University of Chicago Law School
Chicago Unbound
1&*2-"./1+*.("0&1&1*&2*./42*.("6".%
/,*$7
/1+*.("0&12
Mortgage Re%nancing, Consumer Spending, and
Competition: Evidence from the Home A$ordable
Re%nancing Program
Sumit Agarwal
Gene Amromin
Souphala Chomsisengphet
Tomasz Piskorski
Amit Seru
/,,/63)*2".%"%%*3*/.",6/1+2"3 )<02$)*$"(/4.#/4.%4$)*$"(/&%4
)/42*.(!,"6!".%!0/,*$7
"13/'3)& "6/--/.2
)*$"(/.#/4.%*.$,4%&2#/3)6/1+2*.01/(1&22".%;.",5&12*/.2/'"13*$,&2,&"2&#&"6"1&3)"3
"-/1&1&$&.35&12*/./'3)*2"13*$,&-"7#&"5"*,"#,&/.)*$"(/.#/4.%/1&,2&6)&1&
:*2/1+*.("0&1*2#1/4()33/7/4'/1'1&&".%/0&."$$&22#73)&/1+*.("0&12"3)*$"(/.#/4.%3)"2#&&."$$&03&%'/1*.$,42*/.*.
1&*2-"./1+*.("0&1&1*&2*./42*.("6".%/,*$7#7"."43)/1*8&%"%-*.*231"3/1/')*$"(/.#/4.%/1-/1&*.'/1-"3*/.0,&"2&$/.3"$3
4.#/4.%,"64$)*$"(/&%4
&$/--&.%&%*3"3*/.
4-*3("16",&.&-1/-*./40)",")/-2*2&.(0)&3/-"28*2+/12+*-*3&14*.$&.3 "//13("(&&;.".$*.(
/.24-&10&.%*.(".%/-0&3*3*/.5*%&.$&'1/-3)&/-&9/1%"#,&&;.".$*.(1/(1"-1&*2-"./1+*.("0&12&1*&2
*./42*.(
Electronic copy available at: http://ssrn.com/abstract=2662906
CHICAGO
KREISMAN WORKING PAPER ON HOUSING LAW AND POLICY NO. 27
M
ORTGAGE REFINANCING, CONSUMER SPENDING, AND
COMPETITION: EVIDENCE FROM THE HOME AFFORDABLE
REFINANCING PROGRAM
Sumit Agarwal, Gene Amromin, Souphala Chomsisengphet, Tomasz Piskorski,
Amit Seru, & Vincent Yao
THE LAW SCHOOL
THE UNIVERSITY OF CHICAGO
October 2015
This paper can be downloaded without charge at the Kreisman Working Papers Series in Housing Law
and Policy: http://chicagounbound.uchicago.edu/housing_law_and_policy
and The Social Science Research Network Electronic Paper Collection.
Electronic copy available at: http://ssrn.com/abstract=2662906 Electronic copy available at: http://ssrn.com/abstract=2662906
Mortgage Refinancing, Consumer Spending, and Competition:
Evidence from the Home Affordable Refinancing Program
Sumit Agarwal
a
Gene Amromin
b
Souphala Chomsisengphet
c
Tomasz Piskorski
d
Amit Seru
e
Vincent Yao
f
AUGUST 2015
Abstract
We examine the ability of the government to impact mortgage refinancing activity and spur consumption
by focusing on the Home Affordable Refinancing Program (HARP). The policy allowed intermediaries to
refinance insufficiently collateralized mortgages by extending government credit guarantee on such loans.
We use proprietary loan-level panel data from a large market participant with refinancing history and
social security number matched consumer credit records of each borrower. A difference-in-difference
empirical design based on eligibility requirements of the program reveals a substantial increase in
refinancing activity by the program: more than three million eligible borrowers with primarily fixed-rate
mortgages – the predominant contract type in the U.S. -- refinanced their loans under HARP. Borrowers
received a reduction of around 140 basis points in interest rate, on average, due to HARP refinancing,
amounting to about $3,500 in annual savings per borrower. There was a significant increase in the durable
spending by borrowers after refinancing, with larger increase among more indebted borrowers. Regions
more exposed to the program saw a relative increase in non-durable and durable consumer spending, a
decline in foreclosure rates, and faster recovery in house prices. A variety of identification strategies
reveal that competitive frictions in the refinancing market may have partly hampered the program’s
impact. On average, these frictions reduced take-up rate among eligible borrowers by 10%-20% and cut
interest rate savings by 16-33 basis points, with larger effects among the most indebted borrowers who
were the key target of the program. These findings have implications for future policy interventions, pass-
through of monetary policy through household balance sheets, and design of the mortgage market.
