Kim, Lee, Scharlemann, Vickery, Intermediation Frictions in Debt Relief: Evidence from CARES Act Forbearance (NY Fed Staff Report 1035; J. Financial Economics 158 (2024), 2022-10)
Summary
Federal Reserve Bank of New York Staff Reports, no. 1035, dated October 2022, titled Intermediation Frictions in Debt Relief: Evidence from CARES Act Forbearance, by You Suk Kim, Donghoon Lee, Tess Scharlemann, and James Vickery. The abstract states that the paper studies how mortgage servicers shaped the implementation of mortgage forbearance during the COVID-19 pandemic, and that small servicers and nonbanks facilitated fewer forbearances. It states that easier access to forbearance substantially increased mortgage nonpayment but also reduced delinquencies outside of forbearance, and that part of the liquidity was used to reduce credit card debt while most was saved or used for nondurable consumption. A note says the paper was previously released as a FEDS Note in March 2022.
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
“Kim, Lee, Scharlemann, Vickery, Intermediation Frictions in Debt Relief: Evidence from CARES Act Forbearance (NY Fed Staff Report 1035; J. Financial Economics 158 (2024), 2022-10).”
Original at www.newyorkfed.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/kim-lee-scharlemann-vickery-intermediation-frictions-in-debt-relief-ev-f053f17b8c/
, SHA-256 be8f339879cb4f3b1a2c8598cd77881f5fbd542ef8ea17233f70c5c4902930bb.Full text
Intermediation Frictions in
NO. 1035
OCTOBER 2022 Debt Relief: Evidence from
CARES Act Forbearance
You Suk Kim | Donghoon Lee | Tess Scharlemann |
James Vickery
Intermediation Frictions in Debt Relief: Evidence from CARES Act Forbearance
You Suk Kim, Donghoon Lee, Tess Scharlemann, and James Vickery
Federal Reserve Bank of New York Staff Reports, no. 1035
October 2022
JEL classification: G21, G23, G28
Abstract
We study how intermediaries—mortgage servicers—shaped the implementation of mortgage forbearance
during the COVID-19 pandemic and use servicer-level variation to trace out the causal effect of
forbearance on borrowers. Forbearance provision varied widely across servicers. Small servicers and
nonbanks, especially nonbanks with small liquidity buffers, facilitated fewer forbearances and saw a
higher incidence of forbearance-related complaints. Easier access to forbearance substantially increased
mortgage nonpayment but also reduced delinquencies outside of forbearance. Part of the liquidity from
forbearance was used to reduce credit card debt, but most was saved or used for nondurable consumption.
Key words: mortgage, forbearance, liquidity, nonbank, CARES Act, COVID-19
_________________
Lee: Federal Reserve Bank of New York (email: donghoon.lee@ny.frb.org). Kim, Scharlemann: Federal
Reserve Board (emails: you.kim@frb.gov, tess.scharlemann@frb.gov). Vickery: Federal Reserve Bank
of Philadelphia (email: james.vickery@phil.frb.org). This paper was previously released by the Federal
Reserve Board as a FEDS Note in March 2022. The authors thank seminar participants at the Korea
Development Institute, Yonsei University, 2022 SFS Cavalcade, 2022 Fed System Micro Conference,
Federal Reserve Board, AREUEA Virtual Seminar, FDIC Consumer Research Conference, 2022 ASSA
meetings, 2021 Philadelphia Fed Conference on Consumer Behavior in Credit and Payments Markets,
and 2021 AREUEA National Meetings. They also thank discussants Gene Amromin, Neil Bhutta, Meta
Brown, Greg Buchak, Susan Cherry, Lara Lowenstein, and Kyle Mangum, as well as Larry Cordell,
Arpit Gupta, Joe Nichols, and other Federal Reserve colleagues and industry practitioners for comments
and insights about institutional details.
This paper presents preliminary findings and is being distributed to economists and other interested
readers solely to stimulate discussion and elicit comments. The views expressed in this paper are those of
the author(s) and do not necessarily reflect the position of the Federal Reserve Banks of New York or
Philadelphia, the Federal Reserve Board, or the Federal Reserve System. Any errors or omissions are the
responsibility of the author(s).
To view the authors’ disclosure statements, visit
https://www.newyorkfed.org/research/staff_reports/sr1035.html.
1 Introduction
Financial intermediaries are often crucial for implementing public policy, particularly in
the case of debt relief and emergency lending programs.1 Intermediaries have valuable
data, technology, systems, and relationships that can help ensure successful policy out-
comes. On the other hand, misaligned incentives or other frictions may prevent policies
from being implemented as intended “on the ground”.
In this paper we study the role of a particular type of intermediary — mortgage ser-
vicers — in implementing a large debt relief program providing forbearance to mortgage
borrowers during the COVID-19 pandemic. We find that servicers significantly influ-
enced forbearance outcomes, and that variation in servicer behavior is systematically re-
lated to servicer liquidity constraints, size and organizational form. We also use servicer
variation to trace out the causal effects of forbearance for households. Easier access to for-
bearance increased household liquidity by inducing borrowers to pause their payments.
Part of this liquidity infusion was used to pay down high-cost credit card debt, but funds
were primarily used for precautionary saving and/or nondurable consumption.
The forbearance program, authorized by the CARES Act in March 2020, allowed bor-
rowers with federally-backed mortgages to temporarily pause their mortgage payments
without incurring fees, penalties or additional interest and without negative consequences
for their credit history. The borrower simply needed to attest to a pandemic-related hard-
ship to qualify for forbearance; no documentation of income loss was required.
Despite this universal eligibility, a quarter of the mortgages in our sample that became
past-due during the pandemic did not successfully enter into forbearance. Furthermore
the frequency of these “missing” forbearances varied significantly across mortgage ser-
vicers for otherwise equivalent loans. Our analysis focuses on “government” mortgages
1 Examples include emergency business loans under the Paycheck Protection Program (Granja et al., 2020),
mortgage modifications under the Home Affordable Modification Program (Agarwal et al., 2017a), and
streamlined mortgage refinancing under the Home Affordable Refinancing Program (Agarwal et al., 2015).
1
securitized through Ginnie Mae; this is the segment of the mortgage market which serves
the highest-risk borrowers and which, because of institutional factors, poses the greatest
liquidity risk to servicers.
Specifically, using loan-level data we estimate that the conditional probability that a
past-due borrower did not enter forbearance varies between 10% and 60% across ser-
vicers, with a weighted interquartile range of 15 percentage points. Several pieces of
evidence indicate that this variation reflects servicer behavior rather than unobserved
borrower characteristics. The magnitude of these servicer effects is also heterogeneous
across borrowers, and older and low-credit score borrowers appear “hard to reach” in
that they are less likely to enter into forbearance and are not especially responsive to be-
ing matched with a “high-forbearance” servicer.
Investigating these cross-servicer differences, we find that small servicers, nonbanks,
and in particular nonbanks with low cash buffers, were significantly less likely to facilitate
forbearance. These facts suggest that liquidity constraints, as well as some combination
of scale economies and regulatory risk, were important in shaping servicer behavior. Liq-
uidity constraints are important in our setting because Ginnie Mae servicers must finance
payments to investors and other parties when the borrower stops paying. This risk is
most relevant for nonbank servicers, which rely on short-term wholesale debt and cannot
access government liquidity backstops.
The clear benefits of forbearance for borrowers suggest that servicer practices limiting
forbearance uptake also reduced borrower welfare. Consistent with this interpretation,
we show that borrowers were significantly less satisfied with servicers that facilitated
fewer forbearances, based on complaints filed with the Consumer Financial Protection
Bureau (CFPB).
We then use servicer-level variation in forbearance availability to study the causal
effect of forbearance on borrowers. We sort servicers into high (above median) and low
2
(below median) forbearance-availability groups based on the likelihood a past-due loan
received forbearance conditional on loan and borrower characteristics. Then we compare
borrower outcomes between these groups before and after the CARES Act in a difference-
in-differences framework using dynamic mortgage data linked to borrower credit reports.
Studying payment outcomes, we find that assignment to a high-forbearance servicer
reduced the likelihood of the borrower being past-due but not in forbearance by up to
one-quarter (or 0.4 percentage points), with the largest effects early in the pandemic.
However it also caused a much larger number of borrowers to stop making their pay-
ments. Quantitatively, the fraction of past-due borrowers was as much as 5 percentage
points higher at high-forbearance servicers in 2020. (There was no difference prior to
COVID-19.) This finding, that easier access to debt relief induced nonpayment, is remi-
niscent of research on strategic mortgage default, especially Mayer et al. (2014). In our
setting however, several pieces of evidence suggest that these marginal nonpayers were
primarily motivated by precautionary liquidity concerns rather than a purely strategic
calculation to exploit forbearance as an interest-free loan.
Our results therefore indicate that forbearance provided significant liquidity to house-
holds by enabling borrowers to pause their payments. Furthermore the cross-servicer
variation in effective program generosity is quantitatively important. We estimate that
borrowers at high-forbearance servicers deferred an additional $300 in mortgage pay-
ments from April-November 2020, equivalent to $6,000 per marginal forbearance. In
aggregate, switching all Ginnie Mae borrowers from low-to-high forbearance servicers
would increase deferred payments over this short period by $3.1 billion, equivalent to an
effect of ≈ $10 billion if generalizing our estimates to the entire mortgage market.
Part of this liquidity infusion from mortgage payment deferral was used by borrow-
ers to reduce credit card debt, although the effect is limited to less liquidity-constrained
households, defined as a below-median credit card utilization rate. For this group, credit
3
card paydown accounts for about one-fifth of deferred mortgage payments. There is no
evidence that funds were used to establish new auto tradelines, a proxy for auto pur-
chases. We therefore conclude that, at least for borrowers on the margin, funds from pay-
ment deferral were mainly used for precautionary saving or nondurable consumption.
We also confirm that the CARES forbearance program worked as intended to shield bor-
rowers’ credit from adverse consequences of nonpayment, finding a precisely-estimated
effect of forbearance on credit scores close to zero.
We conclude by considering policy implications of our results. Our findings, as well as
those of other researchers, suggest that the CARES Act forbearance program successfully
reached most borrowers in need without inducing widespread strategic behavior or other
serious unintended consequences. However, our results also highlight scope to improve
access and reduce variation in forbearance outcomes unrelated to borrower fundamen-
tals. For example, one possibility would be auto-enrollment in forbearance for borrowers
in observable financial distress (e.g., those drawing unemployment insurance).
Our results also speak to the policy debate about the systemic risk posed by non-
bank mortgage companies, and the debate about the costs and benefits of large banks.
Our results suggest that exposure to liquidity risk reduced nonbank servicers’ willing-
ness to provide liquidity to borrowers. Conversely, large bank servicers had the highest
propensity to facilitate forbearance, likely due to tight post-crisis regulatory scrutiny, scale
economies, and ample liquidity due to access to deposits and the lender of last resort.
1.1 Related literature
We contribute to several strands of literature. First, a number of papers study the be-
havior and incentives of mortgage servicers, in particular studying the 2008 crisis and its
aftermath. Aiello (2021) finds that financial constraints reduced servicers’ propensity to
modify delinquent mortgages, while Agarwal et al. (2017a) show that servicers offered
4
HAMP modifications at divergent rates due to variation in organizational structure and
incentives. Agarwal et al. (2011) and Kruger (2018) find servicers were more likely to
modify mortgages retained in portfolio than loans serviced for other investors. We bring
new data to bear and study a streamlined debt relief program designed to overcome the
frictions that plagued mortgage modification in the wake of the Great Recession. Never-
theless, we still observe large cross-servicer differences in outcomes and find that financial
frictions shaped servicer behavior and borrower outcomes.
Second, a growing literature studies forbearance and other forms of government fi-
nancial assistance during the COVID-19 pandemic. Cherry et al. (2021), An et al. (2022)
and Zhao et al. (2020) present a wealth of information on forbearance takeup, finding that
mortgage forbearance is higher for vulnerable borrowers and those experiencing negative
income shocks. Like us, Cherry et al. (2021) find that nonbanks provided forbearance at
lower rates. Cherry et al. (2022) find that better-capitalized nonbanks were more likely to
provide forbearance, and trace out how nonbanks adjust their balance sheets in response
to the shock of the pandemic. Research on other pandemic relief programs also finds
variation in outcomes across financial intermediaries (e.g., Granja et al. 2020).
Third, we contribute to research on the effects of consumer debt relief (e.g., Agar-
wal et al., 2017a). Our finding that forbearance access induces nonpayment is related
to Mayer et al. (2014), who find that mortgage borrowers strategically default to qualify
for debt relief. Related to our analysis of the effects of payment deferral, several papers
find that mortgage payment reductions reduce delinquency and increase consumption,
among other effects (Abel and Fuster, 2021; Ganong and Noel, 2020; Di Maggio et al.,
2017; Scharlemann and Shore, 2016; Fuster and Willen, 2017). One important distinction
between our setting and these studies, however, is that forbearance represents a payment
deferral, not a permanent reduction in the payments owed.
