Journal of Financial Economics 144 (2022) 590–610

Contents lists available at ScienceDirect

Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec

Financially constrained mortgage servicersR

Darren J. Aiello
Marriott School of Business, Brigham Young University, 683 TNRB, Provo, UT 84602, USA

a r t i c l e i n f o a b s t r a c t

Article history: Financially constrained mortgage servicers destroyed substantial MBS investor value during
Received 27 November 2019 the financial crisis through their management of delinquent mortgages. Servicers advance
Revised 12 May 2021
to investors monthly payments missed by borrowers. In order to minimize this obligation
Accepted 12 May 2021
to extend financing to distressed borrowers, constrained servicers aggressively pursued
Available online 5 October 2021
foreclosures and modifications at the expense of investors, borrowers, and future mortgage
JEL classification: performance. When agency frictions between the servicer and the investor are higher, the
G21 servicer’s financial constraints matter more. IV regressions suggest that, on average per
G23 defaulted loan, servicers’ financial constraints are responsible for 20% of the total investor
value reduction during the financial crisis.
Keywords:
Mortgage servicing © 2021 Elsevier B.V. All rights reserved.
Securitization
Real estate
Financial constraints

1. Introduction of important financial institutions (Mian and Sufi, 2009;

The 20 07–20 09 financial crisis demonstrated the ex-
tent to which mortgage instruments (and hence, mort-
gage default risk) were concentrated on the balance sheets
R
I am indebted to my advisors Mark Garmaise, Andrea Eisfeldt, Stuart
Gabriel, Barney Hartman-Glaser, and William Mann for their advice and
guidance. Additionally, an anonymous referee as well as my discussants—
Gonzalo Maturana, Doug McManus, Jack Liebersohn, and Richard Green—
have made valuable and essential contributions to this paper. For ex-
cellent research assistance, I thank Paige Nelson. I also thank Antonio
Bernardo, Bruce Carlin, Alex Fabisiak, Chady Gemayel, Valentin Haddad,
Ryan Huddleston, Mahyar Kargar, Jason Kotter, Grant McQueen, Todd Mit-
ton, Tyler Muir, Taylor Nadauld, Christopher Palmer, Nimesh Patel, Ryan
Pratt, Jeb Robinson, Jacob Sagi, Tyler Shumway, Clinton Tepper, Ivo Welch,
Tim Witten, and Geoff Zheng, as well as seminar and conference partici-
pants at UCLA Anderson’s Finance Brown Bag and Student Seminar series,
the 2018 UCI-UCLA-USC Urban Research Symposium, Brigham Young Uni-
versity, Notre Dame University, University of Oregon, Philadelphia Federal
Reserve Bank, Federal Reserve Board, University of North Carolina-Chapel
Hill, and University of North Carolina-Charlotte for helpful comments and
suggestions. All errors are my own. I also thank the UCLA Ziman Center’s
Rosalinde and Arthur Gilbert Program in Real Estate, Finance, and Urban
Economics for their generous funding.
E-mail address: d.a@byu.edu
Diamond and Rajan, 2009). These institutions, investors in
mortgage-backed securities (MBS), delegate the manage-
ment of mortgage loans to intermediaries called mortgage
servicers. This paper is the first to study the extent of the
incentive misalignment between servicers and investors,
how that agency conflict gets exacerbated in the presence
of financial constraints, and the substantial consequences
of this combination during the financial crisis. A financially
constrained servicer both modifies and forecloses more
aggressively than is optimal from the investor’s perspec-
tive. Additionally, the extent to which a servicer’s financial
constraints matter varies directly with the severity of the
agency friction between the servicer and the investor. By
describing this undocumented and economically substan-
tial friction, I demonstrate the outsized role mortgage ser-
vicers play in the economics of this market.
If a delinquent mortgage borrower cannot self cure,
refinance the loan, or sell the property, the servicer is
authorized to intervene by either modifying the terms of
the note or foreclosing on the property on behalf of the
investor. Servicers are required to advance the monthly
principal and interest from borrowers to investors when

https://doi.org/10.1016/j.jfineco.2021.09.026
0304-405X/© 2021 Elsevier B.V. All rights reserved.
D.J. Aiello
borrowers do not make their payment. These advances
incentivize the servicer to take costly modification or
foreclosure action when faced with a delinquent mort-
gage. While these default interventions are designed to
benefit the investor, they are also directly beneficial to the
servicer. Upon completion of an intervention (the signing
of a modification agreement or the sale of a foreclosed
property), a servicer avoids advancing more delinquent
payments and recovers all previous advances on that loan,
regardless of the proceeds or performance of the spe-
cific loan. By providing short-run financing to delinquent
borrowers in the form of these advances, servicers give
borrowers the opportunity to recover on their own. How-
ever, servicers are either less willing or less able to take
this option when they are financially constrained. A finan-
cially constrained servicer intervenes in delinquent loans
at a higher rate—producing foreclosures and modifications
in excess of the investor value maximizing level.
While securitization offers benefits to investors over
holding loans in portfolio, it also introduces an agency con-
flict between the servicer and the investor (Wong, 2018).
As an institutional feature, servicing advances exist to ease
these agency conflicts, but an underappreciated—and pre-
viously undocumented—fact is that advances make the bal-
ance sheets of the servicers important. It is exactly when
the agency frictions between the servicer and the investor
are highest that the servicer’s financial constraints matter
to their decision making.
But how important is this effect? For every standard
deviation increase in financial constrainedness, servicers
are nine percentage points more likely to foreclose after
a borrower’s first episode of delinquency. Because mov-
ing quickly to modify a defaulted loan will also benefit a
constrained servicer (offering a much quicker path to re-
covery of advances), for every standard deviation increase
in financial constrainedness, servicers are five percentage
points more likely to modify a loan as well. What is not
clear, however, is whether these interventions are produc-
tive. I show that, at a minimum, these actions taken by
constrained servicers are bad for investors. They are also
bad for borrowers: I exploit quasi-random assignment of
financial constraints to demonstrate that these are foreclo-
sures and modifications where, in the absence of the effect
of servicer financial constraints, the borrower would have
self cured, refinanced, or sold the property instead.
The severity of the agency conflict over whether or not
to intervene in a defaulted loan varies directly with how
uncertain the servicer and investor are over what the op-
timal action to take is, directly impacting the cost of mon-
itoring borne by the investor. To test whether servicers
act differently according to the severity of agency frictions,
I exploit variation in the costs investors bear to moni-
tor servicer actions and show that there is no difference
in the decision making between constrained and uncon-
strained servicers for low monitoring cost (low agency fric-
tion) loans. Alternatively, when intervention may or may
not be optimal and the investor must monitor the servicer,
the high agency friction environment leads the servicer to
act differently depending on its financial constrainedness.
To determine that the distortion of a servicer’s de-
cisions, in the presence of financial constraints, deterio-
591
Journal of Financial Economics 144 (2022) 590–610
rates rather than improves incentive alignment with the
investor, I construct a measure of the net present value
(NPV) of the investor’s loan cash flows. Financially con-
strained servicers’ aggressive and premature intervention
in defaulted loans, and their actions conditional on in-
tervention, destroy a large amount of investor value. On
average, a servicer’s financial constraints caused $11,896
(20.21%) of the $58,866 average total investor value de-
stroyed per defaulted loan in the US during the financial
crisis.
Finally, I explore the manner in which a servicer’s ca-
pacity constraints (e.g., staffing limitations in the face of
large volumes of delinquent loans) interact with their fi-
nancial constraints. I demonstrate that a capacity con-
strained servicer has a severely curtailed ability to respond
to exogenous changes in their financial constraints by in-
creasing the rate at which they modify and foreclose on
the delinquent loans in their portfolio.
I utilize a largely unexplored dimension of mortgage
performance data. I start with a rich dataset that follows
the performance of approximately 90% of the mortgage
loans sold into non-agency residential mortgage-backed
securities (RMBS) transactions pre-crisis. These non-agency
RMBS loans account for the bulk of the delinquencies,
modifications, and foreclosures that occur during the fi-
nancial crisis. To investigate the impact of servicer behav-
ior and incentives on default outcomes without contamina-
tion from prior servicer interventions, I condition on loans
that defaulted and observe the outcome related to the first
stretch of borrower delinquency.
In this paper, the amount of advances a mortgage ser-
vicer makes for missed monthly payments is used as an
indicator of financial constraints. An alternate measure of
a servicer’s financial constrainedness, the credit default
swap spread for a servicer’s bond issuance, is discussed in
Section 4.3. Measuring the causal impact of servicing ad-
vances on servicer decisions and loan outcomes is prob-
lematic because advance levels are endogenous to servicer
actions. I construct plausibly exogenous variation in ser-
vicer financial constraints by instrumenting servicing ad-
vances with a servicer’s unique exposure to the timing
of housing price returns in geographies far removed from
the focal loan (Granja et al., 2017). A servicing-portfolio-
balance-weighted average of zip-level housing price re-
turns, excluding the core-based statistical area (CBSA)
where the focal loan resides, directly impacts the volume
of advances a servicer makes (instrument relevance). How-
ever, the servicer portfolio’s housing price return, in ge-
ographies separate from the focal loan, does not influence
loss mitigation decisions or outcomes except through its
direct effect on the advances made by the servicer (ex-
clusion restriction). Particular concerns related to this ex-
clusion restriction, such as unobservable borrower quality,
dynamic servicer learning, the existence of operational ca-
pacity constraints, and servicer ownership of MBS tranches,
are addressed in detail in Section 4. Additionally, I in-
clude the housing price return for the zip code where
the loan resides as a control, as well as fixed effects at
the servicer level and for an interaction between default
year, zip code, and borrower credit quality (Prime, Alt-A,
and Subprime). I then compare loans of the same credit
D.J. Aiello
quality that defaulted in the same zip code and calendar
year.
Whereas financial constraints and agency costs are
well studied, the interaction between them is not.
Whited (1992) and Eisfeldt and Rampini (2007) pro-
vide models of financial constraint-driven myopia that
are accurate representations of the mortgage servicers’
behavior in this study. A servicer’s financial constraints
are examined in Kim et al. (2018). Most relevant to this
study, Chodorow-Reich and Falato (2021) demonstrate that
commercial banks were less likely to grant waivers to cor-
porations in violation of loan covenants and consequently
increased renegotiation and forced accelerated debt re-
payment when they were in worse health. To the best of
my knowledge this paper is the first to demonstrate the
relationship between a servicer’s financial constraints and
the realized agency costs borne by the investor.
While mortgage securitization is rife with agency prob-
lems at all levels (Griffin et al., 2019; Kruger and Matu-
rana, 2020), this paper also addresses the important de-
cision by a servicer to either intervene or provide an op-
portunity for a borrower to solve the problem themselves.
Agarwal et al. (2017) find that servicers of different quality
have varied appetites and ability to perform modifications.
Various influences on the decision to either modify or fore-
close have been studied, with the impact of securitization
being the most extensively studied (Piskorski et al., 2010;
Agarwal et al., 2011; Adelino et al., 2013; 2014; Kruger,
2017). This paper examines how a servicer’s financial con-
straints, conditional on all these other effects, influence the
agency frictions related to managing a delinquent loan.
The remainder of the paper proceeds as follows.
Section 2 introduces the data and Section 3 provides in-
dustry context. Section 4 describes and discusses the iden-
tification strategy and Section 5 examines my central find-
ings. Section 6 concludes.
2. Data
The data describe 6,940,990 U.S. residential mortgage
loans securitized into non-agency MBS that first went at
least one-month delinquent (defaulted) in 2013 or before.
The data are sourced from BlackBox, which covers about
90% of pre-crisis privately securitized residential mortgage
loans.
These loans have a total of 351 million monthly loan
observations. These monthly records are collapsed to the
loan level for all loans that went delinquent, entered
bankruptcy, or were ever modified, foreclosed, or liqui-
dated in a manner other than a borrower pay off in full.
The loans in the sample are tracked beginning with their
final timely payment. Resolution of a particular episode
of borrower delinquency occurs through an action by
either the borrower or the servicer. Borrowers either self
cure when they make at least three consecutive current
payments with no servicer intervention, pay off the loan in
full, or enter bankruptcy. The servicer ends an episode of
delinquency by either modifying, foreclosing, repurchasing,
or liquidating the loan. Focus is restricted to the first
delinquency.
592
Journal of Financial Economics 144 (2022) 590–610
Restricting attention to defaulted loans with valid loan
and borrower characteristic variables and servicer level
advance and instrument values leaves 6,940,990 unique
loans. Column (1) of Panels A and B of Table 1 contains
the mean value for the various loan-level measures within
the overall BlackBox population (with valid covariate and
servicer values, see Section 4) for comparison to the re-
mainder of the table, which deals with only loans in sam-
ple. The 78.2% incidence of delinquency within this popu-
lation is due to the low-bar definition of delinquency uti-
lized in this study (ever at least 30-days past due), the
non-conforming nature of the assets, and the crisis setting.
The sample loans are selected primarily by condition-
ing on delinquency, as reflected in the comparison be-
tween the sample and BlackBox population measures in
Panel A of Table 1. Defaulted loans are larger, were orig-
inated later, had lower credit scores and higher loan-to-
value (LTV) ratios at origination, and were more likely to
be adjustable rate. Panel B addresses the loss mitigation re-
sults that occurred in sample. Sample proportion statistics
are mutually exclusive1 and collectively exhaustive. They
report the first event to occur post-default.2 These out-
comes are either borrower driven (self cure or pay off in
full), servicer driven (modification, foreclosure, or liquida-
tion), or due to something largely outside of the servicer’s
control (bankruptcy or repurchase). In addition to outcome
proportions, the average (and standard deviation of the)
months to completion and Investor NPV of the action (see
Section 4.4) are also reported. Panel C of Table 1 reports
the average values of loan and borrower characteristics
for various outcomes conditional on delinquency. Servicer
names are cleaned to remove differing entity name permu-
tations and to collapse subsidiaries.
Geographic indicators are found at the zip code level
within the BlackBox dataset. US Postal Service, Census Bu-
reau, and US Department of Agriculture data are used to
match zip codes to CBSA, commuting zones, and states.
CBSA consist of a core urban area and the surrounding
territory with “a high degree of social and economic in-
tegration with the core” (Census Bureau, 2016). CBSA are
either metropolitan statistical areas (MSA) or smaller mi-
cropolitan statistical areas (μSA). Commuting zones are de-
rived from the US Department of Agriculture’s Economic
Research Service and fulfill largely the same role as CBSAs,
but are generally much larger in geographic area. Commut-
ing zones and states cover the entire geography of the US.
Rural counties that are not included in a CBSA are consid-
ered as separate “CBSA-like” geography observations.
3. Institutional background
When a mortgage loan is originated, two separable as-
sets are created: the note itself and a mortgage servicing
1
A small number of delinquencies in the data resulted in a simultane-
ous modification and foreclosure. All results are robust to excluding these
loans, classifying them as one or the other, or both.
2
Because this study focuses on the first delinquency, these sample
proportions are not representative of the ultimate outcomes. Self cures
or modifications that ultimately redefaulted could have had a second
episode of delinquency (or a third, etc.) that ended in any of these possi-
ble outcomes.
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 1
Summary statistics.
This table reports summary statistics relating to the loan population utilized in this paper. Column (1) of Panels A and B report results
relating to the population of BlackBox loans with valid covariate and servicer values. The remainder of the columns relate to the loans
that experienced at least one qualifying delinquent episode. Panel A reports summary statistic data. Panel B reports performance statistics
at the loss mitigation outcome level associated with the first delinquent episode. The average months to completion are the number of
months between first default and the month of the borrower’s third consecutive current payment, the pay off date, modification effective
date, foreclosure sale date, liquidation date, bankruptcy petition filing date, or the date of repurchase as applicable. Panel C replicates
Panel A statistics grouping by loss mitigation outcomes.