Keywords: Financial Crisis, HARP, Debt, Refinancing, Consumption, Spending, Household Finance,
Mortgages, Policy Intervention
JEL Classification Codes: E65, G18, G21, H3, L85
_____________________________________
*First Version: March 2014. The paper does not necessarily reflect views of the FRB of Chicago, the Federal
Reserve System, the Office of the Comptroller of the Currency, or the U.S. Department of the Treasury.
Acknowledgements: The authors would like to thank Charles Calomiris, Erik Hurst, Tullio Jappelli, Arvind
Krishnamurthy, Chris Mayer, Emi Nakamura, Stijn Van Nieuwerburgh, Tano Santos, Amir Sufi, Adi Sunderam and
seminar participants at Columbia Business School, NBER Summer Institute, CEPR Gerzensee Summer Symposium,
and Stanford Institute for Theoretical Economics, for helpful comments and suggestions. Zach Wade provided
outstanding research assistance. Piskorski acknowledges funding from the Paul Milstein Center for Real Estate at
Columbia Business School and the National Science Foundation (Grant 1124188). Seru acknowledges funding from
the Initiative on Global Markets at Booth School of Business at the University of Chicago. a: National University of
Singapore; b: Federal Reserve Bank of Chicago; c: OCC; d: Graduate School of Business, Columbia University and
NBER; e: Booth School of Business, University of Chicago and NBER and f: Georgia State University.
Electronic copy available at: http://ssrn.com/abstract=2662906 Electronic copy available at: http://ssrn.com/abstract=2662906
1
I. Introduction
Mortgage refinancing is one of the main channels through which households can benefit from
decline in the cost of credit. Indeed, because fixed rate mortgage debt is the dominant form of
financial obligation of households in the U.S and many other economies, refinancing constitutes
one of the main direct channels for transmission of simulative effects of accommodative
monetary policy (Campbell and Cocco 2003, Scharfstein and Sunderam 2014, Keys, Pope, and
Pope 2014). Consequently, in times of adverse economic conditions, central banks commonly
lower interest rates in order to encourage mortgage refinancing, lower foreclosures, and stimulate
household consumption. However, the ability of such actions to influence household
consumption through refinancing depends on the ability of households to access refinancing
markets and on the extent to which lenders compete and pass-through lower rates to consumers.
While there is no work that that has systematically analyzed these issues, the importance of the
first factor became apparent in aftermath of the recent financial crisis when many mortgage
borrowers lost the ability to refinance their existing loans (Hubbard and Mayer 2009).
1
Faced
with a situation in which close to half of all borrowers in the economy were severely limited
from accessing mortgage markets, the federal government launched a large-scale refinancing
initiative called the Home Affordable Refinancing Program (HARP). In a nutshell, HARP
allowed eligible borrowers with insufficient equity to refinance their mortgages by extending
explicit federal credit guarantee to lenders. This paper uses HARP as a laboratory to examine the
government’s ability to impact refinancing activity and spur household consumption.
Our paper has two objectives. First, we want to quantify the impact of HARP on mortgage
refinancing activity and analyze consumer spending and other economic outcomes among
borrowers and regions exposed to the program. In doing so, we hope to assess consumer
behavior around refinancing among borrowers with Fixed Rate Mortgages (FRMs), the
predominant contract type in the U.S. Second, after demonstrating that a substantial number of
eligible borrowers did not benefit from the program, we will analyze the importance of
competitive frictions in the refinancing market in hampering HARP’s reach. This sets us apart
from prior work that has focused on borrower specific factors, like inattention, in explaining their
sluggish response to refinancing incentives (Andersen, Campbell, Nielsen, and Ramadorai 2014).
We use a proprietary dataset from a large secondary market participant to execute our analysis.
The dataset covers more than 50% of conforming mortgages (more than 20 million) issued with
guarantees of the Government Sponsored Entities (GSEs). This loan-level panel data has detailed
information on loan, property, and borrower characteristics and monthly payment history.