Finally, our results shed light on the behaviour of nonbank mortgage companies (for
5
other contributions see Buchak et al., 2020; Gete and Reher, 2020; Jiang et al., 2020; Buchak
et al., 2018) and large banks (see e.g., Huber, 2021). We also contribute to a broader liter-
ature studying how financial constraints, size, and organizational frictions affect product
quality and firm outcomes (e.g., Matsa, 2011; Kugler and Verhoogen, 2011; Rose, 1990).
2 Forbearance and the CARES Act
The CARES Act was signed into law on March 27, 2020, and included significant relief
for mortgage borrowers. Homeowners with federally-backed mortgages became eligible
for up to 180 days of forbearance, renewable for an additional 180 days upon request.2,3
Borrowers in forbearance were able to skip their mortgage payments without accruing
unscheduled interest, late fees or penalties, or risking foreclosure. Missed payments were
also not reported as delinquencies to credit bureaus, protecting borrowers’ credit scores.
Eligibility under the CARES Act is very broad, extending to any agency mortgage
borrower experiencing a direct or indirect financial hardship related to the pandemic.
Importantly, the borrower simply needed to attest to a hardship — no documentation or
other proof of income loss was required. Forbearance was not automatic, however; the
borrower had to request and obtain it from their servicer.
The CARES Act is silent about what should occur at the end of the forbearance pe-
riod, but in the weeks after its passage, regulators and the mortgage agencies stated that
a range of options would be available and that a lump-sum repayment of skipped pay-
ments would not be required (e.g., Freddie Mac, 2020). In April 2020, the FHA announced
2 The CARES Act applies directly to “agency” mortgages backed by Fannie Mae, Freddie Mac, the FHA, VA,
and other federal agencies, which make up about 70% of US mortgage debt. Many nonagency borrowers
were still able to obtain forbearance from their servicers, but Cherry et al. (2021) find that the nonagency
forbearance rate was about 25% lower, by studying loans on either side of the conforming loan limit.
3 The forbearance programs were subsequently extended in February 2021. Homeowners already in
forbearance became eligible for a further six months of forbearance, and the enrollment window to request
forbearance was extended to 6/30/2021 (The White House, 2021; Federal Housing Finance Agency, 2021).
6
a “partial claim” program for borrowers resuming payments after exiting forbearance in
which accumulated missed payments could be transferred into a subordinate interest-free
note due at the time the mortgage is paid off (Department of Housing and Urban Devel-
opment, 2020a,b). Borrowers not able to resume payments would be eligible instead for
a loan modification. Fannie Mae and Freddie Mac announced a similar payment deferral
option in May (Federal Housing Finance Agency, 2020). Since deferred payments do not
accrue interest, these programs effectively provide an interest-free loan to the borrower.
Despite these public assurances, there was significant uncertainty and confusion among
borrowers and servicers about post-forbearance options, particularly in the early months
of the pandemic (e.g., Wall Street Journal, 2020; Consumer Financial Protection Bureau,
2021a,b). For example, some servicers incorrectly told borrowers that a lump-sum repay-
ment would be expected upon forbearance exit.
Our analysis focuses on the $2 trillion of “government” mortgages insured by the FHA
and VA, all of which were covered by the CARES Act. This segment of the mortgage mar-
ket is of particular interest because it disproportionately serves low-income and high-risk
borrowers, and because FHA loans in particular experienced a much higher forbearance
and nonpayment rate than the market as a whole. It is also the segment where inter-
mediation frictions are likely to be most severe, because FHA loans are much riskier for
mortgage servicers due to institutional factors (see section 5 for detailed discussion).
2.1 Forbearance trends
Figure 1 traces out the evolution of mortgage forbearance and nonpayment over 2020.
The top panel based on credit bureau data shows that forbearance was rare prior to the
pandemic but rose sharply starting in April, just after the CARES Act was passed. The
aggregate forbearance rate peaked in May at 7.3 percent, falling to 5.2 percent by De-
7
Figure 1: Share of mortgages in forbearance and past-due. “Past-due” is defined as
any loan behind schedule, including mortgages in forbearance where the borrower has
paused their payments with the lender’s consent. Dollar-weighted aggregate statistics
constructed using data from the Federal Reserve Bank of New York (FRBNY) Consumer
Credit Panel / Equifax (panel a) and Black Knight McDash (panel b). Aggregate statistics
reflect agency mortgages covered by the CARES Act as well as mortgages held in portfolio
by banks and other investors and loans securitized through the nonagency market.
(a) Fraction of mortgages in forbearance
(b) Fraction of mortgages 60+ days past due
8
cember.4 The fraction of loans 60+ days past-due also rises and falls along similar lines
(bottom panel), as does the 30+ days-past-due rate (figure A.1 of the Internet Appendix).
Note: “past-due” in this context includes both borrowers who paused their payments
in forbearance and delinquent mortgages not in a forbearance plan. In practice not all
past-due borrowers entered forbearance, and conversely some borrowers entered into
forbearance as a precaution but then kept making their scheduled payments.
As Figure 1 shows, the FHA forbearance and nonpayment rate was much higher than
for the market as a whole, reflecting the lower-income FHA borrower population and
high share of first-time homebuyers. VA mortgages behaved similarly to the overall mar-
ket, while forbearance and nonpayment was relatively low for the typically prime loans
securitized by government sponsored enterprises (GSEs) Fannie Mae and Freddie Mac.
2.2 Forbearance implementation and the role of servicers
One might assume that servicers played a limited and passive role in implementing the
CARES Act forbearance program, given its streamlined design and the lack of required
documentation of hardship. But in practice, qualitative and anecdotal evidence suggests
that servicers varied widely in terms of their level of communication with borrowers
and the information they provided, as well as their systems for receiving and processing
forbearance applications.
For example, Consumer Financial Protection Bureau (2021a) provides a detailed ac-
count of forbearance-related servicing deficiencies observed by CFPB supervisors, includ-
ing: (1) Providing incomplete or false information, such as telling consumers that only
delinquent borrowers qualified for forbearance, that a fee must be paid for forbearance,
or that a lump-sum repayment was required; (2) Incorrectly sending collection or default
4 Other data sources paint a similar picture but indicate a somewhat higher forbearance rate. Survey data
from Mortgage Bankers Association (2020) indicates a peak forbearance rate of 8.55% in June 2020, while
Black Knight (2020) reports a peak forbearance rate of 8.8%, also in June.
9
notices, assessing fees, or initiating foreclosures for borrowers in forbearance; (3) Chang-
ing borrowers’ preauthorized funds transfers without consent, or failing to implement
borrowers’ instructions to freeze payments; (4) Failure to process forbearance requests in
a timely way; (5) Enrolling borrowers in automatic or unwanted forbearance; (6) Failure
to set up an appropriate post-forbearance plan. We heard similar anecdotes in numerous
meetings with credit counselling agencies we arranged as background for this project.
Media reports also highlighted similar issues and described how the wave of forbear-
ance requests early in the pandemic overwhelmed many servicers’ capacity, leading to
long telephone hold times, non-operational servicer websites, and misleading informa-
tion provided to borrowers (e.g. Wall Street Journal, 2020).
These servicing issues are also evident in a sharp rise in forbearance complaints. Con-
sumer Financial Protection Bureau (2021b) calculates based on the CFPB’s repository that
complaints related to forbearance spiked from 3-4% of all mortgage complaints in January
and February 2020 to a peak of 21% in April, remaining persistently at 12-15% over the
rest of 2020 and early 2021. Complaints most commonly cited communication failures,
confusing or incorrect information about post-forbearance options, problems in payment
and forbearance reporting on borrowers’ monthly statements, and delays and denials in
putting the borrower in a post-forbearance repayment plan.5
At the other end of the scale, many servicers took significant steps to streamline the
forbearance process, such as providing a prominent button or link on their website to
a simple online application, and following up with delinquent borrowers frequently to
make them aware of forbearance (e.g., one practitioner told us of a large bank servicer
5 To give a sense of the issues, the following are three complaints taken from the public CFPB database: (1)
“I tried to reach out to <XXX> to request a forbearance ... Unfortunately, I was hung up on two times. I spent
almost 3 hours on hold.”; (2) “My initial 6 month forbearance has been approved, but I’ve been unable to make
contact with the servicer to extend the forbearance. I’ve sent emails, left voice messages and tried online to extend the
forbearance. They do not respond. I’m scared and I need help.”; (3) “I have been trying for over a month to apply for
a 6-month mortgage forbearance plan ( as allowed under the Federal Cares Act ) with <XXX>. If you go to their
website to apply, it doesn’t matter if you are on a mobile device OR hard wired laptop OR desktop computer, it will
not actually let you apply for a forbearance. When you submit, it says ” CRITICAL ERROR ”.”
10
making such calls on a daily basis). We quantify the cross-servicer variation in forbear-
ance policies and outcomes more systematically in the following section.
3 Data and summary statistics
To measure the effects of servicers on forbearance outcomes, we assemble a novel dataset
combining loan-level data on mortgage characteristics and performance, Ginnie Mae for-
bearance records, regulatory data on bank and nonbank servicers, complaints data, and
credit bureau data on borrower liabilities and credit performance. We in fact utilize two
different matches between these underlying datasets, as described below. Further details
on each data source can be found in section A of the Internet Appendix.
Loan-level mortgage data are drawn from eMBS. The data include the universe of se-
curitized FHA and VA mortgages and report the servicer for each loan as well as loan
characteristics and performance. We append data on each loan’s forbearance status and
forbearance terms from Ginnie Mae’s forbearance register. We also match each servicer
by name to servicer-level characteristics (e.g., size, liquidity ratio). For independent mort-
gage banks (“nonbanks”) these characteristics are drawn from the mortgage call report
(MCR) collected by the Conference of State Bank Supervisors, while for banks, they are
drawn from FR Y-9C and Call reports. eMBS itself is also used to calculate some servicer
characteristics (e.g., aggregate servicing volume). We also match the data to forbearance-
related complaints by borrowers aggregated to the servicer level from the CFPB com-
plaints database (see section 5.3 for more details).
For the analysis in section 6 which uses servicer-level variation to study the effects
of forbearance on borrowers, we instead use a merge between eMBS, loan-level data
from Black Knight McDash, and the Equifax Credit Risk Insight Servicing and McDash
(CRISM) dataset. This allows us to trace out effects on nonmortgage outcomes such as
11
credit card debt and a proxy for auto purchases. We match eMBS and McDash/CRISM
based on loan characteristics (for details, see Internet Appendix Section A.1.) From eMBS,
we retain the loan’s forbearance status and an anonymized servicer identifier. (Due to
data use restrictions, we cannot merge servicer characteristics into the CRISM data.) From
CRISM, we draw payment behavior, updated credit scores, geographic data, and addi-
tional details about the borrower’s balance sheet. From McDash we draw mortgage-level
information including some loan and borrower controls not available in eMBS (e.g., prop-
erty location is available at the zip code rather than state level).
3.1 Summary statistics
Table 1 presents loan-level summary statistics from eMBS, reflecting the population of
FHA and VA loans securitized into Ginnie Mae MBS pools as of January 2020. The dataset
includes 10.1 million mortgages, of which about 70% are FHA loans. FHA loans have
higher loan-to-value (LTV) ratios, higher debt-to-income (DTI) and lower average credit
scores, reflecting the lower-income, higher-risk FHA borrower population.
About 5% of loans were at least 30 days past due just before the onset of the pandemic.
Nonpayment then increased sharply, with 18% of loans being 30 days or more past-due at
some point between March and November 2020 (21% of FHA loans and 11% of VA loans).
16% of FHA loans entered forbearance at some point between March and November,
compared to 8% of VA loans. 24% of loans were paid off between March and November,
primarily due to refinancing in response to falling mortgage rates. (Note: we use the term
“past-due” to refer to any loan in arrears relative to its contractual repayment schedule.
This includes loans in forbearance where payments have been paused with the servicer’s
consent, as well as delinquent mortgages not in forbearance.)
Panel C of table 1 reports forbearance and nonpayment statistics for mortgages that
were current as of January 2020. Notably, 26% of loans that became past-due during the
12
Table 1: Summary statistics. Loan-level summary statistics for the eMBS sample. Reflects
the population of active FHA and VA mortgages securitized into Ginnie Mae MBS pools
as of January 2020.