Panel A: summary of loan characteristics

BlackBox Sample
Mean Average Median Std. Dev. 10th% 90th%
(1) (2) (3) (4) (5) (6)

Original Loan Amount ($) 248,454 248,798 188,800 221,908 50,250 504,400
Loan Amount at Default ($) 243,996 185,311 219,018 48,386 499,861
Note Date 3/5/2005 5/23/2005 9/1/2005 19mos, 9 days 6/1/2003 12/1/2006
Default Date 8/13/2007 7/1/2007 25mos, 3 days 3/1/2005 3/1/2010
Original Note Rate 7.32% 7.32% 7.13% 2.59% 4.75% 10.65%
Credit Score 674.37 668.84 670.00 68.64 576.00 761.00
LTV 71.19 72.02 80.00 22.44 26.70 95.00
Adjustable Rate 0.60 0.63
Owner Occupied 0.82 0.82
Purchases 0.41 0.41
Single Family 0.71 0.70
Full Documentation 0.31 0.31

Panel B: summary of loan outcomes

BlackBox Sample Average Months SD of Months Average SD of

Proportion Proportion to Completion to Completion NPV NPV
(1) (2) (3) (4) (5) (6)

Defaulted 0.782 184,385 202,589

Self Cure 0.263 7.07 5.09 153,749 179,134
Paid-In-Full 0.396 2.37 1.87 251,460 231,944
Modified 0.071 8.31 9.03 161,797 179,350
Foreclosed 0.240 24.51 19.38 122,774 149,536
Liquidation 0.002 12.03 13.01 45,737 112,863
Bankruptcy 0.028 120,481 133,593
Repurchased 0.000 208,398 211,931

Panel C: summary of loan characteristics by outcome

Self Cure Paid-In-Full Modified Foreclosed Liquidated

(1) (2) (3) (4) (5)

Average Original Loan Amount ($) 220,250 263,701 272,545 255,020 116,405
Average Loan Amount at Default ($) 217,057 253,211 273,511 255,714 114,245
Average Note Date 5/4/2005 11/17/2004 2/6/2006 1/22/2006 1/28/2009
Average Default Date 6/18/2007 2/9/2007 10/23/2007 3/5/2008 1/24/2009
Average Original Note Rate 7.41% 7.05% 7.24% 7.66% 8.87%
Average Credit Score 656.77 679.32 666.36 665.91 691.10
Average LTV 71.52 69.94 75.14 75.17 39.03
Proportion, Adjustable Rate 0.58 0.66 0.62 0.67 0.57
Proportion, Owner Occupied 0.83 0.84 0.88 0.77 0.76
Proportion, Purchases 0.39 0.39 0.40 0.46 0.44
Proportion, Single Family 0.71 0.71 0.71 0.68 0.61
Proportion, Full Documentation 0.38 0.27 0.37 0.29 0.19

right. The mortgage servicer is the entity that owns the
mortgage servicing right. Servicing a mortgage is the pro-
cess of turning borrower cash flow claims into a uniform
mortgage asset. For securitized loans, the note is placed in
an off-balance sheet special purpose entity of the issuer.3
The servicing right is either sold alongside the note to an-
other party (“servicing released”) or retained by the orig-
3
Alternatively, an entity such as a bank holds the note on its balance
sheet.
inator (“servicing retained”). A servicing right that is sold
can either end up in the hands of the securitization issuer’s
captive primary servicer or a party that intermediated be-
tween the originator and the issuer (these loans are then
primary serviced by an entity unaffiliated with either the
securitization issuer or loan originator). The owner of this
servicing right is referred to as the “servicer.” The servicer
manages all borrower facing activities including payment
collection, managing loan and escrow accounts, and pro-
viding loss mitigation services in the case of default.

593
D.J. Aiello
Loss mitigation activities consist of encouraging delin-
quent borrowers to make their payments, working out re-
payment plans, providing modifications or other loss miti-
gation alternatives, such as deed-in-lieu of foreclosure or
short sales, and managing the foreclosure process. Ser-
vicers undertake these activities in compliance with the
relevant securitization guidelines and have contracted to
act in the best interest of the investor.4 The assumption
that modification is always better than foreclosure, even
from the investor’s perspective, while ubiquitous in the lit-
erature, is not accurate. Maturana (2017) shows this as-
sumption to be true on the level of an average effect but
allows for the self-evident possibility that, locally, it is of-
ten optimal to seize a borrower’s home rather than provide
a hopeless modification.
When a borrower fails to make a monthly mortgage
payment on loans in a non-agency securitization (the fo-
cus of this paper), the servicer is obligated to advance the
principal and interest to the investor out of its own funds.
This creates a receivable on the balance sheet of the ser-
vicer. If the borrower makes their past due payment, the
receivable is cleared on the servicer’s balance sheet. Upon
completion of a modification agreement or the sale of a
foreclosed property, a mortgage servicer has the first claim
on any cash flows from that related loan to recover its ad-
vances. To the extent that these cash flows are insufficient
to reimburse the servicer for its previous advances, the ser-
vicer has first priority claim on all cash flows related to
any other loan in the securitization. If the servicer were to
modify the loan, it capitalizes all outstanding advances into
the investor’s loan balance and the servicer recovers its ad-
vances from the investor by reducing pool cash flows in
the month of modification.5 In the event of a foreclosure,
the servicer recovers advances outstanding from the pro-
ceeds of the sale of the property or, if that is insufficient,
from pool cash flows in the month of foreclosure sale. Ul-
timately, the servicer is always made whole.
Panel B of Table 1 shows that the average months
from delinquency to completion of modification (at which
point a servicer is able to recover 100% of its outstand-
ing advances) is only 8.31 months. Alternatively, the aver-
age foreclosure liquidation occurs 24.51 months after the
borrower’s first delinquency.6 This difference of 493 days
means that, on average, a modification returns a servicer’s
4
Previous studies (Cordell et al., 2008; Mooradian and Pichler, 2017;
Huang and Nadauld, 2019; Herndon, 2021) have looked at agency conflicts
in a mortgage servicer’s contract.
5
Modifications always bring the paid-to-date of a loan to current. This
is accomplished either through a capitalization as discussed, forgiveness
of the borrower’s debt, or a loan workout agreement where partial pay-
ments are made until the borrower is caught up. What matters from the
point of view of the servicer is that once the paid-to-date is brought cur-
rent, the servicing advance receivable is cleared.
6 bulk of delinquent mortgages, is the correct setting in which to evaluate
Both of these statistics are related only to the actions the servicers
took on the borrower’s first delinquency. A foreclosure is measured at the
time the servicer first places a loan into foreclosure. A foreclosure liqui-
dation is when a previously foreclosed property is eventually sold, either
at a foreclosure auction or through the real estate owned (REO) process.
A foreclosure liquidation is distinct from a charge-off or short-sale liqui-
dation, which are generally labeled as unqualified “liquidations” herein.
594
Journal of Financial Economics 144 (2022) 590–610
outstanding advances 2.95 times faster than a foreclosure.7
This leads a financially constrained servicer to prefer mod-
ification over foreclosure even in the presence of sub-
sidiaries that collect fees from the investor during the fore-
closure process.8
3.1. Servicer advances calculation
Previous studies on the financial constraints of interme-
diaries are largely limited to the observations that financial
constraints increased intermediary risk taking during the
savings and loan crisis in the U.S. (Kroszner and Strahan,
1996; Esty, 1997a; 1997b), and the effects that regulatory-
induced financing constraints have on insurance companies
(Lee et al., 1997; Ellul et al., 2011; Koijen and Yogo, 2014;
Merrill et al., 2021).
In this paper, the amount of advances a mortgage ser-
vicer makes for missed monthly payments is used as an
indicator of financial constraints. Advances consist solely of
the principal and interest payment not made by a delin-
quent borrower. Taxes and insurance as well as fees and
other foreclosure and modification-related costs paid by
the servicer are not included. A loan has an outstanding
servicing advance when its paid-to-date is less than the
monthly activity date and its scheduled balance is less
than the actual balance. The loan-level servicing advance
outstanding amount is calculated by summing all monthly
principal and interest (P&I) constants for each paid-to-
date currently outstanding. The loan-month outstanding
advance distribution is right-tail winsorized at the 0.01%
level.
These loan-level servicing advances are summed, for
each month, to the servicer level to calculate Advancest,s ,
representing the outstanding servicing advances in month
t for servicer s.9 Advancest,s is then scaled by the total
portfolio outstanding actual loan balance for servicer s at
time t , Port f ol ioBal ancet,s , to calculate the ratio of the out-
standing advances to the outstanding portfolio balance,
Advancest,s
AdvF ract,s = . (1)
Port f olioBalancet,s
Actual balance scaled servicer-month outstanding servicing
advances, AdvF ract,s , are then right tail winsorized at the
1% level. The principal variable of interest and the main
measure of the financial constrainedness of a servicer, 1-
Year Change in Advances as a Percentage of Loan Portfolio
7
Mortgage modification, particularly after the influence of the crisis,
became so streamlined and uniform that its perception as being more
costly and lengthy as compared to foreclosure is not accurate.
8
This preference that financially constrained servicers have for mod-
ifications over foreclosures is entirely driven by the role that servicer
advances play in the incentives of the servicer. Chomsisengphet and
Lu (2015) present evidence that, in settings where the burden of servicing
advances on a servicer is lessened or non-existent (such as agency securi-
tizations and bank-owned portfolios), foreclosures are preferred to mod-
ifications. However, the non-agency securitization market, containing the
the impact of servicer financial constraints on loss mitigation outcomes.
9
The servicer-loan match is static in the data and represents the ser-
vicer of the loan at the time of securitization. Changes in servicer are
not reported in any available datasets relating to this loan population.
Mayock and Shi (2018) provide evidence that servicing transfer activity
was largely only relevant beginning in 2011.
D.J. Aiello
Balance, is calculated by taking the difference in levels be-
tween the scaled value AdvF ract,s in the month preceding
loan default and the scaled value one year prior,
AdvF ract,s = AdvF ract−1,s − AdvF ract−12,s , (2)
and is standardized separately for each regression
(Cookson and Niessner, 2020).
4. Empirical strategy
My principal empirical specification involves two types
of outcome variables: a servicer’s action decisions regard-
ing a delinquent loan and loan outcomes that occur fol-
lowing a servicer’s decision. These outcomes are examined
within the framework of two dimensions of fixed effects.
First, a fixed effect at the servicer level removes the aver-
age effect of the servicer, held constant over time. Second,
an interacted fixed effect dimension restricts attention to
loans that first defaulted in the same calendar year, are lo-
cated in the same zip code, and are of the same general
credit quality (Prime, SubPrime, or Alt-A). Additionally, all
relevant loan and borrower characteristics are included as
controls. Included within this framework is a measure of
a servicer’s financial constrainedness that is an economi-
cally and statistically significant predictor of both servicer
actions and loan outcomes. From the standpoint of the in-
vestor, the optimal servicer action should not be influenced
by the balance sheet of the servicer. The fact that it is,
however, is indicative of an agency friction relating to the
conflict between the servicer’s objective function and the
obligation to maximize investor value. A servicer’s financial
constraints are measured through the amount of borrower
monthly payments the servicer has been required to ad-
vance in the immediate past, scaled by the servicer’s port-
folio size (see Section 3.1).
This paper estimates the causal impact of a servicer’s
financial constraints on a servicer’s actions. However, the
actions that a servicer takes regarding its delinquent loan
portfolio are going to have a direct impact on the amount
of servicing advances that it has to make. Advances are en-
dogenous to servicer actions. Additionally, changes in ad-
vance levels can be informative about the quality of the
servicer or can inform the servicer about the quality of its
assets, consequently changing our interpretation of these
behaviors or prompting the servicer to adjust its actions
accordingly. Finally, the level of servicer advances, loan
outcomes, optimal loss mitigation strategies, and the bor-
rower’s ability to refinance (Palmer, 2015) can all be in-
fluenced by macroeconomic or regulatory conditions. For
these reasons, I instrument outstanding advances with a
servicer portfolio’s unique exposure to housing price re-
turns (Griffin et al., 2020) over the 12-month period pre-
ceding the loan default. The calculation of this instru-
ment is detailed in Section 4.1. Variation in this instru-
ment stems from local and regional variation in the foot-
print of servicer portfolios (Loutskina and Strahan, 2011).
Housing price returns are measured at a precise geographic
level—the zip code of the property—and even national ser-
vicers have important variations in their lending behavior
and experience distinct competitive environments at these
(and even smaller) geographic levels (Aiello et al., 2017).
595
Journal of Financial Economics 144 (2022) 590–610
This variation translates to important differences in the ge-
ographic diversity of their servicing portfolios. This instru-
ment then measures the extent to which a particular ser-
vicer is exposed to exogenous housing price shocks. This
instrument is relevant because housing price declines in a
particular geographic region are strongly associated with
increased delinquencies, which will in turn cause a ser-
vicer to make additional advances. These first-stage results
are reported in Section 4.3 and demonstrate that the in-
strument is both predictive of servicer advances and en-
compasses a meaningful level of the variation in those ad-
vances.
For this instrument to satisfy the exclusion restriction,
housing prices cannot directly impact a servicer’s actions
or loan outcomes. In the case of my instrument, I exclude
the portion of a servicer’s portfolio in the CBSA immedi-
ately surrounding the focal loan and include as a control
the relevant zip code level housing price return for the
12-month period preceding loan default. Thus my exclu-
sion restriction—as satisfied by assumption—is that, condi-
tional on the local zip code housing price return, the ser-
vicer portfolio housing price return (in geographies sepa-
rate from the loan) does not influence loss mitigation de-
cisions and outcomes except through their effect on ser-
vicing advances outstanding and the balance sheet of the
servicer. Consequently, the manner in which a servicer’s
unique exposure to far away housing prices—conditional
on local housing prices—impairs its balance sheet is what
provides me with plausibly exogenous variation in a ser-
vicer’s financial constraints. Concerns related to the exclu-
sion restriction can be easily grouped into four broad cat-
egories: unobservable borrower characteristics, differential
servicer learning, servicer operational constraints, and ser-
vicers as investors.
First, of particular concern is the possibility that unob-
servable borrower characteristics are correlated with the
measure of financial constraints, perhaps through system-
atic correlation between a servicer’s geographic footprint
and servicer quality or because a low-quality servicer se-
lected into a low-quality portfolio. Servicer-level fixed ef-
fects remove any time-invariant average effects while the
second dimension of fixed effects (interacted loan default
year, property zip code, and borrower credit quality cate-
gory) and borrower and loan covariates ensure that I am
comparing very similar loans on a large number of impor-
tant dimensions. That no missing, important, unobservable
borrower quality characteristics exist after these fixed ef-
fects and controls remains an identifying assumption. I fur-
ther address concerns that the unobservable quality of the
borrower may vary systemtically with the level of servicer
financial constraints in Section 5.7. While time-invariant
servicer characteristics are removed with a fixed effect,
the type or size of the servicer may still be important for
the interpretation of the results. Results remain largely the
same when the sample is restricted to only the largest
servicers, as well as to servicers that are subsidiaries of
large financial institutions rather than specialized, mort-
gage only, entities.
Second, a concern remains that servicers with earlier
exposure to loan defaults may behave differently than ser-
vicers that did not have similar exposure, not because
D.J. Aiello
of lasting balance sheet impacts, but rather because they
have learned (or learned differently) about the behavior of
borrowers. Murfin (2012) demonstrates that, in the com-
mercial loan market, banks that have recently experienced
higher numbers of defaults in their loan portfolio are more
likely to write stricter contracts with future borrowers. The
dataset I utilize is updated on a monthly basis and is
widely available to all market participants in real time. A
servicer’s ability for marginal learning from its direct expe-
rience with defaults in a particular geography is not likely
to be relevant conditional on the learning it can glean from
industry-wide and nation-wide defaults, which are con-
trolled for with a year fixed effect. To the extent a concern
remains, I include a measure of the change in the count of
delinquent loans in a particular servicer’s portfolio over the
12-month period immediately preceding the loan’s default
as a percentage of all loans in the portfolio. Any learning
on the part of the servicer is likely to be a function of the
volume of delinquencies with which it has experience, not
the dollar amount of advances made on those delinquen-
cies.
Third, the possible existence of a servicer operational
constraint could confound the causal interpretation of this
instrument. As a servicer’s portfolio realizes worse hous-
ing price returns, their advance volume increases, and they
suffer from an increase in delinquent mortgages. As oper-
ational capacity constraints are reached, the servicer may
behave differently, adversely impacting the investor’s inter-
ests. I address this through the inclusion of the change in
the count of portfolio delinquencies over the year prior to
the focal loan’s delinquency (as a percentage of all loans
in the servicer’s portfolio) as a measure of capacity con-
straints. Operational constraints bind on the basis of loan
count volume not advance balance. While contamination
of my mechanism from operational constraints is relevant,
the direction of my main finding—that constrained ser-
vicers are more likely to take a loss mitigation action in
order to reduce their constraint—rules out the importance
of this effect.
Finally, some servicers are also among the many in-
vestors in the securitizations it services. When this is the
case however, the servicer most likely owns an equity (or
subordinate) tranche. Consequently, the servicer is more
likely to make decisions that support the short-term cash
flow of the securitization at the expense of the bulk of the
investors (Huang and Nadauld, 2019). In order for these
“servicer-investors” to represent a violation of the exclu-
sion restriction in my study, a servicer’s equity ownership
level must be negatively correlated with the housing price
returns it will later experience—a servicer would have to
hold more equity in worse mortgage pools.
4.1. Instrument construction
Log housing price returns for zip code z in month t,
ZipHP Rett,z , are gathered from the Zillow Home Value In-
dex (All Homes), hereafter ZHVI. The index uses a hedo-
nic regression framework that updates the value of every
home within a particular region in response to every trans-
action (Hartman-Glaser and Mann, 2016).
596
Journal of Financial Economics 144 (2022) 590–610
The zip log housing price returns, ZipHP Rett,z , are
weighted by the outstanding scheduled balance of the
servicer’s total mortgage portfolio available in BlackBox,
Svcr PortSchedBalt,z,s(i ) . This represents the unique exposure
of a particular servicer s(i ) to housing price returns in
month t for a given CBSA, g(i ), and is calculated using the
following:
OneMonthSvcrPortHP Rt,g(i ),s(i )