Importantly, this data contains unique identifiers (Social Security Numbers) for each borrower
1
CoreLogic estimates that in early 2010, close to a quarter of all mortgage borrowers owed more than their houses
were worth and another quarter had less than 20% equity, a common threshold for credit without external support.
2
allowing us to construct their refinancing history, determine the present and prior mortgage
terms during the refinancing process including fees charged by GSEs for insuring credit default
risk (g-fee), the servicer responsible for their prior and current mortgage, as well as accurately
capture various forms of consumer debt using their linked credit bureau records.
We start our analysis by assessing the impact of the program on the mortgage refinancing rate.
To get an estimate of the counterfactual level in the absence of the program, we exploit variation
in exposure of similar borrowers to the program. Specifically, we use high loan-to-value (LTV)
loans sold to GSEs (the so-called conforming loans) as the treatment group since these loans
were eligible for the program. Loans with observationally similar characteristics, but issued
without government guarantees (non-agency loans), serve as a control group since these
mortgages were ineligible for the program. Using difference-in-differences specifications we
find a large differential change in the refinancing rate of eligible loans relative to the control
group after the program implementation date. Thus, by addressing the problem of limited access
to refinancing due to insufficient equity, HARP led to substantial number of refinances (more
than 3 million). We also quantify the extent of savings received by borrowers on HARP
refinances and find around 140 basis points of interest rate savings were passed through on the
intensive margin. This amounts to about $3,500 in annual savings per borrower -- a 20%
reduction in monthly mortgage payments.
We also analyze the consumer spending patterns among borrowers who refinanced under the
program. Our analysis based on new auto financing patterns suggests that borrowers significantly
increased their durable (auto) spending (by about $1,600 over two years) after the refinancing
date, about 20% of their interest rate savings. This increase in spending is substantially larger
among more indebted and less creditworthy borrowers. We augment this analysis by assessing
how outcome variables, accurately measured at the zip code level -- such as non-durable and
durable consumer spending, foreclosures, and house prices—changed in regions based on their
exposure to the program. The advantage of this data is that it allows us to measure household
consumption directly, including credit card spending, as well as analyze the broader effects of
the program (e.g. on house prices). We find that regions more exposed to the program
experienced a meaningful increase in durable and non-durable consumer spending (auto and
credit card purchases), relative decline in foreclosure rate, and faster recovery in house prices.
Although the first part of the paper illustrates that the program had considerable impact on
refinancing activity, it also shows that a significant number of eligible borrowers did not take
advantage of the program. In the second part of our analysis, we investigate the role of
intermediary competition in impacting HARP’s reach and effectiveness.
There are at least a couple of reasons why competitive frictions could play an important role in
the program implementation. First, to the extent that an existing relationship might confer some
3
competitive advantage to the incumbent servicer -- whether through lower (re-) origination costs,
less costly solicitation, or better information regarding borrower conditions -- such advantages
could be enhanced under the program since it targeted more indebted borrowers. Second, in an
effort to encourage servicer participation, the program rules imposed a lesser legal burden on
existing (incumbent) servicers.
To shed light on the importance of such factors, we start by comparing the interest rates on
HARP refinances to the interest rates on regular conforming refinances originated during the
same period (HARP-conforming refi spread). The latter group serves as a natural counterfactual,
as the funding market for such loans – extended to creditworthy borrowers with significant
housing equity -- was quite competitive and remained fairly unobstructed throughout the crisis
period. Thus, the spread captures the extent of pass-through of lower interest rates to borrowers
refinancing under the program relative to those refinancing in the conforming market.
Importantly, we take advantage of our detailed data on fees charged by GSEs for insuring credit
risk of loans (g-fees) to precisely account for differences in interest rates due to differential
creditworthiness of borrowers refinancing in the two markets.
We find that, on average, a loan refinanced under HARP carries an interest rate that is 16 basis
points higher relative to conforming mortgages refinanced in the same month. This suggests a
more limited pass through of interest savings under HARP relative to the regular conforming
market. The markup is substantial relative to the mean interest rate savings on HARP refinances
(140 basis points). Notably, this spread persists when we account for a host of observable loan,
borrower, property and regional characteristics. Moreover, the spread increases substantially
with the current LTV of the loan, reaching more than 30 basis points for high LTV loans. These
findings emerge despite the fact that in computing this spread we removed g-fees that account
for differential mortgage credit risk due to higher LTV ratios. This finding is consistent with the
idea that higher LTV loans, those with very limited refinancing options outside the program, may
confer higher advantage to incumbent lender. In addition, we find that loans refinanced under the
program by larger lenders – ones who are likely to have significant monopoly power in several
local markets -- carry higher spreads.