(1) (2) (3)
FHA VA Total
A. Ex-ante loan characteristics:
Unpaid balance ($000, as of Jan 2020) 150,580 207,148 167,304
Original loan-to-value (LTV) (%) 92.93 94.71 93.42
Original debt-to-income (DTI) (%) 41.08 38.45 40.23
Original credit score 682.18 714.80 692.69
Loan age (years, as of Jan 2020) 5.46 4.05 5.02
30+ days past-due in Jan 2020 0.06 0.03 0.05
60+ days past-due in Jan 2020 0.02 0.01 0.02
B. Forbearance & past-due rates during pandemic (Mar-Nov 2020):
Ever 30+ days past due 0.21 0.11 0.18
Ever 60+ days past-due 0.15 0.08 0.13
Ever paid off 0.19 0.34 0.24
Ever in forbearance 0.16 0.08 0.14
C. Conditional forbearance & past-due rates during pandemic (Mar-Nov 2020):
Forbearance | nonpayment (for loans current in Jan 2020):
Ever in forbearance among loans ever 30+ days past-due 0.74 0.70 0.74
Ever in forbearance among loans ever 60+ days past due 0.91 0.88 0.91
Nonpayment | forbearance (for loans current in Jan 2020):
Ever 30+ days past-due among loans ever in forbearance 0.84 0.84 0.84
Ever 60+ days past due among loans ever in forbearance 0.71 0.72 0.71
N. Obs. 6,943,846 3,185,050 10,128,896
pandemic failed to enter into a forbearance plan. This is quite striking given that any
FHA or VA borrower experiencing financial stress related to the pandemic was eligible
for forbearance, and given that forbearance effectively provided a subsidy to the bor-
rower because no interest was charged on deferred balances. This fraction of “missing”
forbearances is significantly lower — 9% — for loans entering serious delinquency (60+
days past due), but still well above zero. Conversely, 16% of borrowers remained cur-
rent on their payments despite entering into forbearance. Most borrowers in forbearance
skipped multiple payments however, with 71% becoming at least 60 days past due.
13
4 Servicer-level variation in forbearance outcomes
We measure cross-servicer variation in forbearance outcomes by estimating the following
cross-sectional linear probability model using eMBS loan-level data:
forbearancei = Xi β + ξ s + ei . (1)
The dependent variable is an indicator for whether mortgage i entered forbearance
from March-November 2020, ξ s is a vector of servicer fixed effects, and Xi is a set of loan
controls (e.g., LTV and credit score bins) to account for forbearance demand. (Coefficient
estimates on these controls are reported in table A.2 of the Internet Appendix.)
Our baseline model estimates equation 1 on the sample of borrowers that were cur-
rent prior to the onset of the pandemic (January 2020) but missed at least one payment
from March to November. This set of borrowers would unambiguously benefit from for-
bearance, but as we have discussed, around a quarter of them became past-due without
successfully entering into a forbearance plan.
Figure 2 plots the distribution of the servicer fixed effects (ξbs ), showing very wide
variation in forbearance outcomes across servicers for observably similar mortgages.6 For
the figure we normalize the fixed effects to show the probability that a past-due loan
with sample average characteristics fails to enter into forbearance. The likelihood that the
borrower “falls through the cracks” ranges from under 10% to almost 60%. This variation
is not simply due to disparate outcomes among very small servicers. Weighting by loan
count, the “no forbearance” probability is 38% for a servicer at the 90th percentile of the
distribution compared to only 12% for a servicer at the 10th percentile.
The bottom panel of figure 2 presents the same histogram conditioning on more se-
6 Indeed, these estimated servicer fixed effects are highly jointly statistically significant (f-statistic = 435).
Moreover, including the servicer fixed effects doubles the R2 of our model (comparing columns 1 and 2 in
table A.2).
14
rious nonpayment (60+ days past due). The share of “missing” forbearances is signifi-
cantly smaller for this group, but in proportionate terms the cross-servicer variation is
even more stark — the likelihood of not receiving forbearance is six times higher for a
“low-forbearance” servicer at the 90th percentile of the distribution compared to a “high-
forbearance” servicer at the 10th percentile (18% compared to 3%).
4.1 Alternative estimates of servicer effects
The servicer effects presented above are quite robust to alternative modelling choices.
First, we estimate an alternative set of servicer fixed effects using the eMBS-CRISM matched
sample. This allows us to control for a finer set of controls incorporating information from
borrower credit reports, including bins of borrower age, an updated credit score and non-
mortgage debt balances.7 This approach also produces a similarly wide dispersion of
servicer effects estimates (see figure A.2 of the Internet Appendix).
Second, we then use the eMBS-CRISM model to measure how sensitive the fixed ef-
fects are to the set of controls used, comparing specifications with i) no controls, ii) the
controls available in eMBS only, and iii) the full set of eMBS-CRISM controls. We find that
the three resulting sets of servicer fixed effects are highly positively correlated (see figure
A.3 of the Internet Appendix).
Third, within the eMBS sample, we estimate the servicer effects three other ways aside
from the two presented in figure 2: i) including all mortgages in the sample, rather than
just the loans that became past due during the pandemic; ii) restricting the sample to
borrowers that became past-due early in the pandemic (February or March), prior to the
passage of the CARES Act; and iii) including lender fixed effects, so that servicer fixed
7 Coefficients on loan and borrower controls for this specification are reported in table A.3 of the Internet
Appendix. Note that terms-of-use restrictions on the CRISM dataset prevent us from retaining servicer
information in the merged eMBS-CRISM dataset; we are however permitted to retain anonymous servicer
identifiers, which is what we use to estimate the servicer fixed effects.
16
effects are identified only from mortgages where there was a transfer of servicing. This
third approach is motivated by the fact that borrowers do select their lender (in ways that
may be correlated with unobservables) but do not control whether the servicing on their
loan is subsequently sold to a third party. These alternative fixed effects are strongly posi-
tively correlated with our main estimates in figure 2 (figure A.4 of the Internet Appendix).
4.2 Servicer behavior or omitted borrower characteristics?
We interpret these striking differences in forbearance outcomes as being due to variation
in servicer policies and practices. But an alternative explanation is that they reflect unob-
served differences in forbearance demand. For instance borrowers at “high-forbearance”
servicers may be more liquidity constrained and therefore benefit more from an extended
payment holiday, or may be more financially literate. Our estimated fixed effects condi-
tion on a rich set of borrower and loan controls, particularly for the eMBS-CRISM sample,
but of course do not control for all factors that may affect forbearance demand.8
However, three additional pieces of evidence suggest the servicer fixed effects we mea-
sure are not driven by unobserved borrower heterogeneity:
1. Mortgages managed by high- vs low- forbearance servicers have similar ex-ante
characteristics, measured in either the eMBS or eMBS-CRISM samples (see Inter-
net Appendix tables A.5-A.7). Borrower non-mortgage loan balances are also sim-
ilar (e.g., auto, credit card and student loan balances are all within 10%), and the
two groups of loans experienced similar macroeconomic conditions during the pan-
demic (e.g., the 12-month change in the county unemployment rate differs by only
0.2%). Loans managed by low-forbearance servicers are somewhat younger (4.5 vs
8 Servicer forbearance policies per se were not likely to have been important in borrowers’ mortgage choice
prior to the pandemic, because borrowers cannot typically directly choose their servicer, and because
forbearance is an infrequent event particularly given the stable economy and rising home prices leading
into the pandemic. Even so, it is still possible there could be nonrandom assignment of borrowers to
servicers in a way that is correlated with borrowers’ demand for forbearance during the pandemic.
17
6.0 years in the eMBS-CRISM sample); but within age bins mortgages look very
similar on observables, and our regressions also always include loan age controls.
2. There is little or no difference in mortgage nonpayment between high and low-
forbearance servicers in the months prior to the pandemic. The same is true for
credit card and auto delinquencies in the eMBS-CRISM matched sample. We mea-
sure this by estimating account-level delinquency models where the dependent
variable is equal to 1 if a borrower current at t-1 becomes delinquent in month t.9
Controlling for loan characteristics, differences in delinquency transitions for bor-
rowers managed by high-vs-low forbearance servicers are economically small, not
consistently signed, and generally not statistically significant (see figure A.5 and ta-
ble A.9 of the Internet Appendix). In contrast, during the pandemic itself, borrowers
assigned to high-forbearance servicers become much more likely to stop paying their
mortgages, as seen in figure A.5 and as discussed in detail in section 6.
These findings are evidence against the hypothesis that high-forbearance-servicer
borrowers were riskier on unobservables, because such an explanation would pre-
dict a higher nonpayment rate not just during the pandemic, but also prior to it.
They also speak against the story that high-forbearance-servicer borrowers were
more financially literate, because this would be expected to produce a lower pre-
COVID delinquency rate in line with Gerardi et al. (2013) and Agarwal et al. (2017b).
3. Estimated servicer fixed effects are generally insensitive to the set of borrower and
loan controls used, as shown in section 4.1. In other words, there is little evidence of
selection on observables. Among these results, servicer effects are robust to includ-
9 Measuring transitions into delinquency is preferable to measuring the stock of delinquent loans, for two
reasons: i) servicer quality can affect the length of time a mortgage remains delinquent, e.g., better-quality
servicers may make it easier for their borrowers to cure or obtain a loan modification; ii) servicers have the
option to purchase seriously delinquent loans out of Ginnie Mae pools — such loans would no longer
appear in the eMBS data after they are repurchased. This could create a selection effect since e.g., since
banks are more likely to repurchase loans than nonbanks.
18
ing lender fixed effects so that identification comes only from servicing transfers.
This suggests that cross-servicer variation in forbearance outcomes is not due to
non-random matching of borrowers to lenders, because although borrowers pick
their lender, they do not control whether the servicing is sold after origination.
4.3 Heterogeneity
Do servicers affect the behavior of all borrowers equally? Table 2 uses the eMBS-CRISM
matched sample to study variation in the effect of servicers on forbearance outcomes
across borrowers with different characteristics, finding evidence of significant hetero-
geneity. Column 1 of the table shows that, overall, a past-due borrower is 13.6 percentage
points more likely to enter into forbearance when matched to a “high-forbearance” ser-
vicer, defined by a servicer fixed effect that is above the median. Columns 2-6 then interact
this “high-forbearance” dummy with various borrower characteristics.
Our prior is that assignment to a high-forbearance servicer would have a larger effect
for borrowers with a low propensity to seek forbearance, where there is more scope for
behavior to change. This is what we see in columns 2 and 3; borrowers with low cur-
rent mortgage balances, and borrowers in locations less-hard hit by the COVID recession
measured by the local change in unemployment, are both less likely to obtain forbearance
overall and more sensitive to whether they are matched with a high-forbearance servicer.
We observe a different pattern, however, for two relatively vulnerable groups: older
borrowers and borrowers in poor current financial health as indicated by a low credit
score measured just prior to the pandemic (columns 4 and 5). These two groups have
low overall forbearance takeup, but their behavior is either less responsive to mortgage
servicer assignment, or at least no more responsive, than the sample as a whole. (Results
look similar in the multivariate specification in column 6.)
These results speak to the idea that a subset of borrowers are “difficult to reach” —
19
Table 2: Heterogeneity in servicer effects. Regressions studying heterogeneity in the
effects of servicers on forbearance outcomes by borrower characteristics. Data are the
matched eMBS-CRISM sample, restricted to borrowers who were current in January 2020
but missed at least one payment between March and November. High-forbearance ser-
vicer is defined as a servicer with an above-median estimated fixed effect; similarly other
explanatory variables are dummies equal to 1 if the variable is above its sample median
value. Standard errors are clustered at the servicer level.
Dependent variable = 1 if mortgage received forbearance, = 0 otherwise
(1) (2) (3) (4) (5) (6)
High-forbearance servicer 0.136∗∗∗ 0.182∗∗∗ 0.147∗∗∗ 0.120∗∗∗ 0.137∗∗∗ 0.189∗∗∗
(0.021) (0.024) (0.024) (0.022) (0.023) (0.029)
High-forbearance servicer:
× High unpaid loan balance -0.054∗∗∗ -0.044∗∗∗
(0.010) (0.009)
× High ∆ unemp rate (yoy) -0.021∗∗ -0.015∗∗
(0.009) (0.006)
× High updated credit score 0.020 0.003
(0.013) (0.011)
× High borrower age -0.003 0.002
(0.009) (0.008)
High unpaid loan balance 0.123∗∗∗ 0.069∗∗∗
(0.009) (0.010)
High ∆ unemp rate (yoy) 0.061∗∗∗
(0.006)
High updated credit score 0.057∗∗∗ 0.068∗∗∗
(0.012) (0.009)
High borrower age -0.020∗∗∗ -0.013∗
(0.007) (0.007)
Zipcode FE N N N N N Y
Other controls N N N N N Y
N. Obs. 431,478 411,939 431,083 431,478 431,008 405,464
Adj. R2 0.02 0.04 0.03 0.03 0.03 0.09
20
they are less likely to seek forbearance overall, and are also not particularly responsive to
servicer efforts to make forbearance easier to obtain. This is apparent in our later results
as well; although easier access to forbearance induces many borrowers to pause their
payments, it only moderately reduces the number of past-due borrowers outside of the
forbearance safety net. Finally we note that although we find significant heterogeneity,
the overall effect of servicers on forbearance outcomes is broadly based rather than being
driven by a particular group, seen by the fact that the uninteracted “high-forbearance”
dummy remains positive and highly significant in all columns of table 2.
5 Servicer characteristics and forbearance outcomes
Next we examine the economic factors shaping servicer behavior by studying how a ser-
vicer’s propensity to provide forbearance, as measured by its fixed effect, varies with
servicer characteristics such as size, liquidity and organizational form.