/ g(i ) ZipHP Rett,z · Svcr P ortSchedBalt,z,s(i )
z∈
=  , (3)
/ g(i ) Svcr P ortSchedBalt,z,s(i )
z∈
where OneMonthSvcrPortHP Rt,g(i ),s(i ) is the one month ser-
vicer portfolio balance weighted log housing price return
for all zip codes in month t for servicer s(i ) outside of
the CBSA, g(i ), associated with focal loan i. The values
of OneMonthSvcrPortHP Rt,g(i ),s(i ) over the 12-month period
preceding the month of default, t, are then summed, as:
12
Svcr PortHP Rt,g(i ),s(i ) = OneMonthSvcrPortHP Rt−n,g(i ),s(i ) ,
n=1
(4)
where Svcr PortHP Rt,g(i ),s(i ) is the instrument used through-
out the principal empirical specifications, 1-Year Servicer
Portfolio Housing Price Return (Excluding Loan CBSA).
4.2. Empirical approach
The empirical strategy is a two-stage least squares
regression approach. In the first stage, I regress
AdvF ract,s(i) on SvcrPortHPRt,g(i),s(i) . The regressions
include a set of loan and borrower characteristic controls
and fixed effects at both the servicer and zip-year-credit
category level. For loan i, defaulting in month t, located in
zip code z(i ) of loan i, CBSA g(i ) of loan i, and serviced by
the servicer s(i ) of loan i, the first-stage regression is:
AdvF ract,s(i) = βF S · SvcrPortH PRt,g(i),s(i) + γ1 · ZipHPRett,z(i)
+ γ2 · DQCountt,s(i ) + γ3 · controlsi + i . (5)
The controls include the housing price return of zip
code z(i ) for the 12-month period preceding month t,
ZipHP Rett,z(i ) , and the change in the volume of delinquen-
cies at servicer s(i ) over the 12-month period preceding
month t, DQCountt,s(i ) . Additionally, servicer-level fixed
effects and a threefold interaction of fixed effects for the
year and zip code of default and the broad credit cate-
gory (Prime, Alt-A, and SubPrime) of the loan are included
utilizing the estimator of Correia (2016). Finally, a vector
of loan and borrower controls includes FICO, original note
rate, original loan-to-value, original loan balance, lien po-
sition, percentage of mortgage insurance coverage, as well
as indicators for whether the loan was an ARM, owner oc-
cupied, a purchase loan, for a single-family property, fully
income documented, interest only, balloon, or had mort-
gage insurance.
This first-stage regression yields the variation in servic-
ing advances attributable to the housing price movements
in geographies removed from the focal loan. The second-
stage estimates the causal impact of financial constraints
D.J. Aiello
on servicer behavior and loan outcomes,

Yi = βIV · Ad vF ract,s(i) + γ4 · ZipHPRett,z(i)
+ γ5 · DQCountt,s(i ) + γ6 · controlsi + νi , (6)
where Ad vF ract,s(i) is the fitted value from Eq. (5) repre-
senting the variation in servicing advances attributable to
housing price movements in geographies removed from fo-
cal loan i. The coefficient βIV estimates the causal impact
of AdvF ract,s(i ) on Yi .
To estimate the causal impact of financial constraints
on servicer behavior, and to test whether there is com-
plementarity between that relationship and the level of
agency friction between the servicer and the investor,
I implement a two-stage least squares setting analo-
gous to Eq. (6) wherein I instrument AdvF ract,s(i ) using
Svcr PortHP Rt,g(i ),s(i ) and the interaction of AdvF ract,s(i )
and an indicator for whether the focal loan was “high qual-
ity,” IHQ , with an interaction between Svcr PortHP Rt,g(i ),s(i )
and the indicator IHQ .
4.3. First-stage results
Column (1) of Table 2 presents a first-stage estimate
of Eq. (5) of −8.213. The results imply that a reduction
in housing prices at the servicer portfolio level is associ-
ated with an increase in servicing advances over the same
time period, which is precisely what is expected from the
economic motivation underlying this choice of a first-stage
setup.
The first-stage result in column (1) of Table 2 is at
the defaulted loan observation level. The dependent vari-
able and the instrument, which are changes in advances
and servicer portfolio housing price return respectively, are
better thought of as servicer-level variables. Accordingly, I
also test the first stage at a servicer-month level to en-
sure that the relationship between advances and housing
price returns is general. For the servicer-month specifica-
tion, a generalized instrument with no geographic hold-
out is used. Additionally, loan and borrower characteristic
controls are excluded and fixed effects are included sep-
arately at the servicer and year level. Column (2) shows
that the estimated first-stage coefficient with this setup
is −3.385. The difference in the point estimates between
columns (1) and (2) of Table 2 stem largely from differ-
ential weighting on servicers with large numbers of delin-
quent loans in a particular month. I utilize the loan-level
first-stage estimates reported in column (1) of Table 2 in
a two-stage least squares framework primarily because
it allows for the exclusion of housing price returns in
geographies surrounding the focal loan, as well as be-
cause its larger magnitude results in more conservative
IV estimates.
As an alternative measure of servicer financial con-
strainedness, I also utilize, for ten of the largest servicers,
the 5-year credit default swap spread rate for the subor-
dinate bond issuances of the servicer from Markit. Bond
issuances by the immediate (bank level) parent company
are chosen over those of the parent bank holding com-
pany where available. Columns (3) and (4) of Table 2 re-
port the first-stage results obtained from regressing the
597
Journal of Financial Economics 144 (2022) 590–610
credit default swap spread on the instrument at the loan-
month level and servicer-month level, respectively. Col-
umn (3) shows a coefficient magnitude approximately 34%
greater than that of the servicer advance measure of finan-
cial constraints (implying IV coefficient estimates of about
25% less than those reported throughout the paper). While
the CDS spread measure of servicer financial constrained-
ness works, the servicer advances measure is preferable
due to its availability for a more diverse cross-section of
servicers, as well as its closer conceptual proximity to the
salient financial constraint faced by a manager in charge of
the decision process related to delinquent loans.
4.4. NPV calculation
This paper directly measures the magnitude of a
realized agency cost borne by an investor in MBS.
Ang et al. (20 0 0) and Singh and Davidson (2003) have
done this within the context of corporations and,
while not directly related to an agency problem,
Ang et al. (2017) demonstrate that the short-term bud-
getary constraints of a municipality can lead it to take
actions that destroy a significant amount of value for itself.
While similar in spirit to the measure of investor losses
found in Maturana (2017), who uses realized liquidation
losses plus modification concessions as a percentage of
the principal balance, this paper directly measures the
impact of a servicer’s actions using realized investor cash
flows. The results are qualitatively identical using either
calculation, subject to differences in the treatment of the
timing of cash flows and losses.
The monthly loan performance data provides a rich op-
portunity to observe actual NPV outcomes of mortgage
servicer decisions for the investor.10 The NPV to the in-
vestor of a servicer’s action is calculated by discounting
the monthly loan-level cash flows to the investor back to
the last current payment before the qualifying delinquent
episode. Monthly investor cash flows at the loan level are
calculated by adding the monthly interest payment11 to
the change in the loan’s scheduled balance and subtract-
ing any current period loss amounts. These cash flows are
discounted at a rate equal to the note rate on the par-
ticular loan at origination. For loans still active as of the
end of sample in December 2015, the remaining outstand-
ing payments due, at the P&I constant in December 2015,
are treated as an annuity. If the loan is delinquent, is in
bankruptcy or foreclosure, or is an REO property, a haircut
of 50% is applied to the annuity value.
10
The NPV to the investor remains the focus of this paper due to data
limitations. The cash flows to the servicer are largely opaque. Government
cash incentives to modify, late fees from the borrower, foreclosure man-
agement fees, and REO marketing fees, as well as the costs of different
loss mitigation techniques (amongst many other examples) are unobserv-
able. In addition, the haircuts and costs of borrowing relating to the fund-
ing of the servicing advance obligation are unavailable.
11
The monthly interest payment is calculated as the loan balance
owned by the investor at the beginning of the month (the beginning
scheduled balance) times the interest rate net of servicing fees, divided
by 12. If the rate net of all servicing fee strips is not available then the
gross rate is used instead.
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 2
First stage.
This table reports regression results relating to the first-stage estimates obtained by regressing a measure of servicer
financial constraints, the 1-Year Change in Advances as a Percentage of Loan Portfolio Balance in columns (1) and (2),
and the Credit Default Swap Spread in columns (3) and (4), on the instrument, the 1-Year Servicer Portfolio Housing
Price Return all measured at the time of borrower first delinquency, and a vector of controls. Columns (1) and (3)
represent Eq. (5) at a loan level and column (1) is the specification used as the first stage in the two-stage least squares
setup that underlies all of the instrumental variable results. Columns (2) and (4) perform the same regressions at the
servicer level, restricting attention to servicer-months that also appear in columns (1) and (3). First-stage coefficient
estimates from the servicer level are smaller than at the loan level, implying that the use of the loan-level specification
results in a more conservative IV estimate. The instrument utilized for the loan-level specifications is the housing price
variation of the servicer’s portfolio for geographies outside of the CBSA in which the loan resides, while the servicer-
level specifications utilize no geographic holdout. Servicer-level 1-Year Change in Delinquency Count As % of Servicer
Portfolio Loan Count and fixed effects are included in all specifications. Additionally, the servicer-level specifications
have year fixed effects while the loan-level specifications include the 1-Year Zip Housing Price Return for the zip code
where the loan resides, as well as a vector of loan and borrower characteristic controls and an interacted zip-year-credit
category fixed effect. Values of both the 1-Year Change in Advances as a Percentage of Loan Portfolio Balance measure as
well as the Credit Default Swap Spread have been standardized for the sample utilized in each regression. t-statistics
in parentheses are heteroskedasticity-robust and clustered at the servicer level. Columns (1) and (3) are additionally
clustered at the zip level. ∗ ∗ ∗ , ∗ ∗ , and ∗ indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