Next, we exploit variation within HARP borrowers that relates the terms of their refinanced
mortgages to the interest rate on their legacy loans, i.e., rate on the mortgage prior to HARP
refinancing. The main idea underlying this test is that, in the presence of limited competition,
incumbent lenders would extract more surplus from borrowers with higher legacy rates since
such borrowers could be incentivized to refinance at relatively higher rates. Indeed, borrowers
with higher legacy rates experience substantially smaller rate reductions on HARP refinances
compared with otherwise observationally similar borrowers with lower legacy rates. These
results survive when instrumenting the legacy interest rate on a mortgage with the 10-year U.S.
4
Treasury yield prevailing at the time of origination of the legacy mortgage to obtain variation in
the borrower’s legacy rate that is exogenous to individual and regional characteristics.
Finally, we take advantage of the change in the program rules introduced in January 2013 that
relaxed the asymmetric nature of higher legal burden for new lenders refinancing under the
program relative to incumbent ones. In essence, this change was aimed to alleviate the barriers to
competition in the HARP refinancing market. We use a difference-in-difference setting around
the program change to directly assess how changes in competition in the refinancing market
impacted intensive (mortgage rates) and extensive (refinancing rates) margins. There is a sharp
and meaningful reduction in the HARP-conforming refi spread (by more than 30%) around the
program change. Moreover, there was a concurrent increase in the rate at which eligible
borrowers refinanced under the program (6%) relative to refinancing rate in the conforming
market. These estimates imply that the refinancing rate among eligible borrowers would be about
10 to 20 percentage points higher if HARP refinances were priced similar to conforming ones
(accounting for variation in g-fees). The effects are the largest among the group of the borrowers
that were the main target of the program – i.e., those with the least amount of home equity. These
are also the borrowers, who as we show in the first part of the paper, also displayed larger
increase in spending conditional on program refinancing. Overall, these results suggest that by
adversely altering refinancing activity, competitive frictions may have significantly reduced the
program effect on consumption of eligible households, especially those targeted by the program.
Our paper is closely related to a recent literature that examines the importance of institutional
frictions and financial intermediaries in effective implementation of stabilization programs,
particularly in housing markets. In particular, focusing on the Home Affordable Modification
Program (HAMP), Agarwal et al. (2012) provide evidence that servicer-specific factors related to
their preexisting organizational capabilities – such as servicing capacity -- can importantly affect
the effectiveness of policy intervention in debt renegotiation that rely on such intermediaries for
its implementation.
2
In contrast, our work suggests that competition in intermediation market
may also play a role in effective implementation of some stabilization policies.
Our paper is also related to the growing literature on the pass-through of monetary policy,
interest rates, and housing shocks through household balance sheets (e.g., Hurst and Stafford
(2004), Mian, Rao, and Sufi (2013), Keys, Piskorski, Seru, and Yao (2014), Di Maggio, Kermani
and Ramcharan (2014), Chen, Michaux, Roussanov (2014), Auclert (2015), Agarwal,
Chomsisengphet, Mahoney, and Strobel (2015) and Beraja et al. (2015)). Within this literature
2
Since HARP requires servicer participation for its implementation, such factors (e.g. servicer capacity constraints)
could also affect the program reach. Notably, refinancing activity, the target of HARP, is a relatively a routine and
fairly standardized activity that servicers have significant experience doing. In contrast, HAMP’s objective was to
stimulate mortgage renegotiation, a more complex activity that servicers have limited experience with and one that
requires significant servicing infrastructure. Consequently, relative to HAMP, competitive frictions could play a
more important role in HARP implementation compared with servicer organizational capabilities.
5
we provide a novel assessment of the largest policy intervention in refinancing market during the
recent crisis. Our findings also complement those of Scharfstein and Sunderam (2014) who show
that, in general, refinancing markets with a higher degree of lender concentration experienced a
substantially smaller pass-through of lower market interest rates to borrowers.
3
We also contribute to the vast literature on studying consumption responses to various fiscal
stimulus programs. Some studies include Shapiro and Slemrod (1995, 2003), Jappelli et al.