5.1 Economic drivers of servicer behavior
Economic forces that may shape servicer forbearance practices include:
1. Liquidity constraints. When a borrower stops making payments, the mortgage
servicer is required to temporarily finance and advance payments on the borrower’s be-
half, including principal, interest, taxes and insurance. Servicers facing binding liquidity
constraints therefore may wish to discourage borrowers from entering forbearance, to
limit these cash outflows.10 The liquidity risk of nonpayment is particularly significant
for FHA loans, because FHA servicers must forward payments for a much longer period
10 Note that it is nonpayment rather than forbearance per se that creates a liquidity drain on the servicer.
While the two do not mechanically go hand-in-hand, we show empirically below that easier access to
forbearance does in fact causally lead to higher nonpayment, almost one-to-one.
21
and face significant delays and costs before being reimbursed, and also because FHA bor-
rowers have higher default risk (Pence, 2022; Kim et al., 2018).11
Nonbank mortgage companies are much more exposed to liquidity risk than banks,
because they rely on short-term wholesale funding rather than insured deposits and do
not have access to government liquidity backstops such as Federal Home Loan Bank ad-
vances or the discount window (Jiang et al., 2020). Reflecting this fragility, there were
widespread fears in the early months of the pandemic about nonbank liquidity and the
possibility of runs and a wave of nonbank failures (Pence, 2022; Loewenstein, 2021).
2. Regulatory and legal risk. Mortgage intermediaries were forced to pay out large
legal settlements after the Great Recession, and today face much stricter regulation.12 It
therefore seems plausible that legal, regulatory and reputational risk could induce ser-
vicers to adopt “borrower-friendly” practices that make forbearance easier to obtain.
Large commercial bank servicers are likely to be most concerned about these risks, be-
cause these firms are highly visible, face the toughest regulatory scrutiny, and were sub-
ject to the largest post-crisis legal settlements (Buchak et al., 2018).
3. Capitalization and risk-shifting. Servicers face a risk-return tradeoff in the sense
that improving servicing quality and customer satisfaction (e.g., generous servicing poli-
cies, better training or technology) is costly in the short run but may reduce future legal
risk and improve retention in the long run. Undercapitalized servicers may thus have
weaker incentives to act in the borrower’s best interests by making forbearance simple to
obtain, in line with the classic risk-shifting hypothesis of Jensen and Meckling (1976).
11 FHA servicers must typically forward payments until loan termination or modification or until the loan is
repurchased by the servicer using its own funds, unlike GSE loans where advances are capped at four
months. FHA servicers also face significant delays before being reimbursed for payment shortfalls, and
Tozer (2019) estimates they are also typically not compensated for about $10,000 in costs per FHA claim.
12 Additional post-crisis regulation includes national servicing standards, higher bank capital requirements
on servicing rights, and supervisory oversight from the new Consumer Financial Protection Bureau
(CFPB). Legal risk is also much more salient given the scale of post-crisis settlements (Buchak et al., 2018).
In related work, Fuster et al. (2021b) find that CFPB supervision and enforcement results in more
consumer-friendly mortgage servicing practices.
22
4. Size, scale and technology. A large literature finds that size and organizational
form play a key role in shaping financial intermediary behavior (e.g., Berger et al., 2005).
In our context, e.g., scale economies in technology adoption may have meant that large
servicers entered the pandemic with more sophisticated online servicing platforms, facil-
itating borrower communication and enabling mass processing of forbearance requests.
Or conversely, small, nimble servicers may have been able to adjust their practices more
quickly than large bureaucratic organizations with several layers of management.
5.2 Empirical analysis
To investigate which factors are most relevant empirically, we regress the servicer fixed
effects estimated previously on servicer characteristics drawn from mortgage call reports
(for nonbank mortgage companies), Y-9C and bank call reports (for banks or nonbanks
controlled by a bank), and servicer-level aggregations of eMBS loan-level data.13
Estimates are reported in table 3 and reveal several patterns. First, borrowers at large
servicers are significantly more likely to enter into a forbearance plan, whether size is
measured by the log of servicing assets (measured using eMBS) or balance sheet size
(taken from regulatory reports). Second, organizational form matters. Nonbank mort-
gage companies are about 9 percentage points less likely to offer forbearance to a past-due
borrower, while credit unions are about 13 percentage points more likely. Third, the level
of internal liquidity at the start of the pandemic, measured by the ratio of cash to total as-
sets, is strongly positively correlated with forbearance provision, but only for nonbanks.
These results support the view that liquidity constraints shaped servicer behavior. As
discussed above, nonbanks were highly exposed to liquidity risk early in the pandemic
13 We match financial institutions by name across these data sources. Data on financial structure from the
National Information Center and other sources is used to cross-validate the accuracy of the match. Our
analysis focuses on banks, credit unions and nonbank mortgage companies, and excludes government and
government-sponsored enterprises such as state housing authorities and Federal Home Loan Banks.
23
Table 3: Determinants of servicer effects. Servicer-level regression of servicer forbear-
ance fixed effects on characteristics drawn from bank and nonbank Call reports and
eMBS. Column 1 is based on all servicers including banks, credit unions and nonbanks.
Columns 2-4 reflect nonbank mortgage company servicers only. Columns 5-7 reflect bank
servicers only. Weighted least squares, weighted by number of borrowers that were cur-
rent in January 2020 but past due between March and November. Robust standard errors.
Dependent variable: servicer fixed effect. (Higher value ⇔ higher P(forbear | nonpay))
All Nonbank mtg companies Banks
(1) (2) (3) (4) (5) (6) (7)
Servicer characteristics
log(Servicing assets) 0.035∗∗∗ 0.030∗∗∗ 0.027∗∗∗ 0.038∗∗∗ 0.039∗∗∗
(0.006) (0.008) (0.005) (0.009) (0.010)
log(Assets) 0.019∗∗∗ 0.025∗
(0.005) (0.013)
Cash / assets 0.919∗∗∗ 1.047∗∗∗ -0.661 -0.890
(0.185) (0.191) (0.511) (0.651)
Securities / assets 0.100 0.186∗∗ 0.251 0.454
(0.085) (0.090) (0.354) (0.314)
Capital / assets 0.032 0.080 1.079 0.763
(0.104) (0.113) (0.699) (0.798)
Servicing growth -0.045 -0.003 -0.002 -0.019 -0.118 -0.085 -0.104
(0.049) (0.058) (0.048) (0.048) (0.076) (0.082) (0.080)
Servicer type
Nonbank mortgage company -0.084∗∗∗
(0.025)
Credit union 0.186∗∗∗
(0.032)
N. Obs. 152 98 98 98 45 45 45
24
when most forbearance applications were received. We find that nonbank servicers were
significantly less likely to provide forbearance, particularly for small nonbanks with low
cash balances that faced the greatest liquidity risk. Banks in contrast were not liquidity-
constrained because they have access to ample backstop sources of funding for mortgages
and also experienced large deposit inflows after the onset of COVID-19 (Li et al., 2020).
The high forbearance rate for large servicers, both banks and nonbanks, is also strik-
ing. Although it is difficult to pinpoint the mechanism underlying this result, large ser-
vicers may benefit from scale economies in technology investments (e.g., a well-designed
online platform), may have better access to capital markets and more resources to train
servicing staff, or may take a “borrower-friendly” approach because they are more likely
to be targeted by financial regulators, particularly in the case of large banks.14
5.3 Servicing quality: evidence from CFPB complaints
It seems clear given the program’s design that past-due FHA and VA borrowers would
unambiguously benefit from entering into forbearance; this in turn suggests that ser-
vicer practices limiting forbearance uptake also reduced borrower welfare for our sam-
ple. To investigate further, we study whether borrowers were less satisfied with “low-
forbearance” servicers based on the frequency of mortgage forbearance-related complaints
for government loans submitted by borrowers to the CFPB complaint platform.15
14 Results in table 3 are related to contemporaneous analysis by Cherry et al. (2022) also studying the
relationship between forbearance provision and servicer characteristics using a different but overlapping
sample. Cherry et al. (2022) do not investigate the role of liquidity constraints, which is likely to be
particularly important for our sample given the much greater liquidity risk associated with Ginnie Mae
mortgages (as discussed in section 5.2). In other respects our results are consistent with Cherry et al.
(2022); e.g., both studies find that large servicers and bank servicers are more likely to provide forbearance.
15 We identify forbearance-related complaints using a similar approach to Consumer Financial Protection
Bureau (2021b), searching for complaints with a narrative field containing the string “forbear” or “defer”,
restricting the sample to complaints related to a mortgage which is a government loan, to be consistent
with our Ginnie Mae sample. As a cross-check, we confirm that we identify a comparable total sample to
Consumer Financial Protection Bureau (2021b). We match CFPB complaints data to our main servicer
dataset by name. We exclude from the sample any servicer for which we are unable to find a match;
results are however similar if we retain these servicers and code them as having zero complaints.
25
Table 4: Servicer forbearance practices and CFPB complaints. Servicer-level regression
of relationship between volume of forbearance-related CFPB complaints and servicer for-
bearance practices (as measured by servicer fixed effects). Outcome variable is the num-
ber of forbearance-related mortgage complaints for government loans per thousand Gin-
nie Mae mortgages serviced. Weighted least squares, weighted by size of Ginnie Mae
servicing portfolio as of January 2020. Robust standard errors.
Dependent variable: Complaints per thousand loans serviced
All lenders Nonbanks only
(1) (2) (3) (4) (5) (6)
Servicer forbearance propensity -0.222∗∗∗ -0.235∗∗∗ -0.622∗∗
(0.073) (0.077) (0.303)
Servicer characteristics
log(Servicing assets) -0.016∗∗ -0.006 -0.031
(0.007) (0.008) (0.041)
Cash / assets -0.410∗ -1.090∗∗
(0.229) (0.485)
Securities / assets -0.356∗∗ -0.891∗∗∗
(0.175) (0.301)
Capital / assets -0.020 0.440
(0.140) (0.524)
Servicing growth 0.079 0.095 0.731∗
(0.060) (0.061) (0.409)
Frac. govt. loans that are FHA 0.069∗∗ 0.072∗∗ 0.089 -0.337 -0.486
(0.032) (0.036) (0.077) (0.503) (0.562)
Frac. all loans that are FHA
Servicer type
Nonbank mortgage company 0.002 0.012 -0.033
(0.020) (0.022) (0.034)
Credit union 0.070∗∗ 0.000
(0.028) (0.025)
N. Obs. 129 129 129 125 92 92
26
Results are presented in Table 4. The dependent variable is the frequency of com-
plaints per thousand Ginnie Mae mortgages serviced. The rate of complaints is signif-
icantly higher for low-forbearance servicers. This is direct evidence of poorer servicing
quality for these firms. The inverse relationship between forbearance provision and com-
plaints is particularly strong for nonbanks (column 5).
When we replace the servicer fixed effect in the regression with servicer characteris-
tics (columns 3, 4 and 6), we find again that servicer liquidity matters. Servicers with
lower cash balances and smaller securities portfolios were the object of a higher rate of
forbearance-related complaints. Again, these relationships are concentrated among non-
bank servicers. These results and those of the prior table show how liquidity constraints
can lead to a deterioration of servicing quality, consistent with earlier evidence on fore-
closures and modifications from the period of the Great Recession (Aiello, 2021).
6 Effects of forbearance on borrowers
In this section we use cross-servicer variation to estimate the causal effect of forbearance
access on borrower outcomes, including payment behavior, nonmortgage debt, auto pur-
chases and credit scores. We then draw out implications and lessons from our results for
the overall design and effectiveness of the CARES Act forbearance program.
We use the CRISM-eMBS matched sample for this portion of the analysis. This sample
allows us to observe nonmortgage debt and other outcomes from credit reports, measure
geography more finely and control for a richer set of borrower and loan controls, and track
payment status even for nonperforming loans repurchased from Ginnie Mae pools.16
16 Ginnie Mae MBS issuers have the option to repurchase nonperforming mortgages out of securitized pools
at par if the borrower has missed at least three payments. This was an attractive option during the
pandemic because many pools were trading at a premium to par given record-low interest rates. Since
eMBS is a dataset of loans in securitized pools, it does not include data on performance after loans are
repurchased. We do continue to observe performance in the CRISM-eMBS matched sample, however.
27
6.1 Empirical strategy
We use a difference-in-difference approach to compare outcomes for borrowers matched
to servicers with a high-versus-low propensity to provide forbearance. “High-forbearance”
servicers are again defined as those with an above-median servicer fixed effect, estimated
as described in Section 4.1 using the merged eMBS-CRISM data. We use the six months
up to March 2020 to establish the absence of differential pre-trends between high- and
low-forbearance servicers. We attribute differences in borrower outcomes and behavior
observed after March 2020 to variation in forbearance access post-CARES Act.
We trace out the dynamic effects of servicer behavior by regressing various borrower
outcomes (e.g., payment status) on a “high-forbearance servicer” dummy interacted with
a set of time dummies as well as a set of borrower and loan controls and geography by
time fixed effects. Specifically we estimate:
Yit = β t SiH + Zit γ + αs + αztτ + ε it (2)
where Yit is the borrower outcome in question for loan i in month t; β t are coefficients
on a vector of time dummies that are interacted with SiH , a high-forbearance servicer
dummy; Zit is a vector of loan and borrower controls including loan characteristics at
origination, the borrower’s updated credit score (measured by the Equifax Risk Score) as
of January 2020, updated principal balance, loan age, borrower age, and loan type (FHA
vs. VA); αs is a vector of servicer dummies; and αztτ is a vector of zipcode × month ×
origination year (τ) fixed effects to account for the time-varying geographic effects of the
pandemic separately for different loan cohorts. Standard errors are clustered by servicer.