1-Year Change in Advances as a

Percentage of Loan Portfolio Balance Credit Default Swap Spread

Loan Servicer Loan Servicer

(1) (2) (3) (4)

1-Year Servicer Portfolio −8.213∗ ∗ ∗ −11.01∗ ∗ ∗

Housing Price Return (−5.00) (−5.48)
(Excluding Loan CBSA)

1-Year Servicer Portfolio −3.385∗ ∗ ∗ −9.131∗ ∗ ∗

Housing Price Return (−4.95) (−5.54)
(No Geo Holdout)

1-Year Zip Housing Price Return Yes — Yes —

1-Year Change in Delinquency Count Yes Yes Yes Yes

As % of Servicer Portfolio Loan Count

Loan/Borrower Characteristic Controls Yes — Yes —

Servicer FE Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes — Yes —
Year FE — Yes — Yes

Number of observations 6,868,451 9348 1,102,186 494

Adj. R2 0.432 0.184 0.561 0.580

Incremental Adj. R2 0.026 0.009 0.035 0.051

5. Results
In this section, I examine the actions that a servicer
takes in response to an increase in its financial con-
strainedness. A servicer has significant latitude in its abil-
ity to perform its obligations towards the MBS investor but
is contractually obligated to act in the investor’s best in-
terests. The actions taken under financial constraints have
the potential to significantly impact the value of the as-
set. A servicer’s obligation to advance principal and in-
terest payments to the investor in the event of borrower
delinquency represents a significant burden on its balance
sheet. A financially constrained servicer faces increased in-
centives to relieve or avoid its advance obligation. When a
borrower becomes delinquent, the servicer faces a choice:
either give the borrower an opportunity to self cure or ini-
tiate a default intervention, such as a modification or a
foreclosure. Given that the servicer’s decision to allow the
borrower more time to self cure involves the servicer ef-
fectively extending financing to a delinquent borrower, a
financially constrained servicer will be less willing or less
able to do so. Sections 5.1 and 5.2 demonstrate this effect.
More financially constrained servicers are more likely to
modify and foreclose on delinquent borrowers, and, con-
sequently, these borrowers are significantly less likely to
pay off their loan on their own. The financial constrained-
ness of mortgage servicers caused 442,857 foreclosures—
foreclosures that would have not occurred during the life
of a particular loan had servicers been less constrained
or had their constraints not mattered to the extent that
they did. The impact of hundreds of thousands of fore-
closures, through a spillover effect on the economy, rep-
resents a tremendously important indirect cost of financial
distress.
The agency conflict between the servicer and the in-
vestor in a securitization related to whether the ser-
vicer should intervene (modify or foreclose) is exacerbated
in the presence of financial constraints. In Section 5.3,
I exploit variation in the ease in which an investor is
able to monitor the servicer’s intervention decisions. For
delinquent loans of very high-quality the servicer’s deci-
sion is much less complicated. These borrowers should

598
D.J. Aiello
be given a more significant opportunity to self cure or
payoff their delinquent loans, implying lower monitor-
ing costs and lower agency frictions for these loans. For
these low agency friction loans, a servicer’s financial con-
straints do not significantly impact that servicer’s decisions
nor the outcomes of delinquent loans. For loans of aver-
age quality, when monitoring costs and agency frictions
are relatively higher, the servicer’s financial constraints do
matter.
The outcomes related to the behavior of a financially
constrained mortgage servicer, while undesirable from a
social welfare perspective, may have been to the benefit
of the investor. In order to quantify the value of a partic-
ular servicer’s actions to the investor, I measure the NPV
of all cash flows to the investor subsequent to that action.
The servicer is contractually required to maximize the in-
vestor’s NPV when making loss mitigation decisions on its
behalf. Section 5.4 discusses results where the NPV of the
realized cash flows of the delinquent mortgage loans is re-
gressed on the level of servicer financial constraints. Over-
all, more financially constrained servicers take actions that
are costly to the investor.
Even though constrained servicers modify and foreclose
on loans of better average quality (these borrowers were
more likely to pay off if left alone), their performance con-
ditional on intervention was worse than those performed
by less constrained servicers. Section 5.5 demonstrates the
impact that a servicer’s level of financial constrainedness
has on the performance of its default interventions, con-
ditional on a particular action being taken. Returning to
investor NPV, a more precise measure of investor benefit,
Section 5.6 demonstrates that individual modifications and
foreclosures performed by more financially constrained
servicers are, on average, worse. These investor NPV im-
pacts, conditional on outcome, are significantly smaller
than those that are unconditional on servicer action
or loan outcome (reported in Section 5.4). What finan-
cially constrained servicers did that was most harmful to
investors was to intervene at all.
Section 5.7 demonstrates that the unobservable qual-
ity of borrowers at the time of origination is not corre-
lated with the financial constrainedness of the servicer at
origination—the loans in the sample look identical up until
the first payment is missed (conditional on my regression
specification). The differences being measured only exist
once the loan becomes delinquent and the servicer has the
opportunity to (not) act.
Finally, Section 5.8 explores the manner in which ca-
pacity constraints interact with a servicer’s financial con-
straints, demonstrating that a more capacity constrained
servicer has a decreased ability to respond to its financial
constraints.
5.1. Default intervention
This section analyzes the impact of a servicer’s finan-
cial constraints on the ability and willingness of the ser-
vicer to provide financing to a delinquent borrower. By
modifying or foreclosing on a delinquent borrower, the ser-
vicer avoids advancing the monthly payments to the in-
vestor. This denies the borrower the opportunity to either
599
Journal of Financial Economics 144 (2022) 590–610
bring themselves current or to pay off the loan in full. This
change in behavior can be observed through an increase in
the likelihood of a default intervention amongst more con-
strained servicers.
In Table 3 an indicator for whether the servicer inter-
vened in the loan default, Default Intervention, is regressed
on a measure of servicer financial constrainedness. The co-
efficient is positive and significant (coef. = 0.0113), imply-
ing that an uninstrumented increase in advances is associ-
ated with an increase in the likelihood of servicer default
interventions. The controls in the regression include a se-
ries of loan and borrower characteristic controls, as well
as the housing price return over the year prior to delin-
quency for the zip code of the focal loan and the change in
the count of delinquencies in the servicer’s portfolio over
that same time frame. Fixed effects at the servicer level, as
well as an interaction between zip-code, default year, and
credit category, are also included. Column (2) reports re-
sults relating to a reduced form specification with the de-
fault intervention indicator as an outcome variable. This re-
duced form estimate implies that an increase in the port-
folio housing price return for a particular servicer is asso-
ciated with a decrease in its propensity to intervene with
a distressed borrower. Column (3) of Table 3 reports an IV
estimate of 0.140, implying that a one standard deviation
increase in the financial constrainedness of the servicer is
associated with a 14 percentage point increase in the prob-
ability of that servicer performing a default intervention.
This increase is relative to the servicer financing the bor-
rower in their delinquency, allowing the borrower to pay
off the loan in full or to become current again. Column (4)
presents robustness results related to the use of cumula-
tive levels of, rather than changes in, advances and demon-
strates a qualitatively similar result.
The rightmost portion of Table 3, columns (5)-(8),
present IV specifications using indicator variables for major
loan outcomes. A one standard deviation increase in the fi-
nancial constrainedness of the servicer increases not only
the probability of foreclosure by 9.06 percentage points,
but also the probability of modification by 5.26 percent-
age points. Given that the servicer initiates a foreclosure on
24.04% of the sample and a modification on only 7.13%, the
similar magnitude of the impact of financial constrained-
ness on both options is surprising. While foreclosures in-
crease more than modifications in absolute terms, a con-
strained servicer performs relatively more modifications.
Columns (7) and (8) demonstrate the borrower outcomes
that are being substituted for when the servicer intervenes.
Delinquent borrowers who are given more time by an un-
constrained servicer are more likely to be able to pay off
their loan or bring themselves current. When servicers be-
come constrained, they are either unwilling or unable to
extend advance financing, which leads to more foreclo-
sures and modifications.
5.2. Ultimate outcomes
Section 5.1 demonstrated that a constrained servicer
intervenes more aggressively in the event of a borrower
delinquency. The unit of observation for that specification
was the first episode of delinquency on any particular loan.
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 3
Default intervention.
This table reports results related to the estimation of a linear probability model regressing a series of loan outcome binary indicators (calculated
relative to the outcome of the borrower’s first episode of delinquency) on an instrumented 1-Year Change in Advances as a Percentage of Loan Portfolio
Balance measured at the time of borrower first delinquency, as well as a vector of controls and fixed effects. Column (1) additionally reports an
uninstrumented version of column (3), while column (2) reports the commensurate reduced form regression (Default Intervention indicator directly on
to the instrument). Column (4) duplicates column (3), but instead utilizing cumulative levels rather than changes in advances. A default intervention
is defined as either a foreclosure (column (5)), a modification (column (6)) or a liquidation. The coefficient estimate in column (3) is predominantly
relative to the counterfactuals of Paid-In-Full and Self Cure that are utilized in columns (7) and (8) respectively. Values of the 1-Year Change in Advances
as a Percentage of Loan Portfolio Balance and Advances as a Percentage of Loan Portfolio Balance measures have been standardized for the sample utilized
in each regression. t-statistics in parentheses are heteroskedasticity-robust and clustered at the servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and ∗ indicate statistical
significance at the 1%, 5%, and 10% levels, respectively.