(1998), Souleles (1999), Parker (1999), Browning and Collado (2001), Stephens (2008),
Johnson, Parker, and Souleles (2006), Agarwal, Liu, and Souleles (2007), Aaronson et al. (2012),
Mian and Sufi (2012), Parker, Souleles, Johnson, and McClelland (2013), Gelman et. al. (2014)
and Agarwal and Qian (2014).
4
Our analysis relies on a period with lower interest rates, where
borrowers with insufficiently collateralized mortgages had large incentives to refinance, but were
unable to do so (Hubbard and Mayer 2009). HARP generated an exogenous increase in supply of
refinancing opportunities and we find significant increase in consumer spending among
borrowers and regions exposed to the program. This finding is consistent with the
contemporaneous work by Amromin, Di Maggio, and Kermani (2015) who also find that
borrowers increase their spending after HARP refinancing and comprehensively explore the
mechanisms driving the borrower response. Together, this evidence suggests that consumer
spending response to mortgage refinancing can be an important part of transmission of monetary
policy to the economy since lower rates generally induce more refinancing.
Our paper is also related to the recent empirical literature that studies borrowers’ refinancing
decisions (e.g., Koijen et al. 2009, Agarwal, Driscoll and Laibson 2013; Keys, Pope and Pope
2014, Agarwal, Rosen and Yao 2014; Anderson et. al. 2014). This literature focuses on borrower
specific factors like limited inattention and inertia in explaining their refinancing decisions.
While such borrower specific factors can also help account the muted response to HARP (see
Johnson et al. 2015 for recent evidence), our work emphasizes the importance of financial
intermediaries and the degree of market competition in explaining part of this shortfall.
Finally our work relates broadly to the recent growing literature on the housing and financial
crisis (e.g., Mayer et al. 2009 and 2014; Keys et al. 2010, 2012; Charles, Hurst and Notowidigdo
2013, Favilukis, Ludvingson, and Van Nieuwerburgh 2013; Eberly and Krishnamurthy 2014,
Hsu, Matsa and Melzer, 2014, Melzer 2014, Stroebel and Vavra 2014). We contribute to this
literature by providing the first comprehensive assessment of the largest intervention aimed at
stimulating mortgage refinancing during the Great Recession.
3
Within a broader context on market competitiveness and pricing power, this paper is related to the seminal work of
Rotemberg and Saloner (1987) and to the earlier empirical research on pass through of changes in interest rates by
lenders in different competitive environments (e.g., Neumark and Sharpe 1992).
4
The literature finds mixed evidence: some studies find that the consumption response is essentially zero, while
other find that liquidity constrained consumers respond positively to the fiscal stimulus programs.
6
II. Background and Empirical Strategy
II.A U.S. mortgage markets before and during the Great Recession
The U.S. mortgage markets are characterized by several unique features. First, a majority of
mortgage contracts offer fixed interest rates and amortize over long time periods, commonly set
at 15 or 30 years. Second, most mortgages can be repaid in full at any point in time without
penalties, typically by taking out a new loan backed by the same property (refinancing). Finally,
the majority of mortgages, the so-called conforming loans, are backed by government-sponsored
enterprises or GSEs.
5
The GSEs guarantee full payment of interest and principal to investors on
behalf of lenders and in exchange charge lenders a mixture of periodic and upfront guarantee
fees (called “g-fees”). In practice, both types of g-fees are typically rolled into the interest rate
offered to the borrower and are collected as part of the monthly mortgage payment. The interest
rates charged to borrowers are thus affected by three main components: the yield on the
benchmark Treasury notes to capture prevailing credit conditions, the credit profile of the
borrower that affects the g-fee charged for insurance of default risk (which depends on factors
such as FICO credit score and LTV ratio), and finally, a lender’s markup. In addition the
borrowers need to satisfy a set of criteria to be eligible for conforming financing based on factors
such as loan amount and LTV ratio.
6
Under this institutional setup, a borrower with a FRM might be able to take advantage of
declines in the general level of interest rates by refinancing a loan. The economic gain from
refinancing is clearly affected by potential changes in borrower creditworthiness, as well as the
mortgage market environment. During periods of favorable economic conditions, such as those
between 2002 and 2006, refinancing market functioned smoothly. Borrower incomes and credit
scores remained steady. Home prices increased, allowing equity extraction at refinancing while
maintaining stable LTV ratios. Defaults were rare and supply of mortgage credit was plentiful.