Like our earlier analysis, the estimation sample consists of loans that were active and
current in January 2020, to exclude loans that were already past-due before the pandemic.
We also exclude loans originated after October 2019, the start of our sample period.
28
6.2 Nonpayment
Figure 3 traces out the estimated effect of assignment to a high-forbearance servicer on
mortgage payment outcomes based on the β̂ t s from equation 2.
The top panel of figure 3 shows that easier access to forbearance induces a very sig-
nificant increase in the mortgage nonpayment rate. The fraction of past-due borrowers at
high vs low-forbearance servicers track each other closely through March 2020, but then
diverge sharply after the passage of the CARES Act. Mortgage nonpayment rises sharply
for both groups, but the probability that a borrower becomes past-due is much higher for
borrowers at high-forbearance servicers, by more than half, or in level terms as much as
5 percentage points, at the forbearance peak around May 2020.
Notably however, high-forbearance servicers are associated with a significant reduc-
tion in the number of borrowers that are past-due but not in forbearance (bottom panel of
figure 3). This is particularly evident in the early months of the pandemic.
Table 5 summarizes and further unpacks these effects. The table reports average co-
efficients on the high-forbearance × time dummies during three phases of the pandemic
for five different forbearance and payment outcomes.
The first row of table 5 reports the effect on the forbearance rate itself. In the early
stages of the pandemic (April-July 2020), assignment to a high-forbearance servicer in-
creases the forbearance rate by 5.6 percentage points (pp), a quantitatively important ef-
fect compared to the overall forbearance rate of 8.1 percent at low-forbearance servicers.
The effect declines slightly to 4.5 percent later in 2020.
The second and third rows report estimates for nonpayment and nonpayment outside
of forbearance, summarizing the visual evidence from figure 3. Comparing the first and
second rows, a key takeaway point is that the effects on nonpayment are almost as large as
the effects on forbearance itself (e.g., 4.9pp compared to 5.6pp for the April-July period).
29
Table 5: Forbearance and nonpayment outcomes. Estimates of the average effect of as-
signment to a high-forbearance servicer on five different payment and forbearance out-
comes. Estimates reported in columns (1), (3) and (5) are the average coefficient on the
high-forbearance-servicer × time dummies (estimates of β t from equation 2), over three
phases of the pandemic: a pre-pandemic period (October 2019-February 2020); early pan-
demic (April-July 2020) and later pandemic (August-November 2020), along with the as-
sociated standard error of each mean. For context, Columns (2), (4) and (6) report the
unconditional mean of the dependent variable at low-forbearance servicers during the
period referenced. Standard errors are clustered at the servicer level.
Outcome variable: Pre-pandemic Pandemic
2019:m10-2020:m2 2020:m4 to 2020:m7 2020:m8 to 2020:m11
(1) (2) (3) (4) (5) (6)
Coeff. Mean Coeff. Mean Coeff. Mean
Forbearance 0.002 0.001 0.056∗∗∗ 0.081 0.045∗∗∗ 0.091
(0.002) (0.009) (0.010)
Missed payment 0.002 0.016 0.049∗∗∗ 0.069 0.037∗∗∗ 0.083
(0.002) (0.007) (0.007)
Missed payment, no forbearance -0.001 0.016 -0.004∗∗ 0.017 -0.004∗∗ 0.022
(0.002) (0.002) (0.002)
Forbearance, no missed payment -0.001 0.000 0.003 0.029 0.005 0.029
(0.002) (0.005) (0.006)
Forbearance, missed payment 0.003 0.000 0.053∗∗∗ 0.052 0.041∗∗∗ 0.061
(0.002) (0.007) (0.008)
31
In other words, a first-order effect of easier forbearance access is to induce nonpayment
among borrowers who otherwise would have continued performing on their mortgage.
We discuss the interpretation of this finding in section 6.5.
Easier forbearance access also reduces the number of past-due borrowers not in for-
bearance by about 0.4pp. This is a much smaller absolute effect, although in proportion-
ate terms it represents a significant 20-25 percent reduction in the number of “missing”
forbearances relative to the sample average for low-forbearance servicers.
As we have discussed, some borrowers entered forbearance as a precaution but contin-
ued making their scheduled payments; the fourth row of table 5 shows however that the
share of such borrowers does not differ systematically between low- and high-forbearance
servicers. The final row of table 5 confirms that the primary effect of easier forbearance
access is to increase the share of borrowers who both skipped payments and entered for-
bearance (these estimates are essentially the difference between rows 2 and 3).
These results show that servicer policies significantly affected liquidity provision to
households during the pandemic. Based on our estimates and some auxiliary assump-
tions, we calculate that borrowers matched to high-forbearance servicers deferred an ad-
ditional ≈ $300 in cumulative mortgage payments by November 2020 compared to oth-
erwise equivalent borrowers at low-forbearance servicers (see Figure A.7 in the Internet
Appendix). Since the treatment effect on the forbearance rate itself is about 5pp, this im-
plies, on the margin, deferred payments of $6,000 per additional forbearance, a significant
sum. In aggregate, switching all Ginnie Mae borrowers from low-to-high forbearance
servicers would increase deferred payments over the short period from April-November
2020 by $3.1 billion, equivalent to an effect of ≈ $10 billion if generalizing our estimates to
the entire mortgage market.17 Next we study how this liquidity was used by borrowers.
17 The estimate of $3.1 billion is computed by multiplying the estimate of the cumulative deferred payment
from Figure A.7 in the Internet Appendix by the number of FHA and VA mortgages outstanding as
reported in our table of summary statistics. Our estimate of the aggregate effect of ≈ $10 billion is
32
6.3 Nonmortgage debt
First we study credit card debt, an alternative high-interest-rate form of borrowing often
used by households during periods of financial stress. Liquidity constraints are likely to
play a key role in determining whether funds from forbearance are used to reduce debt
rather than for more immediate needs such as nondurable consumption (e.g., Telyukova,
2013; Gross and Souleles, 2002; Zeldes, 1989). Correspondingly in the spirit of Gross and
Souleles (2002) we split the sample into high- and low- credit-card utilization groups
based on the ratio of drawn balances to total credit limits summed across cards (measured
ex ante over the six months to March 2020) and estimate results separately by group.
Figure 4 shows that the liquidity provided by forbearance did indeed allow some bor-
rowers to reduce credit card balances, with the effect concentrated among less liquidity-
constrained (i.e., low utilization) households.18 For the low-utilization group, assignment
to a high-forbearance servicer reduced credit card debt by 1.2 percent between April and
July and 1.4 percent between August and November (top panel of figure 4). This is an av-
erage effect across all borrowers regardless of forbearance status; scaled by the treatment
effect on the forbearance rate of around 5pp, it represents a 20-30 percent reduction in
credit-card debt for the marginal borrowers induced to enter into forbearance, account-
ing for about one-fifth of the total liquidity infusion provided by forbearance for the low-
utilization group.19 In contrast, we find little or no corresponding change in credit card
debt for high-utilization borrowers (bottom panel of figure 4).
We find no evidence that borrowers used forbearance to pay down other types of
computed by then grossing up this estimate by the fraction of all forbearances that were in the FHA/VA
segment as of the peak in June 2020, taken from Black Knight (2020).
18 The figure is based on estimating our difference-in-difference equation using log(credit card debt) as the
outcome variable. As an alternative specification, figure A.6 in the Internet Appendix presents estimates
using the level of credit card debt; these alternative estimates show similar patterns to figure 4.
19 Given the average credit-card balance of $3,910 for low-utilization borrowers, a 1.4% paydown amounts to
$55 per borrower. By comparison, we estimate that assignment to a high-forbearance servicer results in an
additional ≈ $300 in total cumulative deferred mortgage-related payments up to November 2020 (see
figure A.7 of the internet appendix), 5-6 times larger than the estimated effect on credit card balances.
33
debt like auto loans, student debt, or junior home equity liens (table A.10 in the Internet
Appendix), perhaps because these forms of borrowing are cheaper than credit card debt,
making them a lower priority for payoff. Reported in the same table, we find little or no
effects on delinquency for non-mortgage debt. This may in part reflect the availability of
forbearance for these other debt types.
6.4 Other outcomes
6.4.1 Credit scores
Employing the same methodology, Figure 5 shows that easier access to forbearance did
not damage the credit scores of borrowers at high-forbearance servicers, despite the high
nonpayment rate for this group.20 This outcome is consistent with the intended design
of the CARES Act forbearance program, which stipulated that nonpayment in forbear-
ance should not be reported as a delinquency to credit bureaus. The point estimate is in
fact slightly positive, which could be possible if, e.g,., forbearance reduced non-mortgage
delinquency, although it is not close to statistical significance.
6.4.2 Auto purchases
How much of the liquidity made available through forbearance was used for consump-
tion? Although credit bureau data unfortunately do not in general allow us to measure
consumption directly, we do study a proxy for durable goods purchases often used in the
literature; the establishment of new auto credit trade lines as a measure of automobile
purchases (e.g., Abel and Fuster, 2021; Di Maggio et al., 2017). We find no evidence that
borrowers at high-forbearance servicers were more likely to purchase an automobile —
20 Credit score is measured by FICO score version 5. FICO is a registered trademark of Fair Isaac
Corporation.
35
borrowers to exit forbearance first before refinancing. In practice however, we find little
or no effect on prepayment. This implies that borrowers assigned to high-forbearance
servicers were not diverted from refinancing into forbearance, an outcome that would
have complicated the welfare analysis of the program given the substantial benefits of
refinancing for borrowers.
6.4.4 Bankruptcy
Wang et al. (2020) document a striking decline in consumer bankruptcy filings during
the COVID pandemic. We find no evidence that servicers significantly affected the like-
lihood of bankruptcy; the difference in the bankruptcy rate (including bankruptcy of any
type) between high-and-low forbearance servicers is only about 1 percent of the sample
mean, and quite tightly estimated. This may seem surprising in light of evidence that
bankruptcy filings are very sensitive to cash-on-hand (Indarte, 2020). A possible explana-
tion is that we measure a local treatment effect: borrowers in sufficient distress to be close
to bankruptcy likely had strong incentives to file for forbearance regardless of servicer,
particularly given the complexity of the bankruptcy process compared to forbearance.
6.5 Summary and policy implications
Stepping back, what can we learn from our results about the overall design and effec-
tiveness of the CARES Act forbearance program? For example, was the program too
generous, resulting in widespread moral hazard and strategic default by borrowers not
facing liquidity problems? Or alternatively was the program not streamlined enough,
as indicated by the many borrowers who became delinquent without obtaining forbear-
ance? Cross-servicer variation helps shed light on these questions because, as we have
shown, it produced some quasi-random variation in program generosity on the margin.
37
Our finding that access to forbearance induced nonpayment by borrowers is closely
related to the strategic default literature, and in particular to Mayer et al. (2014) who find
that mortgage borrowers defaulted in order to qualify for generous modifications from
subprime lender Countrywide during the 2008 financial crisis. There are some important
differences between the two settings, however, and several pieces of evidence described
below suggest that the marginal nonpayers in our sample primarily sought forbearance
because of genuine liquidity concerns rather than because of a strategic desire to take
advantage of a source of interest-free borrowing.
First, unlike Mayer et al. (2014) we find that marginal nonpayers look similar on ob-
servables (e.g., nonmortgage balances, credit scores) to “control group” borrowers who
obtained forbearance from low-forbearance servicers; see Table A.8 in the Internet Ap-
pendix. If moral hazard was the main driver of nonpayment, we might instead expect
marginal nonpayers to be financially literate, higher-income borrowers, as Mayer et al.
(2014) do find in their setting. Furthermore, although borrowers at high-forbearance ser-
vicers stay in forbearance slightly longer and are less likely to exit, the effects are small
(e.g., the probability of forbearance exit is 0.31 at high-forbearance servicers compared to
0.35 at low-forbearance servicers.) In other words there is little evidence that borrowers
that obtained forbearance on the margin acted to “max out” the zero-interest financing
provided by payment deferral.
Second, the liquidity made available through forbearance was generally not used to
pay down debt or to purchase automobiles, suggesting deferred payments were primar-
ily used for precautionary savings or nondurable consumption. This interpretation is
also consistent with survey data from April 2020 presented in Anderson et al. (2021).
Households were asked how they would use funds from forbearance; the most popular
response was spending on “necessary” consumption, followed by saving, and then debt
consolidation. Although we find evidence of credit-card debt paydown, this response is
38
limited to less liquidity-constrained households and even for this group accounts for only
about one-fifth of deferred payments.