Default Intervention Foreclosure Modification Paid-In-Full Self Cure

Reduced
OLS Form IV IV IV IV IV IV
(1) (2) (3) (4) (5) (6) (7) (8)

1-Year Change in Advances as a 0.0113∗ ∗ ∗ 0.140∗ ∗ ∗ 0.0906∗ ∗ ∗ 0.0526∗ −0.0959∗ ∗ ∗ −0.0357

Percentage of Loan Portfolio Balance (2.88) (5.07) (4.51) (1.75) (−2.94) (−1.54)

Advances as a Percentage of 0.220∗ ∗ ∗

Loan Portfolio Balance (4.16)

1-Year Servicer Portfolio −1.154∗ ∗ ∗

Housing Price Return (−7.91)
(Excluding Loan CBSA)

1-Year Zip Housing Price Return −0.721∗ ∗ ∗ −0.291∗ ∗ ∗ −0.473∗ ∗ ∗ −0.484∗ ∗ ∗ −0.368∗ ∗ ∗ −0.103∗ ∗ 0.379∗ ∗ ∗ 0.124∗ ∗
(−16.28) (−7.17) (−10.40) (−9.61) (−7.30) (−1.98) (4.85) (1.97)

1-Year Change in Delinquency Count −0.000397 0.000153 0.00496 0.00332 0.00509 −0.000179 −0.00708 0.00125
As % of Servicer Portfolio Loan Count (−0.44) (0.17) (0.85) (1.61) (1.06) (−0.11) (−1.55) (0.55)

Loan/Borrower Characteristic Controls Yes Yes Yes Yes Yes Yes Yes Yes
Servicer FE Yes Yes Yes Yes Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes Yes Yes Yes Yes Yes Yes

Number of observations 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451
Adj. R2 0.260 0.262 0.001

Dependent Variable Average 0.3135 0.3135 0.3135 0.3135 0.2404 0.0713 0.3961 0.2618

But does the impact of a servicer’s financial constraints on
a borrower’s outcome survive past just the first opportu-
nity for the servicer to intervene? This section investigates
the impact that a servicer’s financial constraint has on a
borrower’s ultimate outcomes. For example, were they ever
foreclosed on, did they ever receive a modification, or were
they ever able to pay their loan off in full?
Column (1) of Table 4 regresses an indicator, Ultimately
Foreclosed, measuring whether the servicer ever initiated
foreclosure proceedings at any point in the life of the loan,
on the measure of servicer financial constrainedness at the
time of the borrower’s first delinquency and the standard
controls. This OLS coefficient is small (0.0156) but signif-
icant, demonstrating a positive relationship between the
financial constrainedness of a servicer and the probabil-
ity of it ever initiating foreclosure. In column (2), the Ul-
timately Foreclosed outcome variable is regressed in a re-
duced form setting on the housing price return for loans
in a servicer’s portfolio that exist in geographies outside
of the focal loan’s CBSA in a reduced form setting. Column
(3) demonstrates that a one standard deviation increase in
a servicer’s financial constraints at the time of borrower
delinquency increases the probability that the servicer ever
forecloses on the borrower by 11.7 percentage points. Col-
umn (4) presents robustness results related to the use of
cumulative levels of, rather than changes in, advances.
Additionally, a financially constrained servicer increases
the rate at which loans are ultimately modified by 6.00
percentage points per standard deviation increase in fi-
nancial constrainedness (column (5) of Table 4). The in-
crease in foreclosures is ultimately at the expense of the
loan being able to pay off in full, which is reduced by 9.53
percentage points per standard deviation increase in con-
strainedness (column (6) of Table 4).
In all, a servicer’s financial constrainedness inflicted
over 442,0 0 0 extra foreclosures on the economy at large.
These foreclosures were avoidable in the sense that if the
mortgages had instead been serviced by an unconstrained
servicer (or if the servicer’s financial constraints had not
mattered to the extent they did), the borrower would have
been able to pay off their loan in full, or bring themselves
current.12
12
I do not account for the magnitude of the foreclosure externality it-
self, the neighborhood spillover effect, in my analysis (Gupta, 2019). By
leaving out the impact of a large geographic area around the focal loan,
the causal estimator avoids measuring any spillover effects, as well as any
endogeneity introduced through a servicer internalizing those costs as in
Favara and Giannetti (2017). This exclusion leads to underestimations of
the devastating impact that a mortgage servicer’s financial constraints had
on the economy during the recent financial crisis.

600
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 4
Ultimate outcomes.
This table reports results related to the estimation of a linear probability model regressing a series of loan outcome binary indicators on
an instrumented 1-Year Change in Advances as a Percentage of Loan Portfolio Balance measured at the time of borrower first delinquency,
as well as a vector of controls and fixed effects. Column (1) additionally reports an uninstrumented version of column (3), while col-
umn (2) reports the commensurate reduced form regression (Ultimately Foreclosed indicator directly on to the instrument). Column (4)
duplicates column (3), but instead utilizing cumulative levels rather than changes in advances. Outcome indicators utilized in this table
are calculated in reference to ultimate outcomes on the loan. That is, regardless of the initial result of the borrower’s first episode of
delinquency, was this loan ever placed into foreclosure (column (3)), modified (column (5)), or paid off in full (column (6)). Values of the
1-Year Change in Advances as a Percentage of Loan Portfolio Balance and Advances as a Percentage of Loan Portfolio Balance measures have
been standardized for the sample utilized in each regression. t-statistics in parentheses are heteroskedasticity-robust and clustered at the
servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and ∗ indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Ultimately Ultimately
Ultimately Foreclosed Modified Paid-In-Full

Reduced
OLS Form IV IV IV IV
(1) (2) (3) (4) (5) (6)

1-Year Change in Advances as a 0.0156∗ ∗ ∗ 0.117∗ ∗ ∗ 0.0600∗ −0.0953∗ ∗ ∗

Percentage of Loan Portfolio Balance (3.59) (5.45) (1.78) (−2.94)

Advances as a Percentage of 0.184∗ ∗ ∗

Loan Portfolio Balance (4.43)

1-Year Servicer Portfolio −0.964∗ ∗ ∗

Housing Price Return (−7.53)
(Excluding Loan CBSA)

1-Year Zip Housing Price Return −0.595∗ ∗ ∗ −0.248∗ ∗ ∗ −0.401∗ ∗ ∗ −0.409∗ ∗ ∗ −0.193∗ ∗ ∗ 0.380∗ ∗ ∗
(−7.27) (−3.55) (−5.50) (−5.62) (−3.75) (4.86)

1-Year Change in Delinquency Count 0.00137 0.00158 0.00560 0.00423∗ −0.000716 −0.00708
As % of Servicer Portfolio Loan Count (1.21) (1.33) (1.04) (1.92) (−0.46) (−1.55)
Loan/Borrower Characteristic Controls Yes Yes Yes Yes Yes Yes
Servicer FE Yes Yes Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes Yes Yes Yes Yes

Number of observations 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451

Adj. R2 0.262 0.263 0.003 0.003

Dependent Variable Average 0.3986 0.3986 0.3986 0.3986 0.1646 0.3963

5.3. Agency friction severity
There is an agency conflict between the servicer and
the investor in a securitization regarding whether the ser-
vicer should intervene (i.e. modify or foreclose) on a de-
faulted loan. The severity of this agency conflict varies with
how uncertain the servicer and investor are over what the
optimal action to take is, directly impacting monitoring
cost borne by the investor. I argue that for high-quality
delinquent loans, loans with origination FICOs greater than
or equal to 720 (24.9% of the regression sample) and orig-
ination LTVs less than or equal to 70% (28.8% of the re-
gression sample), the servicer’s decision is “easier”—these
borrowers are more likely (if allowed the opportunity
by their servicer) to bring themselves current, refinance,
or sell their home. These borrowers have more at stake
in the property (LTV≤70) and better financial resources
(FICO≥720).13
Because high-quality loans self cure or payoff with a
much higher frequency than loans of average quality, the
optimal action from the perspective of the investor is to
13
The interaction of this particular high-FICO and low-LTV threshold
designates 8.9% of my regression sample as high-quality. Results are ro-
bust to various permutations of the threshold definition.
not intervene, implying significantly reduced monitoring
costs for the investor. Alternatively, for average loans, in-
tervening may or may not be optimal, so the investor
must monitor to ensure that the servicer is protecting
its interests. This increased uncertainty for average-quality
loans is manifest in a rate of intervention that is much
closer to 50% than that for high-quality loans (33% com-
pared to 17%, respectively). Additionally, interventions in
high-quality loans occur at a much lower rate, implying
that there are fewer interventions to monitor and thus
per-defaulted-loan monitoring costs are lower. Therefore,
agency frictions around intervention decisions are drasti-
cally reduced for high-quality loans, providing an ideal set-
ting to test whether servicers act differently according to
the severity of agency frictions. I test the hypothesis that
there is a complementarity between agency frictions and
financial constraints in a regression setting where I hy-
pothesize two outcomes. First, that high-quality loans expe-
rience interventions at much lower frequencies than loans
of average quality. Second, that the coefficient on the in-
teraction of a high-quality loan indicator and the servicer’s
financial constraints is negative, implying that monitoring
costs for high-quality loans are lower, so the servicer has
much less scope to allow its financial constraints to influ-
ence its decision making.

601
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 5
Financial constraints exacerbate agency frictions.
This table reports results demonstrating that servicers act differently according to the severity of their agency frictions.
Column (1) replicates the instrumental variables specification presented in column (3) of Table 3. Columns (2) and (3)
update column (1) by including a High-Quality Loan Indicator as well as an interaction term between that indicator and the
1-Year Change in Advances as a Percentage of Loan Portfolio Balance. These instrumental variable specifications utilize 1-Year
Servicer Portfolio Housing Price Return (Excluding Loan CBSA) as well as the interaction of that variable with the High-Quality
Loan Indicator to instrument for the advances variable and its interaction. Column (4) replicates the specification in column
(3) in the setting of column (3) of Table 4. Columns (3) and (4) drop from the Loan/Borrower Characteristic Controls set both
the origination FICO and origination LTV and also reduce the interacted zip code, delinquency year, credit category fixed
effect dimension by removing credit category. The value of the F -statistic testing whether the sum of the coefficients for the
advances term and the interaction term is equal to zero, along with the corresponding p-value is also reported. Values of the
1-Year Change in Advances as a Percentage of Loan Portfolio Balance measure have been standardized for the sample utilized
in each regression. t-statistics in parentheses are heteroskedasticity-robust and clustered at the servicer and zip levels. ∗ ∗ ∗ ,
∗∗
, and ∗ indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Default Intervention Ultimately Foreclosed

IV IV IV IV
(1) (2) (3) (4)

1-Year Change in Advances as a 0.140∗ ∗ ∗ 0.146∗ ∗ ∗ 0.138∗ ∗ ∗ 0.114∗ ∗ ∗

Percentage of Loan Portfolio Balance (5.07) (5.16) (4.92) (5.28)
High-Quality Loan Indicator −0.057∗ ∗ ∗ −0.112∗ ∗ ∗ −0.158∗ ∗ ∗
(−3.81) (−9.98) (−11.27)

1-Year Change in Advances as a −0.111∗ ∗ ∗ −0.0943∗ ∗ ∗ −0.0964∗ ∗ ∗

Percentage of Loan Portfolio Balance (−4.34) (−3.45) (−3.58)

× High-Quality Loan Indicator

1-Year Zip Housing Price Return −0.473∗ ∗ ∗ −0.473∗ ∗ ∗ −0.468∗ ∗ ∗ −0.401∗ ∗ ∗

(−10.40) (−10.65) (−9.97) (−5.28)

1-Year Change in Delinquency Count 0.00496 0.00474 0.00470 0.00538

As % of Servicer Portfolio Loan Count (0.85) (0.83) (0.84) (1.06)

Loan/Borrower Characteristic Controls Yes Yes — —

Loan/Borrower Characteristic Controls (Adj) — — Yes Yes
Servicer FE Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes — —
Zip-Year FE — — Yes Yes

Number of observations 6,868,451 6,868,451 6,868,451 6,868,451

Adj. R2 0.003

F -Stat: Advances + Interaction = 0 (0.94) (1.40) (0.40)

p-Value 0.33 0.24 0.53

Columns (2) and (3) of Table 5 report the results of this
test by replicating column (3) of Table 3 (itself displayed
as column (1) of Table 5). Column (3) drops FICO and LTV
from the set of Loan/Borrower Characteristic Controls and
reduces one of the fixed effect dimensions from Zip-Year-
Credit Category to just Zip-Year. Additionally, I include an
indicator for whether the focal loan had both an origina-
tion LTV less than or equal to 70% and a FICO score at orig-
ination equal to or greater than 720, as well as the inter-
action between that indicator and my standard measure of
financial constraints. The financial constraints measure is
instrumented with servicer portfolio housing price returns
and the interaction term is instrumented with the inter-
action of the high-quality loan indicator and the servicer
portfolio housing price return. In this setting, I find an es-
timate of the effect of financial constraints on the prob-
ability of a servicer intervening in a defaulted loan to be
remarkably similar to my main specification—a coefficient
of 0.138 in column (3) of Table 5—indicating that finan-
cially constrained servicers modify and foreclose more ag-
gressively. Additionally, validating my hypothesis that high-
quality loans experience interventions at much lower fre-
quencies than do the average loan, the coefficient on High-
Quality Loan Indicator is negative (−0.112) and significant.
Most importantly, the interaction term has a negative co-
efficient (−0.094) and is significant, indicating that there
is a significantly reduced sensitivity to servicer financial
constraints in high-quality (low agency friction) loans. I
demonstrate that constrained and unconstrained servicers
act identically in their decision-making related to high-
quality loans by testing whether the coefficients on ad-
vances and advances interacted with a high-quality indica-
tor offset each other. I fail to reject the null hypothesis that
constrained and unconstrained servicers act the same in
regards to high-quality loans with an F -stat of 1.40. These
results indicate that when monitoring costs are low, a ser-
vicer’s financial constraints do not impact their decision
making.
The benefit in utilizing this high-quality loan setting to
explore heterogeneous agency frictions is that the variation
occurs at the loan level rather than at the securitization or
servicer level. Examining agency friction variations at the
loan level ensures that the role of advances and financial
constraints are held constant. Other potential dimensions

