payments. To
30
Table A.4: Results on Loan Forgiveness Announcement
(1) (2) (3)
Panel A: Mortgages
Balance 67.296 82.167 49.815
(115.353) (118.103) (123.685)
Payments 0.387 0.413 0.316
(0.923) (0.962) (0.994)
Delinquent -0.000 -0.000 -0.000
(0.000) (0.000) (0.000)
Panel B: Auto Loans
Balance 27.525∗ 24.073 33.066∗
(12.144) (12.369) (13.346)
Payments 0.143 0.118 0.290
(0.281) (0.296) (0.299)
Delinquent 0.000 0.000 0.000
(0.000) (0.000) (0.000)
Panel C: Credit Cards
Balance -33.825∗ -36.026∗ -24.542
(15.309) (14.184) (16.342)
Payments -0.802 -0.883∗ -0.614
(0.417) (0.380) (0.432)
Delinquent 0.000 0.000 0.000
(0.000) (0.000) (0.000)
Panel D: Non-Student Debt
Balance 45.658 53.99943.774
(126.051) (130.493)
(135.634)
Payments -0.964 -1.064
-0.633
(1.087) (1.119)
(1.185)
Delinquent 0.000 0.000 0.000
(0.000) (0.000)
(0.000)
Individual FE Ø Ø Ø
Time FE Ø Ø Ø
State × Time FE Ø Ø
Payment Group × Time FE Ø
Observations 28,789,948 28,789,948 28,789,948
Notes: This table shows the coefficients β1 from the OLS regression yi t = β Eli gi blei Post t +γEli gi blei M onthsPost t +
πEl i gi blei Post For g i veness t + µi + τ t + εi t , where Post t is an indicator for being after the payment pause,
M onthsPost t is the number of months since the start of the payment pause, and Post For giveness t is an indica-
tor for being after the student loan forgiveness announcement. The outcome, yi t , is labelled above each panel, and
the inclusion of fixed effects is noted at the bottom of the table. Standard errors are in parentheses, and are clustered
at the repayment cohort level. Source: TransUnion. ∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01.
31
confirm our results and classification method we employ an alternative method for identifying
DL and FFEL borrowers using lender IDs. The Booth TransUnion Consumer Credit Panel masks
the specific identifies of lenders but does provide a unique ID for each lender. Additionally, the
Booth TransUnion Consumer Credit Panel provides industry codes for each lender. We can
determine whether loans originated prior to 2010 were under the DL program, as these all
used one particular large servicer. This information enables us to estimate lenders as either
FFEL, DL, or private loan lenders.
To identify FFEL lenders we begin by considering lenders with industry codes in banking,
personal finance, education, and government. Because FFEL loans are distributed by private
banks but backed by the federal government it is plausible for credit bureaus to report FFEL
lenders as either private or government entities. Among this subset of lenders we consider
only lenders who stopped lending after 2010 since this was the year the FFEL program was
discontinued and no FFEL loans should exist post 2010. All lenders with the previously men-
tioned industry codes and no record of lending post 2010 are considered FFEL lenders. Loans
from lenders who have industry codes in banking and personal finance but are not classified as
FFEL loans are considered private loans as long as more than 70% of the lender’s student loans
have a co-signer attached. Any lender that is not classified as a FFEL or private loan lender is
considered a DL lender.
Comparing this classification method to our main classification method we find that ap-
proximately 72% of loans are identified as the same type with both methods. Of the loans
whose main classification does not match the alternative classification, 90% are classified as
FFEL in the main classification and DL in the alternative classification. This is likely because
our alternative classification method assumes that all FFEL lenders stopped lending in any form
after 2010. In reality some lenders may have shifted over to private loans or, in the case of
government lenders, direct loans. Thus the alternative method runs the risk of over classifying
direct loans and under classifying FFEL loans. That said, the results using the alternative clas-
sification method are found in Figures B.1 through B.3, and all align closely with the results
using the main classification method and thus strengthen the robustness of our results.
32
Figure B.1: Student Loan Balances and Payments: Alternative Classification
Payments Balances
20 1500
0 1250
20 1000
40 750
60 500
80 250
100 0
120 250
140
500
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
Date Date
Notes: This figure shows the coefficients β t from the OLS regression yi t = t β t Eli gi blei + µi + τ t + εi t , where yi t
P
is student loan balances or payments, along with a 95% confidence interval. We include fixed effects for individual
and initial loan payment decile. The outcome is labelled above each panel. The solid line shows the month before the
payment moratorium announcement, while the dashed line shows the month before the forgiveness announcement.
Standard errors are doubled clustered at the repayment cohort level. Source: TransUnion
33
Figure B.2: Loan Delinquency: Alternative Classification
Student Loan Delinquencies Auto Loan Delinquencies
0.002
0.0008
0.000 0.0006
0.002 0.0004
0.0002
0.004
0.0000
0.006
0.0002
0.008 0.0004
0.0006
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
Date Date
Mortgage Delinquencies Credit Card Delinquencies
0.0020
0.0006
0.0015
0.0004
0.0010
0.0002
0.0005
0.0000
0.0000
0.0002
0.0005
0.0004
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Credit Score
7
6
5
4
3
2
1
0
1
-03 -09 -03 -09 -03 -09 -03 -09
2019 2019 2020 2020 2021 2021 2022 2022
Date
Notes: This figure shows the coefficients β t from the OLS regression yi t = t β t Eli gi blei + µi + τ t + εi t , where yi t is
P
loan delinquency or credit scores, with a 95% confidence interval. We include fixed effects for individual and initial
loan payment decile. The outcome is labelled above each panel. The solid line shows the month before the payment
moratorium announcement, while the dashed line shows the month before the forgiveness announcement. Standard
errors are doubled clustered at the repayment cohort level. Source: TransUnion
34
Figure B.3: Student Loan Balances and Payments: Alternative Classification
Payments Balances
Auto Loans
8
300
6 250
200
4 150
100
2 50
0
0
50
2 100
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Mortgage Loans
2000
15
1500
10
1000
5 500
0 0
500
5
-03 019-09 020-03 020-09 021-03 021-09 022-03 022-09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2 2 2 2 2 2 2 2019 2 2 2 2 2 2 2
Date Date
Credit Card
8 400
6 300
4 200
2 100
0 0
2
100
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
Date Date
All Non-Student Loans
30 2500
25
2000
20
1500
15
1000
10
5 500
0 0
5 500
10 1000
-03 -09 -03 -09 -03 -09 -03 -09 -03 019-09 020-03 020-09 021-03 021-09 022-03 022-09
2019 2019 2020 2020 2021 2021 2022 2022 2019 2 2 2 2 2 2 2
Date Date
Notes: This figure shows the coefficients β t from the OLS regression yi t = t β t Eli gi blei + µi + τ t + εi t , where yi t is
P
loan balances or payments, along with a 95% confidence interval. We include fixed effects for individual and initial
loan payment decile. The outcome is labelled above each panel. The solid line shows the month before the payment
moratorium announcement, while the dashed line shows the month before the forgiveness announcement. Standard
errors are doubled clustered at the repayment cohort level. Source: TransUnion
File and source
- File
- nber-w31247.pdf
- Size
- 508,719 bytes
- SHA-256
- e17807816c1742868eb866c65fc7aa6403875cfe5e4b67b339a7cb26b4e98454
- Our copy
- nber-w31247.pdf
- Original
- www.nber.org