Third, external evidence suggests that forbearance was mostly used by borrowers ex-
periencing negative income or expenditure shocks. Lambie-Hanson et al. (2021) present
survey data that at least three-quarters of borrowers entering forbearance had experi-
enced a job disruption or income loss. Zhao et al. (2020) document using rich adminis-
trative data that borrowers in forbearance had experienced larger income declines, and
were more likely to have lost their jobs or to have have received unemployment benefits.
Further, in aggregate less than one in ten borrowers made use of forbearance, despite the
easy qualification requirements. This suggests opportunistic behavior was relatively rare.
A final point: in our context the benefits of strategic default are fairly modest because
forbearance is only an interest-free payment deferral; it is not debt forgiveness, unlike
the setting in Mayer et al. (2014). Consistent with this point, An et al. (2022) find that
less than one-in-ten borrowers exiting forbearance maximized the interest benefit of pay-
ment deferral by rolling skipped payments into a long-term “partial claim” due when the
mortgage is paid off. Requiring even a simple attestation of economic hardship may have
limited strategic behavior, in line with experimental evidence in Anderson et al. (2021).
Aside from the question of strategic default, our results show the CARES Act forbear-
ance program worked as intended to allow borrowers to pause their payments without
negative effects on their credit scores; furthermore it did not inadvertently prevent bor-
rowers from refinancing. More broadly, the program successfully reached a high propor-
tion of vulnerable borrowers — three-quarters of FHA and VA borrowers who became
past-due successfully obtained forbearance, rising to nine-tenths for seriously past-due
borrowers, a high takeup rate compared to many government programs.
That said, the failure to reach all past-due borrowers highlights the forbearance pro-
gram’s limits, as does the wide variation in outcomes across servicers unrelated to bor-
39
rower fundamentals. These policy shortcomings could cause more serious and long-
lasting problems in a future stress event involving a more prolonged economic and hous-
ing market downturn. In designing future debt relief policies, policymakers may there-
fore wish to consider ways to standardize servicer practices (e.g., more detailed guide-
lines about minimum borrower outreach) or implement forms of forbearance auto-enrolment
for borrowers observably in distress (e.g., tied to unemployment insurance claims). Auto-
enrolment may be particularly valuable for “hard-to-reach” borrowers identified in our
analysis such as those with low credit scores.
Our results, and those of Cherry et al. (2022), also show how debt relief outcomes
are connected to the financial health and regulation of mortgage intermediaries. Non-
banks are now responsible for a majority of mortgage lending and servicing, but these
entities are less regulated than banks and face significant liquidity and run risk (Pence,
2022; Kim et al., 2018). We show that nonbanks were less likely to provide debt relief,
particularly when the nonbank was small or had low liquid asset buffers. In contrast
the more “borrower-friendly” outcomes for large bank servicers may reflect tighter reg-
ulation and oversight of these intermediaries as well as their lower liquidity risk. Our
findings highlight the importance of appropriate nonbank regulation and the potential
benefits of public nonbank liquidity backstops such as that provided in 2020 through the
Ginnie Mae Pass-Through Assistance Program (Ginnie Mae, 2020).
7 Conclusion
We show that mortgage intermediaries played a key role in shaping the implementation
of the CARES Act mortgage forbearance program. Forbearance outcomes varied widely
across servicers for otherwise similar loans. Small servicers, nonbanks, and particularly
nonbanks with low liquid asset buffers facilitated fewer forbearances and saw a higher
40
volume of forbearance-related borrower complaints. Servicer effects are heterogeneous
across borrowers with older and low credit-score borrowers seemingly “hard to reach”.
We also use cross-servicer variation to trace out the causal effects of forbearance for
borrowers, showing that easier access to forbearance resulted in significant liquidity pro-
vision by inducing nonpayment among borrowers who otherwise would have kept mak-
ing their mortgage payments. Several pieces of evidence suggest that these “marginal”
nonpayers generally paused their payments due to liquidity concerns rather than purely
strategic considerations, although part of the liquidity was used to consolidate debt through
credit card debt paydown.
Overall, we interpret our results as evidence that the CARES Act forbearance program
successfully reached most borrowers in need without inducing widespread strategic be-
havior or other unintended consequences, thereby balancing the tradeoffs inherent in any
social insurance program (e.g., see Chetty 2008 on unemployment insurance or Indarte
2020 on personal bankruptcy). That said, significant idiosyncratic variation in outcomes
across servicers as well as the failure to reach a significant share of past-due borrowers in-
dicate program limitations which could be mitigated through changes in design. Further
consideration of optimal program design seems prudent given that forbearance is likely
to be an important debt-relief tool in the future.
41
References
A BEL , J. AND A. F USTER (2021): “How Do Mortgage Refinances Affect Debt, Default,
and Spending? Evidence from HARP,” American Economic Journal: Macroeconomics, 13,
254–91.
A GARWAL , S., G. A MROMIN , I. B EN -D AVID , S. C HOMSISENGPHET, AND D. E VANOFF
(2011): “The role of securitization in mortgage renegotiation,” Journal of Financial Eco-
nomics, 102, 559–578.
A GARWAL , S., G. A MROMIN , I. B EN -D AVID , S. C HOMSISENGPHET, T. P ISKORSKI , AND
A. S ERU (2017a): “Policy Intervention in Debt Renegotiation: Evidence from the Home
Affordable Modification Program,” Journal of Political Economy, 125, 654–712.
A GARWAL , S., G. A MROMIN , S. C HOMSISENGPHET, T. L ANDVOIGT, T. P ISKORSKI ,
A. S ERU , AND V. YAO (2015): “Mortgage Refinancing, Consumer Spending, and Com-
petition: Evidence from the Home Affordable Refinancing Program,” NBER Working
Papers 21512, National Bureau of Economic Research, Inc.
A GARWAL , S., S. C HOMSISENGPHET, AND Y. Z HANG (2017b): “How does working in a
finance profession affect mortgage delinquency?” Journal of Banking Finance, 78, 1–13.
A IELLO , D. J. (2021): “Financially constrained mortgage servicers,” Journal of Financial
Economics.
A N , X., L. C ORDELL , L. G ENG , AND K. L EE (2022): “Inequality in the Time of COVID-19:
Evidence from Mortgage Delinquency and Forbearance,” Tech. rep.
A NDERSON , J., D. H ARRISON , AND M. S EILER (2021): “Reducing Strategic Forbearance
under the CARES Act: an Experimental Approach Utilizing Recourse Attestation,” The
Journal of Real Estate Finance and Economics.
42
B ERGER , A. N., N. H. M ILLER , M. A. P ETERSEN , R. G. R AJAN , AND J. C. S TEIN (2005):
“Does function follow organizational form? Evidence from the lending practices of
large and small banks,” Journal of Financial Economics, 76, 237–269.
B LACK K NIGHT (2020): “Forbearances rise after three weeks of declines,” McDash Flash
Blog June 26.
B UCHAK , G., G. M ATVOS , T. P ISKORSKI , AND A. S ERU (2018): “Fintech, regulatory arbi-
trage, and the rise of shadow banks,” Journal of Financial Economics, 130, 453 – 483.
——— (2020): “Beyond the Balance Sheet Model of Banking: Implications for Bank Reg-
ulation and Monetary Policy,” .
C HERRY, S., E. J IANG , G. M ATVOS , T. P ISKORSKI , AND A. S ERU (2022): “Shadow Bank
Distress and Household Debt Relief: Evidence from the CARES Act,” AEA Papers and
Proceedings, 112, 509–15.
C HERRY, S. F., E. X. J IANG , G. M ATVOS , T. P ISKORSKI , AND A. S ERU (2021): “Govern-
ment and Private Household Debt Relief during COVID-19,” NBER Working Papers
28357, National Bureau of Economic Research.
C HETTY, R. (2008): “Moral Hazard versus Liquidity and Optimal Unemployment Insur-
ance,” Journal of Political Economy, 116, 173–234.
C ONSUMER F INANCIAL P ROTECTION B UREAU (2021a): “COVID-19 Prioritized Assess-
ments, Special Edition,” Supervisory Highlights 23.
——— (2021b): “Mortgage forbearance issues described in consumer complaint,” Com-
plaint Bulletin, April.
D EPARTMENT OF H OUSING AND U RBAN D EVELOPMENT (2020a): “FHA’s COVID-19
Loss Mitigation Options,” Mortgagee Letter 2020-22, July 8.
43
——— (2020b): “HUD Issues New CARES Act Mortgage Payment Relief for FHA Single
Family Homeowners,” Press Release, April 1.
D I M AGGIO , M., A. K ERMANI , B. K EYS , T. P ISKORSKI , R. R AMCHARAN , A. S ERU , AND
V. YAO (2017): “Monetary Policy Pass-Through: Mortgage Rates, Household Con-
sumption, and Voluntary Deleveraging,” American Economic Review, 107, 3550–88.
F EDERAL H OUSING F INANCE A GENCY (2020): “FHFA Announces Payment Deferral as
New Repayment Option for Homeowners in COVID-19 Forbearance Plans,” News re-
lease, May 13.
——— (2021): “FHFA Extends COVID-19 Forbearance Period and Foreclosure and REO
Eviction Moratoriums,” Press release, February 25.
F REDDIE M AC (2020): “Freddie Mac: Lump Sum Repayment is Not Required in Forbear-
ance,” Homeownership blog, April 16.
F USTER , A., A. H IZMO , L. L AMBIE -H ANSON , J. V ICKERY, AND P. W ILLEN (2021a):
“How Resilient is Mortgage Credit Supply? Evidence from the COVID-19 Pandemic,”
Working Paper.
F USTER , A., M. P LOSSER , AND J. V ICKERY (2021b): “Does CFPB Oversight Crimp
Credit?” CEPR Discussion Papers 15681, C.E.P.R. Discussion Papers.
F USTER , A. AND P. S. W ILLEN (2017): “Payment Size, Negative Equity, and Mortgage
Default,” American Economic Journal: Economic Policy, 9, 167–91.
G ANONG , P. AND P. N OEL (2020): “Liquidity versus Wealth in Household Debt Obli-
gations: Evidence from Housing Policy in the Great Recession,” American Economic
Review, 110, 3100–3138.
G ERARDI , K., L. G OETTE , AND S. M EIER (2013): “Numerical ability predicts mortgage
default,” Proceedings of the National Academy of Sciences, 110, 11267–11271.
44
G ETE , P. AND M. R EHER (2020): “Mortgage Securitization and Shadow Bank Lending,”
The Review of Financial Studies, 34, 2236–2274.
G INNIE M AE (2020): “Availability of Pass-Through Assistance Program for Participants
in Ginnie Mae’s Single-Family MBS Program,” All Participants Memo APM 20-03.
G RANJA , J., C. M AKRIDIS , C. YANNELIS , AND E. Z WICK (2020): “Did the Paycheck Pro-
tection Program Hit the Target?” NBER Working Papers 27095, National Bureau of
Economic Research, Inc.
G ROSS , D. B. AND N. S. S OULELES (2002): “Do Liquidity Constraints and Interest Rates
Matter for Consumer Behavior? Evidence from Credit Card Data,” The Quarterly Journal
of Economics, 117, 149–185.
H UBER , K. (2021): “Are Bigger Banks Better? Firm-Level Evidence from Germany,” Jour-
nal of Political Economy, 129, 2023–2066.
I NDARTE , S. (2020): “Moral Hazard versus Liquidity in Household Bankruptcy,” Working
paper.
J ENSEN , M. C. AND W. H. M ECKLING (1976): “Theory of the firm: Managerial behavior,
agency costs and ownership structure,” Journal of Financial Economics, 3, 305–360.
J IANG , E., G. M ATVOS , T. P ISKORSKI , AND A. S ERU (2020): “Banking without Deposits:
Evidence from Shadow Bank Call Reports,” Working Paper 26903, National Bureau of
Economic Research.
K IM , Y. S., S. M. L AUFER , K. P ENCE , R. S TANTON , AND N. WALLACE (2018): “Liquidity
Crises in the Mortgage Market,” Brookings Papers on Economic Activity, 49, 347–428.
K RUGER , S. (2018): “The effect of mortgage securitization on foreclosure and modifica-
tion,” Journal of Financial Economics, 129, 586–607.
45
K UGLER , M. AND E. V ERHOOGEN (2011): “Prices, Plant Size, and Product Quality,” The
Review of Economic Studies, 79, 307–339.
L AMBIE -H ANSON , L., J. V ICKERY, AND T. A KANA (2021): “Recent Data on Mortgage
Forbearance: Borrower Uptake and Understanding of Lender Accommodations,” Re-
search Brief, Federal Reserve Bank of Philadelphia.
L EE , D. AND W. V. DER K LAAUW (2010): “An introduction to the FRBNY Consumer
Credit Panel,” Staff Reports 479, Federal Reserve Bank of New York.
L I , L., P. E. S TRAHAN , AND S. Z HANG (2020): “Banks as Lenders of First Resort: Evi-
dence from the COVID-19 Crisis,” The Review of Corporate Finance Studies, 9, 472–500.
L OEWENSTEIN , L. (2021): “Why Wasn’t there a Nonbank Mortgage Servicer Liquidity
Crisis?” Economic Commentary.