602
D.J. Aiello
along which agency frictions might vary include comparing
agency to non-agency securitizations—or even securitiza-
tions to portfolio loans—but the different advance regimes
in these alternate settings14 confound any apparent differ-
ing sensitivities to financial constraints.
I extend my tests of these hypotheses to the setting
with an outcome variable measuring whether a loan was
ever ultimately foreclosed on, following the specification
for column (3) of Table 4. In column (4) of Table 5 I again
find that financially constrained servicers are more likely
to foreclose (coef. = 0.114), that high-quality loans are less
likely to be foreclosed on (coef. = −0.158), the financial
constrainedness of servicers matter significantly less for
high-quality loans (coef. = −0.096), and that I cannot re-
ject that—in regards to high-quality loans—servicers act the
same regardless of whether or not they are financially con-
strained (F -stat = 0.40).
One potential concern with the above findings is that
there may be differential incentives on the part of the ser-
vicer to intervene that correlate with the quality of the
loan. This is only a potential concern insofar as these in-
centives are related to the financial constraints of the ser-
vicer (otherwise the effects get absorbed in the indicator
for high-quality loans). However, as long as the action the
investor would like the servicer to take is easily observ-
able and contractible, the servicer’s incentives do not mat-
ter. The servicing contract requires the servicer to act in
the investor’s interests, particularly in regards to modifi-
cations. “[T]he Master Servicer may also waive, modify, or
vary any term of any Mortgage Loan. if in the Master Ser-
vicer’s determination such... modification... is not materi-
ally adverse to the interests of the Certificateholders... ”
(RALI Series 20 07-QS6, 20 07). These results demonstrate
that a servicer behaves differently according to the sever-
ity of the agency friction environment. High-quality loans,
as contrasted with loans of average quality, leave much less
scope for the servicer to make a decision, have lower mon-
itoring costs, and consequently have less severe agency
frictions. The servicer’s financial constraints matter signifi-
cantly less in their decision making with high-quality loans
than with loans of average quality.
5.4. Investor value
To characterize the actions of a financially constrained
servicer as pinning down an agency cost, I determine the
impact on the NPV (at the time of first delinquency) of the
future cash flows to the investor. Section 4.4 details the
methodology of these calculations. Because precise data re-
lating to the investor cash flows is available subsequent
to default, the exact impact of a servicer’s level of fi-
nancial constrainedness can be determined. The outcomes
observed, while representing an ex-post outcome, are an
agency cost in the sense that they would not have been in-
14
For example, servicers of loans in Fannie Mae securitizations only ad-
vance the first four delinquent payments and servicers of loans held in
the portfolio of their parent entity never advance to themselves at all un-
der REMIC regulations.
603
Journal of Financial Economics 144 (2022) 590–610
curred had the servicer been less financially constrained.15
The magnitude of this realized agency cost, however, is
also influenced by the poor state realization of the econ-
omy.
Table 6 reports results relating to the overall impact
that a financially constrained servicer’s action (or inac-
tion) had on investor value. A servicer’s decision to push
a modification or a foreclosure on a borrower that would
have been able to self cure or pay-off, in addition to a
servicer’s poor handling of foreclosures and modifications
themselves, destroys an immense amount of value for its
principal. Column (1) of Table 6 presents the results of an
OLS specification, where the NPV of the loan cash flows
after default are regressed on a measure of the financial
constrainedness of the servicer. Columns (2) and (3) of
Table 6 estimate a reduced form and second-stage spec-
ification respectively with the NPV of the future investor
cash flows as the outcome variable. Column (4) presents
robustness results related to the use of cumulative levels
of, rather than changes in, advances.
Column (3) of Table 6 reports results relating to the
overall impact that a financially constrained servicer’s ac-
tion (or inaction) had on investor value. A financially con-
strained servicer destroyed $21,633 of investor value per
loan, for every standard deviation increase in the level of
financial constrainedness. Loans in sample had an average
balance at the time of default of $243,996. Therefore, a one
standard deviation increase in constrainedness reduces in-
vestor value by 8.87% of the outstanding principal value
of the loan at the time of default. Because the expected
NPV of a defaulted loan has to be less than its outstanding
principal balance, this represents a theoretical lower bound
on the actual loss percentage suffered by an investor due
solely to the level of financial constrainedness observed for
the servicer at the time of the first delinquency. The aver-
age NPV for the investor, at the time of first delinquency,
is $185,130. Thus, a true percentage loss figure closer to
11.7% per standard deviation increase in the financial con-
strainedness of the servicer is more likely.
Finally, in column (5) of Table 6, I demonstrate that
for high-quality loans (loans with low agency frictions),
there is no sensitivity in the investor’s NPV outcome to a
servicer’s financial constraints. The coefficient on financial
constraints is −22,604, similar to that reported in column
(3) of Table 6. Furthermore, high-quality loans have much
higher investor NPV relative to average-quality loans (coef.
= 18,485) and the sensitivity to a servicer’s financial con-
straints is eliminated for high-quality loans (coef = 31,648,
F -stat = 2.39, testing whether the financial constraint and
interaction coefficients offset).
Overall, investors that purchased $4.65 trillion in pri-
vate label MBS prior to the financial crisis suffered $81.71
billion in realized agency costs (accounting for 20.21% of
all investor value losses) associated with hiring an agent
that, in the presence of financial constraints, would inter-
vene aggressively in the event of default rather than pro-
15
Even considering that an investor might prefer sacrificing some NPV
to keep a struggling mortgage servicer afloat, the servicer’s financial con-
strainedness itself will still impose a loss that represents an agency cost.
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 6
Investor value.
This table reports results demonstrating the causal impact of a servicer’s financial constraints, as measured at the time of borrower
first delinquency by the 1-Year Change in Advances as a Percentage of Loan Portfolio Balance, on the NPV for the investor. Column (1)
reports the coefficient estimate for an uninstrumented OLS specification, column (2) displays the reduced form specification wherein
Investor NPV is regressed directly on the instrument 1-Year Servicer Portfolio Housing Price Return (Excluding Loan CBSA), and column
(3) utilizes an instrumental variables framework to report the causal estimate. Column (4) duplicates column (3), but instead utilizing
cumulative levels rather than changes in advances. Column (5) replicates the setting in Table 5 including an indicator for whether the
focal loan was high-quality as well as an interaction of that indicator with the advances measure of financial constraints. Advances
are instrumented as in column (3), whereas the interaction term is instrumented with the interaction of the indicator and that same
instrument. The standard controls and fixed effects are utilized throughout. Values of the 1-Year Change in Advances as a Percentage of
Loan Portfolio Balance and Advances as a Percentage of Loan Portfolio Balance measures have been standardized for the sample utilized
in each regression. t-statistics in parentheses are heteroskedasticity-robust and clustered at the servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and ∗
indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Investor NPV

Reduced
OLS Form IV IV IV
(1) (2) (3) (4) (5)

1-Year Change in Advances as a −1,906.0∗ ∗ −21,663.0∗ ∗ ∗ −22,604∗ ∗ ∗

Percentage of Loan Portfolio Balance (−2.03) (−3.62) (−3.67)

Advances as a Percentage of −33,993.8∗ ∗ ∗

Loan Portfolio Balance (−3.27)

High-Quality Loan Indicator 18,485∗ ∗ ∗

(5.87)

1-Year Change in Advances as a 31,648∗ ∗ ∗

Percentage of Loan Portfolio Balance (8.31)
× High-Quality Loan Indicator

1-Year Servicer Portfolio 177,918.3∗ ∗ ∗

Housing Price Return (4.93)
(Excluding Loan CBSA)

1-Year Zip Housing Price Return 99,470.9∗ ∗ ∗ 33,574.3∗ 61,658.2∗ ∗ ∗ 63,225.4∗ ∗ ∗ 60,648∗ ∗ ∗
(6.33) (1.67) (3.46) (3.50) (3.37)

1-Year Change in Delinquency Count −108.3 −186.4 −927.3 −675.5 −857.0

As % of Servicer Portfolio Loan Count (−0.33) (−0.57) (−0.90) (−1.59) (−0.87)

Loan/Borrower Characteristic Controls Yes Yes Yes Yes —

Loan/Borrower Characteristic Controls (Adj) — — — — Yes
Servicer FE Yes Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes Yes Yes —
Zip-Year FE — — — — Yes

Number of observations 6,868,451 6,868,451 6,868,451 6,868,451 6,868,451

Adj. R2 0.830 0.830 0.670 0.671 0.674

Dependent Variable Average 185,130 185,130 185,130 185,130 185,130

F -Stat: Advances + Interaction = 0 (2.39)
p-Value 0.12

viding the financing (and the time) the borrower needed
to help themselves.
5.5. Performance of foreclosures and modifications
Conditional on a servicer undertaking a foreclosure,
does its financial constrainedness impact its performance?
There are many dimensions under which a servicer can in-
fluence the performance of foreclosures once the foreclo-
sure has begun (the “first legal date”). Examples include
the speed at which the servicer is able to get the loan to
the foreclosure auction, the price the servicer receives at
auction, and the expenses the servicer incurs during the
process. Many of these items are difficult to disentangle.
A simple measure of the performance of the servicer, con-
ditional on foreclosure, is the loss severity (sales proceeds
less servicer expenses divided by the outstanding principal
balance, all at the time of liquidation).
Column (1) of Table 7 regresses the foreclosure liqui-
dation loss severity for loans where the servicer success-
fully foreclosed on the borrower after the first episode of
delinquency on the servicer’s financial constrainedness at
the time of first delinquency. For every one standard devia-
tion increase in the servicer’s financial constrainedness the
loss severity on the loan increased by 4 percentage points.
The loss severity represents the average percentage of the
principal balance that the investor wrote-off because of the
foreclosure, and had an average value of 68.03% in sam-
ple. While this result suggests a particular form of investor
value destruction conditional on the servicer initiating a
foreclosure, Section 5.6 investigates this more by utilizing
the investor NPV calculated at the time of first default.

604
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 7
Performance of foreclosures and modifications.
This table reports results relating to the performance and characteristics of foreclosures and modifications performed on loans following their first
episode of delinquency. Column (1) reports, conditional on a loan being foreclosed on during its first episode of delinquency, the results from
regressing the Foreclosure Liquidation Loss Severity on an instrumented measure of servicer financial constraints, the 1-Year Change in Advances as
a Percentage of Loan Portfolio Balance measured at the time of borrower first delinquency, and the standard controls and fixed effects. Columns
(2) and (3) report the results from regressing the percentage point reduction in interest rate associated with a modification on an instrumented
measure of servicer financial constraints, the 1-Year Change in Advances as a Percentage of Loan Portfolio Balance measured at the time of borrower
first delinquency, and the standard controls and fixed effects. Columns (4) and (5) regress indicators of whether the loan again became delinquent
three and twelve months respectively, after the modification brought the loan current on the measure of servicer financial constraints and the
standard controls and fixed effects. Columns (2)-(5) condition on a loan being modified during its first episode of delinquency. Columns (2) and
(3) further condition on the loan being a fixed rate loan both prior and subsequent to the modification, and column (3) on the modification
performed containing some interest rate reduction component. Values of the 1-Year Change in Advances as a Percentage of Loan Portfolio Balance
measure have been standardized for the sample utilized in each regression. t-statistics in parentheses are heteroskedasticity-robust and clustered
at the servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and ∗ indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

Foreclosure Liquidation Modification Interest 3mo 12mo

Loss Severity Rate Reduction Redefault Redefault

FC Only Fixed Rate Mods FR Int Rt Red. Mod Only Mod Only
IV IV IV IV IV
(1) (2) (3) (4) (5)

1-Year Change in Advances as a 0.0400∗ ∗ 1.689∗ ∗ ∗ 0.727∗ 0.0704∗ 0.0618∗

Percentage of Loan Portfolio Balance (2.49) (2.98) (1.86) (1.97) (1.82)

1-Year Zip Housing Price Return −0.108∗ ∗ ∗ −0.829 0.120 −0.0794 −0.0855
(−3.61) (−1.39) (0.39) (−1.34) (−1.11)

1-Year Change in Delinquency Count 5.32e-06 −0.211 −0.419 0.0369∗ ∗ 0.0303∗

As % of Servicer Portfolio Loan Count (0.00) (−0.70) (−0.46) (2.24) (1.67)

Loan/Borrower Characteristic Controls Yes Yes Yes Yes Yes

Servicer FE Yes Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes Yes Yes Yes

Number of observations 1,353,649 110,219 33,325 440,921 440,921

Adj. R2 0.050 0.143 0.004

Dependent Variable Average 68.03% 1.16% 3.05% 0.3498 0.5426

A financially constrained servicer is able to minimize its
future servicing advance obligation and more quickly re-
cover outstanding advances by increasing the rate at which
it modifies loans. To increase the rate of modification, the
servicer increases the benefits of modification to the bor-
rower by offering better terms. An interest rate reduction,
while costly to the investor, is beneficial to both the bor-
rower and servicer. The borrower benefits from a lower
rate and the servicer has lowered the monthly payment he
is obligated to advance in the event the modified loan re-
defaults (the likelihood of which is also reduced given the
lower interest rate). An interest rate reduction does not im-
pact the servicer’s monthly fee strip.
Columns (2) and (3) of Table 7 report results relating to
the impact that financial constraints have on the terms of
completed modifications. Conditional on a fixed-rate mort-
gage being modified, column (2) estimates Eq. (6) where
the dependent variable is the percentage point reduction
in interest rate after the modification. The average mod-
ification reduces the borrower’s rate by 1.16 percentage
points, and, for every standard deviation increase in the
financial constrainedness of the servicer, the borrower re-
ceives an additional 1.7 percentage point decrease in their
mortgage rate. Column (2) of Table 7 reports the impact of
financial constraints on the size of the modification inter-
est rate reduction, conditional only on modification. Col-
umn (3) additionally conditions on the modification in-
cluding an interest rate reduction component. In this case,
the average interest rate reduction, conditional on a modi-
fication with a reduction, is 3.05 percentage points. A one
standard deviation increase in the constrainedness of the
servicer implies a decrease in the post-modification rate of
an additional 0.73 percentage points.
While a financially constrained servicer makes more
modifications and provides the borrower with a lower
monthly principal and interest payment, whether or not
this is beneficial to the investor is unclear. The standard
measure of the success of a particular modification is
whether or not the borrower redefaulted (missing at least
one payment after modification completion) within a spec-
ified time frame. The average redefault rate is increasing in
the redefault window; 34.98% of modified loans in sample
went delinquent within three months after the completion
of the modification. A year after the modification, 54.26%
were again delinquent. While these estimates seem gener-
ally indicative of poorly designed loss mitigation strategies
on behalf of the servicers, the estimates are in line with
previous studies (An and Cordell, 2017) for modification
success rates during the recent financial crisis. Columns
(4) and (5) of Table 7 measure the impact that the fi-
nancial constrainedness of a servicer has on the effective-
ness of its modifications by implementing the instrumen-
tal variables framework of Eq. (6) at the level of a modi-
fied loan utilizing a binary indicator of whether or not the
loan redefaulted. Columns (4) and (5) use 3-month and 12-
month redefault rates, respectively, as the dependent vari-