M ATSA , D. A. (2011): “Running on Empty? Financial Leverage and Product Quality in
the Supermarket Industry,” American Economic Journal: Microeconomics, 3, 137–73.
M AYER , C., E. M ORRISON , T. P ISKORSKI , AND A. G UPTA (2014): “Mortgage Modifi-
cation and Strategic Behavior: Evidence from a Legal Settlement with Countrywide,”
American Economic Review, 104, 2830–57.
M ORTGAGE B ANKERS A SSOCIATION (2020): “Share of mortgage loans in forbearance in-
creases to 8.55%,” Press Release June 15.
P ENCE , K. (2022): “Liquidity in the mortgage market: How does the COVID-19 crisis
compare with the global financial crisis?” Real Estate Economics.
R OSE , N. L. (1990): “Profitability and Product Quality: Economic Determinants of Airline
Safety Performance,” Journal of Political Economy, 98, 944–964.
46
S CHARLEMANN , T. C. AND S. H. S HORE (2016): “The Effect of Negative Equity on Mort-
gage Default: Evidence From HAMP’s Principal Reduction Alternative,” The Review of
Financial Studies, 29, 2850–2883.
T ELYUKOVA , I. A. (2013): “Household Need for Liquidity and the Credit Card Debt Puz-
zle,” The Review of Economic Studies, 80, 1148–1177.
T HE W HITE H OUSE (2021): “Fact Sheet: Biden Administration Announces Extension of
COVID-19 Forbearance and Foreclosure Protections for Homeowners,” Release, Febru-
ary 16.
T OZER , T. (2019): “A Primer and Perspective on Ginnie Mae,” Milken Institute Discussion
Paper.
WALL S TREET J OURNAL (2020): “Struggling Borrowers Want To Pause Their Mortgage
Payments. It Hasn’t Been Easy,” Tech. Rep. April 16.
WANG , J., J. YANG , B. C. I VERSON , AND R. K LUENDER (2020): “Bankruptcy and the
COVID-19 Crisis,” Available at SSRN 3690398.
Z ELDES , S. P. (1989): “Consumption and Liquidity Constraints: An Empirical Investiga-
tion,” Journal of Political Economy, 97, 305–346.
Z HAO , C., D. FARRELL , AND F. G REIG (2020): “Did Mortgage Forbearance Reach the
Right Homeowners? Income and Liquid Assets Trends for Homeowners during the
COVID-19 Pandemic,” Policy Brief, J.P. Morgan Chase Institute.
47
Internet Appendix for:
“Intermediation Frictions in Debt Relief: Evidence from
CARES Act Forbearance”
You Suk Kim, Donghoon Lee, Tess Scharlemann, and James Vickery
September 6, 2022
A Datasets
eMBS loan-level data. eMBS provides information on the characteristics of the popula-
tion of mortgages securitized into agency MBS. The data include standard underwriting
fields such as credit score at origination, loan-to-value ratio, loan amount, mortgage rate,
and property location (state). The data set also includes dynamic information about loan
performance, such as updated principal balance, nonpayment status, and crucial for our
analysis, the servicer identity. Our sample consists of FHA and VA loans, which account
for 92% of all loans securitized into Ginnie Mae MBS.
Ginnie Mae forbearance register. We measure forbearance outcomes using Ginnie
Mae data listing the monthly loan-level forbearance history of loans securitized into Gin-
nie Mae MBS. The file indicates the start date of the forbearance policy, the scheduled
end date, and the number of months of forbearance granted. The data were first released
publicly in June 2020, and were backfilled to the start of the pandemic for loans that were
in forbearance as of June. They have subsequently been updated on a monthly basis.1
Financial Call Reports. Data on servicer characteristics are drawn from quarterly
regulatory filings. For bank servicers we use the bank call reports and FR Y-9C. For inde-
pendent mortgage banks we use mortgage call reports (MCRs) data. MCRs are filed by
financial data companies holding a license through the Nationwide Mortgage Licensing
System, including all bank and nonbank agency MBS servicers. The data include balance
sheet and income data and other information on business activities. Together the bank
and nonbank call report datasets allow us to link servicer characteristics to forbearance
and delinquency outcomes.
Black Knight McDash and CRISM. Black Knight McDash (hereafter “McDash”) in-
cludes loan characteristics and performance for the servicing portfolios of the largest res-
idential mortgage servicers in the US, covering around two-thirds of the servicing mar-
ket. The Equifax Credit Risk Insight Servicing and McDash (CRISM) dataset is a match
between McDash and credit bureau data on nearly 79 million individual consumers, in-
cluding information on other forms of debt (e.g., credit cards, junior liens, and student
loans) for primary borrowers and all co-borrowers on the McDash mortgages.
Federal Reserve Bank of New York (FRBNY) Consumer Credit Panel / Equifax Data
(CCP). The CCP is a representative panel of the credit history of an anonymous 5% sam-
1 One relatively minor reporting issue is that the initial release of the forbearance data only includes loans
that were in forbearance as of June 2020. Thus, the data do not allow us to observe a forbearance spell for
borrowers who entered forbearance in March but had already exited prior to June.
1
ple of the U.S. adult population (see Lee and der Klaauw (2010) for details of the dataset).
Narrative codes in the CCP together with scheduled payment variables allow us to mea-
sure the incidence of mortgage forbearance. The CCP does not include loan performance
data for mortgages in forbearance plans, since that information is not reported to credit
bureaus. We use the CCP to calculate forbearance rates for the overall mortgage market
(Figure 1), and to cross-validate the forbearance information in the Ginnie Mae data.
A.1 Details of eMBS-CRISM merge
Unlike eMBS, CRISM does not report the identity of the servicer. We are however able
to merge CRISM with anonymized servicer identifiers through a fuzzy match between
CRISM/McDash with eMBS loan-level data, matching on mortgage balance at origina-
tion, origination year-month, mortgage rate, credit score, whether a loan is an FHA or VA
loan, and state.2
This eMBS-CRISM matched dataset allows us to trace out the effects of servicer varia-
tion in forbearance practices on other borrower outcomes (e.g., credit card debt and credit
scores). It also enriches the set of available borrower-level characteristics relative to the
eMBS-only dataset. For example, CRISM/McDash includes finer geographic information
on the property location, and allows us to observe the borrower’s refreshed credit score
just prior to the pandemic. A limitation however is that only a subset of loans can be
matched with precision, whereas in eMBS we essentially are able to observe the entire
universe of FHA and VA mortgages.
Table A.1 reports summary statistics of loan characteristics for the full eMBS data and
for the merged eMBS-CRISM dataset. As shown by the number of observations in the two
columns, about 30% of loans in the eMBS data are matched to CRISM; this reflects both
the fact that CRISM does not cover the entire market, and our restrictive matching criteria
(we require essentially an exact match on all fields). The characteristics of matched loans
are however very similar to the full eMBS sample, however.
2 The Federal Reserve’s terms of use agreement with Black Knight does not permit us to retain servicer
characteristics in this merged dataset. We are permitted to retain an anonymized servicer identifier,
however. This allows us to measure servicer-level variation in forbearance outcomes, by estimating fixed
effects for these identifiers.
2
Table A.1: Comparison between eMBS and eMBS-CRISM matched sample. Summary
statistics reflect eMBS data fields, and are measured as of January 2020.
(1) (2)
eMBS eMBS-CRISM match
Ever 30+ days past-due 0.17 0.18
Ever in forbearance 0.13 0.14
Current UPB ($) 171,731.42 173,088.57
Orig LTV (%) 93.43 94.60
Orig DTI (%) 40.36 40.25
Orig credit score 692.56 696.80
Loan age (year) 4.97 5.33
FHA 0.68 0.70
VA 0.32 0.30
N. Obs. 11,015,574 3,068,450
3
B Mortgages 30+ days past due, by segment
Figure A.1: Past-Due Rate, 30+ Days. Fraction of active mortgages that are at least 30
days past due relative to scheduled payments, inclusive of mortgages that are in forbear-
ance. Calculations based on Black Knight McDash servicing data.
4
C Loan-level estimates
C.1 eMBS sample
Table A.2: First-stage forbearance regression. Dependent variable = 1 if mortgage en-
tered forbearance from March-November 2020. Cross-sectional linear probability model.
eMBS loan-level data. Sample is loans active as of January 2020. Sample for columns 1
and 2 restricted to mortgages that became past due from March-November 2020.
Past-due sample Full sample
(1) (2) (3) (4)
Excluding Svcr FE Including Svcr FE Excluding Svcr FE Including Svcr FE
Ever servicer change -0.061∗∗∗ -0.021∗∗∗ -0.002∗∗∗ 0.003∗∗∗
(0.001) (0.002) (0.000) (0.000)
Months since last servicer change 0.000∗∗∗ 0.000∗∗∗ -0.000∗∗∗ -0.000∗∗∗
(0.000) (0.000) (0.000) (0.000)
yes FTHB 0.035∗∗∗ 0.030∗∗∗ 0.023∗∗∗ 0.022∗∗∗
(0.001) (0.001) (0.000) (0.000)
Debt-to-income (DTI) at origination:
25 < dti ≤ 50 0.019∗∗∗ 0.043∗∗∗ 0.021∗∗∗ 0.026∗∗∗
(0.002) (0.002) (0.000) (0.000)
dti > 50 0.058∗∗∗ 0.081∗∗∗ 0.060∗∗∗ 0.064∗∗∗
(0.002) (0.002) (0.001) (0.001)
Loan age (year) 0.001∗∗ -0.016∗∗∗ -0.000∗∗∗ -0.004∗∗∗
(0.000) (0.000) (0.000) (0.000)
Loan age (year) × Loan age (year) -0.000∗∗∗ 0.000∗∗∗ -0.000∗∗ 0.000∗∗∗
(0.000) (0.000) (0.000) (0.000)
Ln(Current UPB) 0.108∗∗∗ 0.104∗∗∗ 0.028∗∗∗ 0.026∗∗∗
(0.001) (0.001) (0.000) (0.000)
Credit score (CS) at origination:
620 < orig cs ≤ 680 0.052∗∗∗ 0.016∗∗∗ -0.009∗∗∗ -0.017∗∗∗
(0.001) (0.001) (0.000) (0.000)
680 < orig cs ≤ 740 0.072∗∗∗ 0.025∗∗∗ -0.050∗∗∗ -0.061∗∗∗
(0.002) (0.002) (0.000) (0.001)
orig cs > 740 0.062∗∗∗ 0.010∗∗∗ -0.081∗∗∗ -0.093∗∗∗
(0.002) (0.002) (0.001) (0.001)
Loan purpose:
refinace 0.033∗∗∗ 0.035∗∗∗ -0.001∗∗∗ 0.002∗∗∗
(0.002) (0.002) (0.000) (0.000)
Loan-to-value (LTV) at origination:
80 < LTV ≤ 95 0.025∗∗∗ 0.026∗∗∗ 0.008∗∗∗ 0.006∗∗∗
(0.002) (0.002) (0.000) (0.000)
95 < LTV ≤100 0.032∗∗∗ 0.034∗∗∗ 0.017∗∗∗ 0.016∗∗∗
(0.002) (0.002) (0.000) (0.000)
LTV > 100 0.036∗∗∗ 0.045∗∗∗ 0.022∗∗∗ 0.019∗∗∗
(0.003) (0.003) (0.001) (0.001)
FHA 0.077∗∗∗ 0.099∗∗∗ 0.064∗∗∗ 0.065∗∗∗
(0.001) (0.001) (0.000) (0.000)
30+ days past-due in Jan 2020 -0.305∗∗∗ -0.300∗∗∗
(0.010) (0.010)
Servicer fixed effects N Y N Y
State fixed effects Y Y Y Y
N. Obs. 1,189,326 1,189,326 9,774,503 9,774,503
Adj. R2 0.05 0.10 0.07 0.08
5
C.2 eMBS-CRISM sample
Table A.3: First-stage forbearance regression: eMBS-CRISM. Dependent variable = 1 if
mortgage entered forbearance from March-November 2020. Cross-sectional linear prob-
ability regression model. eMBS-CRISM matched loan-level sample. Sample is loans that
were active as of January 2020 and became past due from March-November 2020.
(1) (2)
Forbearance | past-due Forbearance | past-due
30 <age ≤ 45 0.0157∗∗∗
(0.003)
45 <age ≤ 60 0.0129∗∗∗
(0.003)
age >60 -0.0278∗∗∗
(0.003)
Riskscore (Feb 2020)+ 0.000225∗∗∗
(0.000)
Ln(Consumer debt)+ 0.00522∗∗∗
(0.000)
Delinq. consumer debt+ -0.00217∗∗∗
(0.000)
Other housing debt+ 0.00356∗∗∗
(0.000)
Delinq. other housing debt+ -0.00252∗∗
(0.001)
Credit utilization+ 0.0185∗∗∗
(0.002)
First Time Homebuyer 0.0244∗∗∗ 0.0294∗∗∗
(0.002) (0.002)
25 <dti ≤ 50 0.0481∗∗∗ 0.0566∗∗∗
(0.004) (0.004)
dti >50 0.0776∗∗∗ 0.0883∗∗∗
(0.004) (0.004)
Age of loan (years) -0.0146∗∗∗ -0.0128∗∗∗
(0.001) (0.001)
Age of loan (years) × Age of loan (years) 0.000148∗∗∗ 0.0000562
(0.000) (0.000)
Ln(Current UPB) 0.0759∗∗∗ 0.102∗∗∗
(0.002) (0.001)
620 <orig cs ≤ 680 -0.00406 0.00728∗∗
(0.003) (0.003)
680 <orig cs ≤ 740 -0.00592∗ 0.0142∗∗∗
(0.003) (0.003)
orig cs >740 -0.02187∗∗∗ 0.0000∗∗∗
(0.003) (0.003)
Refinance 0.0191∗∗∗ 0.0188∗∗∗
(0.003) (0.003)
80 <LTV ≤ 95 0.0172∗∗∗ 0.0180∗∗∗
(0.004) (0.003)
95 <LTV ≤ 100 0.0235∗∗∗ 0.0256∗∗∗
(0.004) (0.003)
LTV >100 0.0238∗∗∗ 0.0281∗∗∗
(0.005) (0.004)
FHA 0.0641∗∗∗ 0.0820∗∗∗
(0.002) (0.002)
N 416298 421941
Servicer fixed effects Yes Yes
State fixed effects No Yes
Zipcode fixed effects Yes No
Standard errors in parentheses
+ Measured as of February 2020
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
6
E Summary statistics: servicer-level sample
Table A.4: Servicer level summary statistics. Servicing assets and growth are measured
using eMBS. Financial characteristics for banks and nonbank mortgage companies are
measured using bank and nonbank Call reports.
(a) All servicers
Mean Std. dev. Median
Servicer forbearance propensity 0.00 0.11 -0.01
Nonbank mortgage company 0.62 0.49 1.00
Credit union 0.01 0.11 0.00
log(Servicing assets) 25.63 1.63 26.07
Servicing growth 0.08 0.23 0.04
Observations 152
(b) Nonbank mortgage companies only
Mean Std. dev. Median
Servicer forbearance propensity -0.04 0.08 -0.04
log(Servicing assets) 25.43 1.49 26.01
log(Assets) 8.86 1.59 9.71
Servicing growth 0.12 0.23 0.16
Cash / assets 0.05 0.04 0.04
Securities / assets 0.08 0.10 0.00
Capital / assets 0.21 0.09 0.19
Observations 98
(c) Banks only
Mean Std. dev. Median
Servicer forbearance propensity 0.07 0.11 0.10
log(Servicing assets) 26.03 1.75 26.60
log(Assets) 13.26 1.77 14.50
Servicing growth 0.02 0.23 -0.06
Cash / assets 0.08 0.03 0.08
Securities / assets 0.20 0.07 0.22
Capital / assets 0.12 0.02 0.12
Observations 45
10
F Borrower characteristics: high-vs-low forbearance servicers
Table A.5: Ex ante borrower characteristics across servicers: eMBS-CRISM matched
sample. Summary statistics measured as of January 2020 for high- and low-forbearance
servicers using the merged eMBS-CRISM data. We define “high-forbearance” servicers as
those with above-median servicer fixed effects (estimated as described in Section 4, using
the merged eMBS-CRISM data).
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 184,736.49 165,272.18
Auto Loan Balance 16,103.48 15,331.09
Credit Card Balance 8,951.05 8,740.05
12-mo change CNTY UR (8/20) 6.07 5.86
FHA 0.69 0.70
FICO V5 (updated) 694.22 703.49
LTV at origination 93.94 94.28
Loan age (year) 4.50 6.01
N. Obs. 1,270,977 1,626,621
11
Table A.6: Ex ante borrower characteristics across servicers by origination year: eMBS-
CRISM matched sample. Summary statistics measured as of January 2020 for high-
and low-forbearance servicers using the merged eMBS-CRISM data. We define “high-
forbearance” servicers as those with above-median servicer fixed effects (estimated as
described in Section 4, using the merged eMBS-CRISM data).
(a) Origination year up to 2013
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 136,335.32 137,011.41
Auto Loan Balance 13,399.83 13,955.28
Credit Card Balance 9,167.34 8,856.06
12-mo change CNTY UR (8/20) 6.16 5.87
FHA 0.83 0.77
FICO V5 (updated) 711.80 707.34
LTV at origination 93.66 93.67
Loan age (year) 8.52 8.66
N. Obs. 350,093 741,283
(b) Origination year from 2014 to 2017
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 185,845.28 180,704.64
Auto Loan Balance 16,547.14 16,564.76
Credit Card Balance 9,320.18 8,996.56
12-mo change CNTY UR (8/20) 6.11 5.88
FHA 0.71 0.68
FICO V5 (updated) 696.65 702.02
LTV at origination 93.92 94.92
Loan age (year) 4.62 4.79
N. Obs. 363,104 553,867
(c) Origination year since 2018
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 221,930.34 209,119.38
Auto Loan Balance 17,511.62 16,346.45
Credit Card Balance 8,574.99 8,052.02
12-mo change CNTY UR (8/20) 5.98 5.81
FHA 0.59 0.59
FICO V5 (updated) 681.59 697.33
LTV at origination 94.11 94.58
Loan age (year) 1.90 2.12
N. Obs. 557,780 331,471
12
Table A.7: Ex ante borrower characteristics across servicers by origination year: eMBS
sample Summary statistics measured as of January 2020 for high- and low-forbearance
servicers using the eMBS sample. We define “high-forbearance” servicers as those with
above-median servicer fixed effects (estimated as described in Section 4, using the eMBS
sample).
(a) Origination year up to 2013
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 114,464.79 118,119.20
12-mo change CNTY UR (8/20) 5.95 5.91
FHA 0.80 0.77
Orig credit score 699.49 705.85
Orig LTV (%) 92.62 92.67
Loan age (year) 10.26 10.11
N. Obs. 1,039,878 1,893,974
(b) Origination year from 2014 to 2017
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 178,017.12 175,241.83
12-mo change CNTY UR (8/20) 6.09 5.87
FHA 0.69 0.61
Orig credit score 690.85 702.36
Orig LTV (%) 93.38 93.06
Loan age (year) 4.62 4.71
N. Obs. 1,150,984 1,200,757
(c) Origination year since 2018
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Current Mortgage Balance 216,955.67 203,721.30
12-mo change CNTY UR (8/20) 6.15 5.77
FHA 0.69 0.58
Orig credit score 683.56 696.60
Orig LTV (%) 94.56 93.46
Loan age (year) 2.05 2.12
N. Obs. 1,843,638 1,326,624
13
Table A.8: Characteristics of borrowers in forbearance: high vs low forbearance ser-
vicers. Summary statistics measured as of January 2020 for borrowers that were ever in
forbearance, for high- vs low-forbearance servicers based on the merged eMBS-CRISM
data.
(1) (2)
Low-Forbearance Servicer High-Forbearance Servicer
Months in forbearance (as of Nov 2020) 5.73 6.60
Ever exited from forebarance 0.35 0.31
Current Mortgage Balance 196,713.93 172,327.63
Auto Loan Balance 18,242.10 17,765.52
Credit Card Balance 10,841.62 11,487.21
12-mo change CNTY UR (8/20) 6.61 6.36
FHA 0.83 0.82
FICO V5 (updated) 648.79 663.06
LTV at origination 94.22 94.77
Loan age (year) 3.89 5.60
N. Obs. 140,001 237,422
14
Table A.9: Pre-CARES Act loan performance. Relationship between high-forbearance
servicer dummy and various measures of loan delinquency in the period prior to the
passage of the CARES Act. eMBS data from December 2019 and January 2020 are used
for the estimates in table (a), and the matched eMBS-CRISM data from December 2019
and January 2020 are used for the estimates in tables (b), (c), and (d). Dependent variable
for tables (a) and (b) is the dummy variable for transitioning from current to 30+ days
delinquent for the mortgage. Dependent variables for tables (c) and (d) are whether a
borrower has a delinquent credit card and auto loan account, respectively. eMBS controls
include the dummy for FHA loans, loan size, dummy for first-time homebuyer, LTV,
credit score, DTI, and dummy for purchase loans. CRISM controls include updated credit
scores and a borrower’s age. Standard errors clustered at the servicer level.
(a) New 30-day mortgage delinquencies (eMBS only)
(1) (2) (3) (4)
High-forbearance servicer -0.0027∗∗ -0.0016∗∗∗ -0.0015∗∗∗ -0.0016∗∗∗
(0.0013) (0.0005) (0.0005) (0.0005)
eMBS controls Y Y Y
State FE Y
Orig Year-Month FE Y
FHA x State x Orig Year-Month FE Y
Nonbank x FHA x State x Orig Year-Month FE Y
Sample mean 0.013 0.013 0.013 0.013
N. Obs. 22,010,182 20,180,908 20,180,907 20,180,906
(b) New 30-day mortgage delinquencies (eMBS-CRISM
match)
(1) (2) (3) (4)
High-forbearance servicer -0.0027 -0.0009 -0.0005 -0.0005
(0.0029) (0.0019) (0.0017) (0.0016)
eMBS controls Y Y Y
CRISM controls Y Y
Zipcode FE Y Y
Orig Year-Month FE Y Y
FHA x Zipcode x Orig Year-Month FE Y
Sample mean 0.014 0.014 0.014 0.014
N. Obs. 6,117,275 5,748,527 5,741,020 5,732,113
(c) Credit card delinquencies (eMBS-CRISM match)
(1) (2) (3) (4)
High-forbearance servicer -0.0077 -0.0018 0.0023∗ 0.0025∗∗
(0.0082) (0.0029) (0.0013) (0.0010)
eMBS controls Y Y Y
CRISM controls Y Y
Zipcode FE Y Y
Orig Year-Month FE Y Y
FHA x Zipcode x Orig Year-Month FE Y
Sample mean 0.100 0.100 0.100 0.100
N. Obs. 6,136,120 5,748,577 5,741,070 5,732,164
(d) Auto loan delinquencies (eMBS-CRISM match)
(1) (2) (3) (4)
High-forbearance servicer -0.0056 -0.0024∗∗ -0.0011∗∗ -0.0009
(0.0034) (0.0011) (0.0005) (0.0005)
eMBS controls Y Y Y
CRISM controls Y Y
Zipcode FE Y Y
Orig Year-Month FE 16 Y Y
FHA x Zipcode x Orig Year-Month FE Y
Sample mean 0.032 0.032 0.032 0.032
N. Obs. 6,136,120 5,748,577 5,741,070 5,732,164
H Additional non-mortgage results
Table A.10: Non-mortgage results. Estimates of the average effect of assignment to a
high-forbearance servicer on various nonmortgage outcomes measured in CRISM. Es-
timates reported in columns (1), (3) and (5) are the average coefficient on the high-
forbearance-servicer × time dummies (estimates of β t from equation 2) over three phases
of the pandemic: a pre-pandemic period (October 2019-February 2020); early pandemic
(April-July 2020) and later pandemic (August-November 2020), along with the associated
standard error of each mean. For context, columns (2), (4) and (6) report the uncondi-
tional mean of the dependent variable at low-forbearance servicers during the period
referenced. For outcome variables related to auto loans, we report ”NA” during the early
pandemic period because the CRISM data on auto loans for the period contains signifi-
cant reporting error. Standard errors are clustered at the servicer level.
Pre-pandemic Pandemic
2019:m10-2020:m2 2020:m4 to 2020:m7 2020:m8 to 2020:m11
(1) (2) (3) (4) (5) (6)
Coeff. Mean Coeff. Mean Coeff. Mean
Log of auto loan balance 0.000061 6.304 NA 6.055 0.002225 6.165
(0.003665) (0.006727)
Log of other consumer loan balance 0.001 3.476 0.000 3.389 0.004 3.360
(0.003) (0.002) (0.005)
Transition to delinquency (credit card) -0.00001 0.012 0.00014 0.008 0.00014 0.007
(0.00010) (0.00022) (0.00028)
Transition to delinquency (auto loan) -0.000144∗ 0.004 NA 0.004 0.000026 0.004
(0.000077) (0.000062)
Transition to delinquency (other consumer loan) -0.00003 0.004 -0.00001 0.003 0.00001 0.003
(0.00006) (0.00009) (0.00008)
Mortgage prepayment 0.0003 0.013 0.0004 0.019 -0.0001 0.023
(0.0004) (0.0005) (0.0009)
Auto loan origination 0.000391 0.018 NA 0.030 -0.000061 0.023
(0.000344) (0.000308)
Bankruptcy 0.000047 0.004 0.000030 0.004 -0.000047 0.004
(0.000075) (0.000039) (0.000105)
17
File and source
- File
- nyfed-sr1035-forbearance.pdf
- Size
- 1,878,322 bytes
- SHA-256
- be8f339879cb4f3b1a2c8598cd77881f5fbd542ef8ea17233f70c5c4902930bb
- Our copy
- nyfed-sr1035-forbearance.pdf
- Original
- www.newyorkfed.org