605
D.J. Aiello
ables. Both coefficients are positive and qualitatively sim-
ilar and suggest that for every one standard deviation in-
crease in the financial constrainedness of the servicer, the
redefault rate increases by 7.04 and 6.18 percentage points
at 3-month and 12-month horizons, respectively.
This increase in redefault rate comes despite the more
generous terms provided to the borrower through an in-
terest rate reduction. This suggests some additional com-
ponent regarding the modification terms provided or ac-
tions performed by financially constrained servicers that
remains unobserved. Section 5.6 discusses results measur-
ing the impact of financial constraints on the investor NPV,
conditional on modification, which will account for the to-
tality of a constrained servicer’s impact on investor out-
comes. Overall, the results in Tables 3 and 7 suggest a hi-
erarchy in the actions of a financially constrained servicer
that is consistent with a financial constraints motivation.
Constrained servicers are more aggressive in performing
foreclosures, at the expense of loans that would have paid
off. Then, because modifications recover advances faster on
average than do foreclosures, some modifications would
have been better off as foreclosures from the perspective of
investors. While Table 3 shows that, in total, more foreclo-
sures are performed than modifications due to the finan-
cial constraints of servicers, in relative terms the increase
in modifications is much greater.
5.6. Conditional investor value
A financially constrained servicer modifies and fore-
closes on loans that, had they been left alone, would have
paid-off or self cured. Consequently, the overall perfor-
mance of these loans, conditional on a servicer’s actions,
could be better than that of loans foreclosed and modified
by unconstrained servicers. However, Section 5.5 demon-
strates that the actions of a servicer under financial stress
have a deleterious impact on the performance of both fore-
closures and modifications. This section demonstrates that,
conditional on a loan being either foreclosed or modified,
greater levels of servicer financial constrainedness at the
time of loan first delinquency reduce investor NPV. Despite
the fact that constrained servicers are modifying and fore-
closing on loans that are in some sense “better,” they still
destroyed value on average.
Column (1) of Table 8 demonstrates that, conditional
on foreclosure, a financially constrained servicer performed
significantly worse than an unconstrained one from the
perspective of the investor. For every standard devia-
tion increase in the financial constrainedness of the ser-
vicer, the investor NPV of the foreclosure was reduced
by $10,727. Column (2) demonstrates that, conditional on
modification, a financially constrained servicer performed
modifications that had worse investor NPV outcomes than
those performed by an unconstrained one. Regressing in-
vestor NPV on an instrumented measure of financial con-
straints results in per standard deviation value destruction
of $6352 on average across all modifications. Column (3)
shows that the financial constrainedness of the servicer
has little impact on the performance of short sale and
charge-off liquidations. Demonstrating that, beyond decid-
ing which loss mitigation tactic to take, the outcome of
606
Journal of Financial Economics 144 (2022) 590–610
these liquidations is invariant to the particular actions of
the servicer.
5.7. Unobservable borrower quality
My identification strategy is subject to the important
caveat that unobservable characteristics of the borrower
and loan at origination look the same across servicers that
will experience different levels of financial constrained-
ness. The observation that a particular servicer is servicing
a particular loan might provide information as to the bor-
rower’s unobserved type, which in turn could influence the
appropriate value-maximizing decision on the part of the
servicer.16 To address this possibility, I first control for all
observable loan and borrower characteristics. Additionally,
I include a servicer-level fixed effect to remove any time-
invariant average effect (such as manager skill), as well
as an interacted fixed effect of year, zip, and credit cate-
gory that ensures I am comparing similar loans. To address
any remaining concerns with this identifying assumption,
I propose three tests that are ultimately suggestive of no
relationship between the financial constrainedness of the
servicer and the unobservable quality of the borrower at
the time of origination.
Columns (1) and (2) of Table 9 regress an indicator for
whether the loan ever went delinquent over a one-year or
a five-year window, respectively, following loan origination
on the financial constrainedness of the servicer at the time
of loan origination and the standard controls (measured at
the time of loan origination, where applicable). No statisti-
cally significant relationship between a servicer’s financial
constrainedness and the probability of their serviced loans
going delinquent is identified.
Column (3) demonstrates that no statistically significant
relationship exists between the length of time it took for a
borrower to go delinquent and the financial constrained-
ness of the servicer at the time of loan origination. The
servicer has little to no opportunity to influence the time
it takes for a borrower to go delinquent. Conditional on
the fixed effect structure, loans serviced by financially con-
strained servicers look identical to those serviced by un-
constrained servicers up until the first missed payment.
They only diverge once they go delinquent and the servicer
has the opportunity to react.
Aiello (2016) shows that a borrower that makes their
initial few monthly payments at least one business day be-
fore their actual due date (generally the first of the month)
on a regular basis has a sharply reduced likelihood of ever
going delinquent in the future. This propensity to pay an
obligation before it is due, “borrower diligence,” is unob-
servable at origination. Following this idea, I measure the
number of payments made before they were due during
an early period of the life of the loan and regress this
on the level of financial constrainedness of the servicer at
loan origination. Columns (4), (5), and (6) of Table 9 re-
port these results. No statistically significant relationship
exists between the financial constrainedness of the servicer
16
See DeMarzo (2005), Aiello (2016), Begley and Purnanandam (2017),
and Adelino et al. (2019) on the information asymmetries found in the
pooling and tranching of mortgages.
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 8
Conditional investor value.
This table reports instrumental variable regression results similar to that of column (3) in
Table 6 with further conditioning based on the servicer action taken after the first episode
of borrower delinquency. Columns (1)-(3) condition on foreclosure, modification, and liqui-
dation (generally a charge-off or a short sale) respectively. Values of the 1-Year Change in
Advances as a Percentage of Loan Portfolio Balance measure have been standardized for the
sample utilized in each regression. t-statistics in parentheses are heteroskedasticity-robust
and clustered at the servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and ∗ indicate statistical significance at
the 1%, 5%, and 10% levels, respectively.

Investor NPV

FC Only Mod Only Liq Only

IV IV IV
(1) (2) (3)

1-Year Change in Advances as a −10,727.0∗ ∗ ∗ −6,382.1∗ ∗ −63.38

Percentage of Loan Portfolio Balance (−3.00) (−2.03) (−0.02)

1-Year Zip Housing Price Return 12,330.9∗ −31,785.5∗ 6,292.2

(1.68) (−1.93) (0.32)

1-Year Change in Delinquency Count 85.28 0.419 11,692.5

As % of Servicer Portfolio Loan Count (0.21) (0.00) (0.30)

Loan/Borrower Characteristic Controls Yes Yes Yes

Servicer FE Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes Yes

Number of observations 1,592,194 440,921 7306

Adj. R2 0.576 0.588 0.655

Dependent Variable Average 123,560 168,093 33,610

Table 9
Unobservable borrower quality.
This table regresses various measures of unobservable borrower quality on an instrumented measure of servicer financial constraints, the 1-Year Change
in Advances as a Percentage of Loan Portfolio Balance, measured at the time of loan origination. Columns (1) and (2) utilize the full sample of loans
in the BlackBox dataset (unconditional on delinquency) and regress an indicator of delinquency over a 1-year and 5-year (post-origination) window,
respectively, on the servicer’s financial constraints at the time of loan origination. Column (3) regresses the number of months between loan origina-
tion and the borrower’s first delinquency on the servicer’s financial constraints at the time of loan origination. Columns (4)-(6) utilize a measure of
unobservable borrower quality related to the frequency with which the borrower makes their monthly payment at least a day before it is due. This
measure has been established in the literature as highly correlated with future loan delinquency and default outcomes and is, crucially, unobservable
at the time of loan origination. Values of the 1-Year Change in Advances as a Percentage of Loan Portfolio Balance measure have been standardized for
the sample utilized in each regression. t-statistics in parentheses are heteroskedasticity-robust and clustered at the servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and
∗
indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

DQ Prob DQ Prob Months To Early Payments, Early Payments, Early Payments,

1-Year 5-Year First DQ Mos 1–6 Mos 4–6 Mos 4–9

IV IV IV IV IV IV
(1) (2) (3) (4) (5) (6)

1-Year Change in Advances as a 0.0309 0.0375 −3.145 −4.371 −0.244 −0.928

Percentage of Loan Portfolio Balance (0.55) (0.77) (−1.45) (−1.03) (−0.87) (−0.97)
At Origination

1-Year Zip Housing Price Return −0.0358 −0.0572 4.807∗ −0.169 0.190 0.501
At Origination (−0.65) (−1.16) (1.84) (−0.06) (0.70) (0.75)

1-Year Change in Delinquency Count As % of 2.65e-04 −2.02e-04 −0.0252 −6.850 −0.0564 −0.203
Servicer Portfolio Loan Count, At Origination (0.57) (−0.35) (−0.68) (−1.09) (−0.83) (−0.88)

Loan/Borrower Characteristic Controls Yes Yes Yes Yes Yes Yes

Servicer FE Yes Yes Yes Yes Yes Yes
Zip-Orig. Year-CreditCat FE Yes Yes Yes Yes Yes Yes

Number of observations 8,889,284 8,889,284 6,363,798 2478 2,514,099 2,227,485

Adj. R2 0.012 0.016 0.013

Dependent Variable Average 0.25 0.74 25.47 1.87 0.93 1.84

607
D.J. Aiello
at the time of origination and a measure of the borrower’s
unobservable quality, as inferred by observing initial pay-
ment habits.
The probability of delinquency, the time to borrower
first delinquency, and the borrower’s payment habits are
measures of quality that have nothing to do with the ser-
vicer. These measures, unobservable at origination, show
that financially constrained servicers do not have loans
that vary, even unobservably, from those serviced by un-
constrained servicers—at least until the borrower goes
delinquent and the servicer has the opportunity to (not)
act.
5.8. Capacity constraints
In this section, I explore the role that capacity con-
straints play in my setting. My first objective is to demon-
strate that my measure of financial constraints, servicer ad-
vances, is not likely to be confounded to a meaningful de-
gree with the capacity constraints of the servicer. Secondly,
I explore the relationship between capacity constraints and
financial constraints in a servicer’s decision-making.
One possible challenge to the interpretation of my
causal mechanism as being related to the financial con-
strainedness of the mortgage servicer is the fact that a
financially constrained servicer is also likely to be opera-
tionally constrained. An influx of delinquent loans not only
requires a large outflow of cash in the form of servicing ad-
vances, but also necessitates a significant operational com-
mitment to appropriately handle these loans. Although, as
discussed in Section 4, the likelihood that this explains my
result remains low due to their direction (i.e., Table 3), I
include throughout the paper the change in the count of
servicer portfolio delinquencies for the year prior to loan
delinquency as a percentage of the total loans in the ser-
vicer’s portfolio, as a control.
My measure of financial constraints, the recent change
in the amount of servicing advances paid out by a servicer
as a percentage of its loan portfolio balance, consists of
three sources of variation: the number, loan balance (size),
and the delinquency severity of the delinquent loans in a
servicer’s portfolio. By controlling for the count of servicer
delinquencies, I allow my servicer housing price return in-
strument to operate exclusively through the dollar amount
of advances extended by the servicer per delinquent loan.
For example, consider two properties experiencing rel-
atively different zip code-level housing price movements.
The property that has experienced a significantly larger de-
cline in housing prices is likely going to experience more
missed payments. These properties, with differential delin-
quency severities, clearly provide drastically different lev-
els of servicer advances, which affect financial constraints
quite differently. It is not clear, however, that the differ-
ence in delinquency severity is a driver of differential ca-
pacity constraints. The cost of intervening17 in the loans
related to these properties is largely fixed with respect to
17
Costs borne by the servicer related to modifying a delinquent loan
include the design of the modified instrument, drafting of the modified
note, solicitation of the borrower, and execution of the modified docu-
ments. In relation to a foreclosure these costs vary by jurisdiction but
608
Journal of Financial Economics 144 (2022) 590–610
the severity of the delinquency. To the extent there ex-
ists differences in the fixed costs of intervening, they are
largely driven by institutional differences correlated (and
removed) with a zip code level fixed effect. Column (1)
of Table 10 replicates column (3) of Table 3 for compari-
son. Column (2) omits any control for capacity constraints,
demonstrating that the principal coefficient related to ser-
vicer financial constraints is largely unaffected by whether
I control for a measure of servicer capacity constraints.
The main result of this paper demonstrates that a ser-
vicer, when faced with an increase in its financial con-
strainedness, will act to aggressively modify and foreclose
on the delinquent loans in its portfolio to recover out-
standing advances on these loans as quickly as possible.
Because servicers face operational constraints with respect
to the number of delinquent loans they are capable of han-
dling, when they are overwhelmed with delinquencies, we
should expect a relatively more capacity constrained ser-
vicer to be significantly less able to respond to increased
financial constraints by increasing the rate at which he is
modifying and foreclosing on delinquent loans.
I test this hypothesis by replicating, in columns (3)
and (4) of Table 10, column (3) of Table 3 split into
two samples. The sample is split between loans that first
went delinquent at a time when the servicer of that loan
was experiencing below median levels of delinquency as
a percentage of their portfolio loan count (column (3) of
Table 10, Low Capacity Constraint), and when they were ex-
periencing above median levels (column (4) of Table 10,
High Capacity Constraint). As these results demonstrate, a
servicer with low capacity constraints is nearly twice as
responsive to their financial constraints in their decision
to intervene as a servicer with high capacity constraints.
Thus, a servicer facing a large volume of delinquent loans
is not able to increase their default intervention activity in
response to their financial constraints to the same extent
as a servicer facing a relatively low level of delinquencies.
6. Conclusion
This paper analyzes the relationship between financial
constraints and agency costs by exploiting an important
aspect of the institutional setting of securitizations. A fi-
nancial intermediary, the servicer, is obligated to advance
to the investor the monthly payments of delinquent bor-
rowers. Advances represented a major source of financial
stress for servicers during the recent financial crisis. In all
structured finance transactions, not just residential mort-
gages, we care deeply about servicer and investor conflicts.
Understanding how, when, and why a servicer’s balance
sheet comes into play is a first-order concern (Loutskina
and Strahan, 2009; Loutskina, 2011).
Mortgage servicers entered the financial crisis with
contracts in place that obligated them to bear a signifi-
cant portion of the short-term risk in the event of large-
scale mortgage defaults. As agents wholly responsible for
the management of mortgages, they were positioned to
often involve court filings and appearances, public notifications, and evic-
tions by local county sheriffs.
D.J. Aiello Journal of Financial Economics 144 (2022) 590–610

Table 10
Capacity constraints.
This table reports, in column (1), a regression identical to column (3) of Table 3. Column (2) mimics column (1),
but with the control for capacity constraints omitted entirely. Columns (3) and (4) report results from specifi-
cations identical to those of column (1), but with additional sample conditioning. Column (3) demonstrates the
results conditioning on loans experiencing their first delinquency in months where the servicer has lower capac-
ity constraints relative to the median measured capacity constraint for that servicer. Column (4), similarly, reports
results in the sample where the servicer has above-median capacity constraints. t-statistics in parentheses are
heteroskedasticity-robust and clustered at the servicer and zip levels. ∗ ∗ ∗ , ∗ ∗ , and ∗ indicate statistical significance
at the 1%, 5%, and 10% levels, respectively.

Default Intervention Default Intervention

Low Capacity High Capacity

Constraint Constraint
IV IV IV IV
(1) (2) (3) (4)

1-Year Change in Advances as a 0.140∗ ∗ ∗ 0.140∗ ∗ ∗ 0.245∗ ∗ ∗ 0.133∗ ∗ ∗

Percentage of Loan Portfolio Balance (5.07) (5.16) (2.76) (4.54)

1-Year Change in Delinquency Count 0.00496 −0.296 0.00249

As % of Servicer Portfolio Loan Count (0.85) (−0.92) (0.45)

1-Year Zip Housing Price Return −0.473∗ ∗ ∗ −0.474∗ ∗ ∗ −0.517∗ ∗ ∗ −0.525∗ ∗ ∗

(−10.40) (−10.45) (−6.86) (−9.57)

Loan/Borrower Characteristic Controls Yes Yes Yes Yes

Servicer FE Yes Yes Yes Yes
Zip-Year-CreditCat FE Yes Yes Yes Yes

Number of observations 6,868,451 6,868,451 3,349,398 3,469,802

Adj. R2

have immense influence on the value of a large volume of
assets owned by major systemic institutions. The destruc-
tion of $81.71 billion of investor value on defaulted loans
with original balances of $1.72 trillion and the infliction
of 442,85718 foreclosures on the economy is of enormous
concern. These servicers traded reputation for short-term
liquidity.
A mortgage servicer faces a decision relating to how
quickly a modification or foreclosure should take place in
the event of delinquency. In waiting, a servicer is forced
to finance the borrower until he self cures. I demonstrate
that financially constrained servicers were either unwilling
or unable to provide that financing and consequently mod-
ified and foreclosed at higher rates as compared to uncon-
strained servicers. These effects are concentrated amongst
loans with higher levels of agency frictions between the
servicer and the investor.
Policymakers at the Federal Reserve acted in 2009 to
relieve these constraints in a manner that was likely effec-
tive at accomplishing its stated goal of “prevent[ing] avoid-
able foreclosures.” One possible motivation for this pol-
icy is the evidence presented in this paper. However, this
policy response to the financial crisis was largely an at-
tempt to prevent foreclosures and increase modifications.
This study shows that relieving a servicer’s financial con-
straints, while effective at reducing foreclosures, will not
increase modifications. Relief will instead lead to some-
thing even better—an increased opportunity for a borrower
to become current or pay off through their own efforts.
These lessons—learned as a result of the financial crisis—
18
Representing 16% of the approximately 2.7 million foreclosures in my
sample.
are universally applicable to the way we manage and think
about the role of servicer balance sheets, and have impor-
tant ramifications for later crises like the COVID-19 pan-
demic.
The agency costs and economic destruction associated
with the actions of important intermediaries under con-
siderable financial stress are important to study. Both in-
vestors and servicers, while likely aware of potential distor-
tionary influences related to servicer financial constrained-
ness, were unlikely to have appreciated the true magnitude
of the value destruction that could stem from these deci-
sions. Likewise, regulators, in acting to extend TALF cover-
age to servicing advance obligations in 2009, were too late
to avoid the swathes of foreclosures that were caused by
servicers’ financial constraints.
References
Adelino, M., Gerardi, K., Hartman-Glaser, B., 2019. Are lemons sold first?
Dynamic signaling in the mortgage market. J. Financ. Econ. 132 (1),
1–25.
Adelino, M., Gerardi, K., Willen, P., 2014. Identifying the effect of secu-
ritization on foreclosure and modification rates using early payment
defaults. J. Real Estate Finance Econ. 49 (3), 352–378.
Adelino, M., Gerardi, K., Willen, P.S., 2013. Why don’t lenders renegoti-
ate more home mortgages? Redefaults, self-cures and securitization.
J. Monet. Econ. 60 (7), 835–853.
Agarwal, S., Amromin, G., Ben-David, I., Chomsisengphet, S., Evanoff, D.D.,
2011. The role of securitization in mortgage renegotiation. J. Financ.
Econ. 102 (3), 559–578.
Agarwal, S., Amromin, G., Ben-David, I., Chomsisengphet, S., Piskorski, T.,
Seru, A., 2017. Policy intervention in debt renegotiation: evidence
from the home affordable modification program. J. Polit. Economy 125
(3), 654–712.
Aiello, D.J., 2016. Information Exploitation? A Pre-Crisis RMBS Issuer’s Pri-
vate Information. Working Paper.
Aiello, D. J., Garmaise, M. J., Natividad, G., 2017. Competing for deal flow
in mortgage markets.

609
D.J. Aiello
An, X., Cordell, L.R., 2017. Regime Shift and the Post-crisis World of Mort-
gage Loss Severities. Working Paper.
Ang, A., Green, R.C., Longstaff, F.A., Xing, Y., 2017. Advance Refundings of
Municipal Bonds. J. Finance.
Ang, J.S., Cole, R.A., Lin, J.W., 20 0 0. Agency costs and ownership structure.
J. Finance 55 (1), 81–106.
Begley, T.A., Purnanandam, A., 2017. Design of financial securities: empir-
ical evidence from private-label RMBS deals. Rev. Financ. Stud. 30 (1),
120–161.
Census Bureau, 2016. Core-based statistical areas. https://www.census.
gov/topics/housing/housing-patterns/about/core-based-statistical-
areas.html. Accessed: 2021-05-03.
Chodorow-Reich, G., Falato, A., 2021. The loan covenant channel: how
bank health transmits to the real economy. J. Finance. Forthcoming
Chomsisengphet, S., Lu, C., 2015. Servicer Distress and Mortgage Renego-
tiations. Working Paper.
Cookson, J.A., Niessner, M., 2020. Why don’t we agree? Evidence from a
social network of investors. J. Finance 75 (1), 173–228.
Cordell, L., Dynan, K., Lehnert, A., Liang, N., Mauskopf, E., 2008. The In-
centives of Mortgage Servicers: Myths and Realities.
Correia, S., 2016. Linear Models with High-Dimensional Fixed Effects: An
Efficient and Feasible Estimator. Working Paper.
DeMarzo, P.M., 2005. The pooling and tranching of securities: a model of
informed intermediation. Rev. Financ. Stud. 18 (1), 1–35.
Diamond, D., Rajan, R., 2009. The credit crisis: conjectures about causes
and remedies. Am. Econ. Rev..
Eisfeldt, A.L., Rampini, A.A., 2007. New or used? Investment with credit
constraints. J. Monet. Econ. 54 (8), 2656–2681.
Ellul, A., Jotikasthira, C., Lundblad, C.T., 2011. Regulatory pressure and
fire sales in the corporate bond market. J. Financ. Econ. 101 (3),
596–620.
Esty, B.C., 1997a. Organizational form and risk taking in the savings and
loan industry. J. Financ. Econ. 44 (1), 25–55.
Esty, B.C., 1997b. A case study of organizational form and risk shifting in
the savings and loan industry. J. Financ. Econ. 44 (1), 57–76.
Favara, G., Giannetti, M., 2017. Forced asset sales and the concentration of
outstanding debt: evidence from the mortgage market. J. Finance 72
(3), 1081–1118. doi:10.1111/jofi.12494.
Granja, J., Matvos, G., Seru, A., 2017. Selling failed banks. J. Finance 72 (4),
1723–1784.
Griffin, J.M., Kruger, S., Maturana, G., 2019. Do labor markets discipline?
Evidence from RMBS bankers. J. Financ. Econ. 133 (3), 726–750.
Griffin, J.M., Kruger, S., Maturana, G., 2020. What drove the 20 03–20 06
house price boom and subsequent collapse? Disentangling competing
explanations. J. Financ. Econ..
Gupta, A., 2019. Foreclosure contagion and the neighborhood spillover ef-
fects of mortgage defaults. J. Finance 74 (5), 2249–2301.
Hartman-Glaser, B., Mann, W., 2016. Collateral Constraints, Wealth Effects,
and Volatility: Evidence from Real Estate Markets. Working Paper
Herndon, T., 2021. Punishment or forgiveness? Loan modifications in pri-
vate label residential mortgage-backed securities from 2008 to 2014.
Rev. Polit. Economy 1–29.
610
Journal of Financial Economics 144 (2022) 590–610
Huang, Y., Nadauld, T.D., 2019. A direct test of agency theories of debt:
evidence from residential mortgage-backed securities. Manage. Sci. 65
(4), 1792–1809.
Kim, Y.S., Laufer, S.M., Stanton, R., Wallace, N., Pence, K., 2018. Liquidity
crises in the mortgage market. Brookings Pap. Econ. Act. 2018 (1),
347–428.
Koijen, R.S., Yogo, M., 2014. The cost of financial frictions for life insurers.
Am. Econ. Rev. 105 (1), 445–475.
Kroszner, R.S., Strahan, P.E., 1996. Regulatory incentives and the thrift cri-
sis: dividends, mutual-to-stock conversions, and financial distress. J.
Finance 51 (4), 1285–1319.
Kruger, S., Maturana, G., 2020. Collateral misreporting in the residential
mortgage-backed security market. Manage. Sci..
Kruger, S.A., 2017. The effect of mortgage securitization on foreclosure and
modification. J. Financ. Econ..
Lee, S.-J., Mayers, D., Smith, C.W., 1997. Guaranty funds and risk-taking
evidence from the insurance industry. J. Financ. Econ. 44 (1), 3–24.
Loutskina, E., 2011. The role of securitization in bank liquidity and funding
management. J. Financ. Econ. 100 (3), 663–684.
Loutskina, E., Strahan, P.E., 2009. Securitization and the declining impact
of bank finance on loan supply: evidence from mortgage originations.
J. Finance 64 (2), 861–889.
Loutskina, E., Strahan, P.E., 2011. Informed and uninformed investment in
housing: the downside of diversification. Rev. Financ. Stud. 24 (5),
1447–1480.
Maturana, G., 2017. When are modifications of securitized loans beneficial
to investors? Rev. Financ. Stud..
Mayock, T., Shi, L., 2018. Adverse Selection in the Market for Mortgage
Servicing Rights. Working Paper.
Merrill, C.B., Nadauld, T.D., Stulz, R.M., Sherlund, S., 2021. Were there fire
sales in the RMBS market? J. Monet. Econ. 122, 17–37.
Mian, A., Sufi, A., 2009. The consequences of mortgage credit expansion:
evidence from the US mortgage default crisis. Q. J. Econ. 124 (4),
1449–1496.
Mooradian, R.M., Pichler, P., 2017. Servicer contracts and the design of
mortgage-backed security pools. Real Estate Econ..
Murfin, J., 2012. The supply-side determinants of loan contract strictness.
J. Finance 67 (5), 1565–1601.
Palmer, C., 2015. Why Did So Many Subprime Borrowers Default During
the Crisis: Loose Credit or Plummeting Prices?.
Piskorski, T., Seru, A., Vig, V., 2010. Securitization and distressed loan
renegotiation: evidence from the subprime mortgage crisis. J. Financ.
Econ. 97 (3), 369–397.
RALI Series 20 07-QS6, 20 07. Residential funding company, LLC. Prospec-
tus Supplement dated April 25, 2007. Available at https://www.vision.
ocwen.com/
Singh, M., Davidson, W.N.I., 2003. Agency costs, ownership structure and
corporate governance mechanisms. J. Bank. Finance 27 (5), 793–816.
Whited, T.M., 1992. Debt, liquidity constraints, and corporate investment:
evidence from panel data. J. Finance 47 (4), 1425–1460.
Wong, M., 2018. CMBS and conflicts of interest: evidence from ownership
changes for servicers. J. Finance 73 (5), 2425–2458.