Each of these components changed dramatically during the Great Recession. Rapidly rising
unemployment rates and the attendant stress to household ability to service debt obligations
impaired income and credit scores. As home prices dropped precipitously, many borrowers were
left with little or no equity in their homes, making them ineligible for conforming loan
refinancing. By early 2010, close to a quarter of all mortgage borrowers found themselves
5
As of the end of 2013, GSE-backed securities (agency mortgage backed securities [MBS]) accounted for just over
60% of outstanding mortgage debt in the U.S. About half of the agency MBS market is backed by Fannie Mae,
slightly less than 30% is backed by Freddie Mac, and the rest is backed by Ginnie Mae, which securitizes mortgages
made by the Federal Housing Administration (FHA) and Veterans Administration (VA). For the purposes of this
paper, our discussion of GSEs will be limited to the practices of Fannie Mae and Freddie Mac. FHA and VA
mortgages also represent an important source of funding, particularly for low down-payment loans to borrowers
with somewhat impaired credit history. However, they remain beyond the scope of this paper.
6
Conforming mortgages cannot exceed the eligibility limit, which has been $417,000 since 2006 for a 1-unit, single-
family dwelling in a low-cost area. In addition, most such loans have LTV ratios at origination no greater than 80%.
7
“underwater”, i.e. owing more on their house than it was worth (CoreLogic data). Refinancing
was also made more difficult by a virtual shutdown of the private securitization market, as
investors fled mortgage-backed securities not explicitly backed by the federal government
leading to a massive exit of lenders from the subprime mortgage industry.
7
Since refinancing
underwater or near-underwater loans would be considered extending unsecured credit and trigger
prohibitive capital charges, balance sheet (portfolio) lending for such borrowers dried up as well.
Overall, due to the environment in the credit industry, borrowers with insufficient home equity
were shut out of refinancing markets, even as countercyclical monetary policy actions drove
mortgage interest rates to very low levels.
II.B The Home Affordable Refinancing Program (HARP) and asymmetric pricing power
In the face of massive disruptions in mortgage markets, the Treasury Department and the Federal
Housing Finance Agency (FHFA) developed a program to allow households with insufficient
equity to refinance their mortgages. This policy action – the Home Affordable Refinancing
Program, or HARP –instructed GSEs to provide credit guarantees on refinances of conforming
mortgages, even in cases when the resulting loan-to-value ratios exceeded the usual eligibility
threshold of 80 percent. Initially, only loans with an LTV of up to 105% could qualify. Later in
2009, the program was expanded to include loans with an estimated LTV at the time of
refinancing up to 125%. Finally in December 2011, the program rules were changed again by
removing any limit on negative equity for mortgages so that even those borrowers owing more
than 125% of their home value could refinance, creating what is referred to as “HARP 2.0”.
After a number of extensions of its end date, HARP is scheduled to end on December 31, 2016.
Given the size of GSE-backed mortgage holdings, opening up refinances for this segment of the
market had the potential to influence household consumption. Although refinancing imposed
losses on the existing investors in mortgage backed securities (MBS) who had to surrender high-
interest paying assets in a low-interest-rate environment, it benefitted borrowers who lowered
their interest payments and received a substantial reduction in the NPV of their mortgage
obligations (Eberly and Krishnamurthy 2014). Consequently, HARP aimed to provide economic
stimulus to the extent that liquidity-constrained borrowers had higher marginal propensities to
consume than MBS investors. It also potentially lowered the likelihood of delinquencies and
subsequent foreclosures that may result in substantial deadweight losses (Mian et al. 2011).
HARP got off to a slow start, refinancing only about 300,000 loans during the first full year of
the program. Overall, more than 3 million borrowers refinanced during the first five years of the
7
Most prominent subprime lenders failed outright (e.g., Countrywide, Washington Mutual, Wachovia, IndyMac,
and Lehman Brothers).
8
program, which amounts to up to between 40 to 60 percent of potentially eligible borrowers as of
the program start date in March 2009.
8
Market commentary pointed to a number of flaws in the program design, which included
frictions with junior liens and origination g-fee surcharges (LLPAs) that limited borrowers’
potential gains from refinancing. Crucially, lender willingness to participate in HARP was
potentially undermined by ambiguities about the program’s treatment of representations and
warranties (R&W).
9
Any mortgage found to be in violation of its R&W can be returned (“put
back”) to the originator, who would then bear all of the credit losses. The risk of put backs
became particularly pronounced in the wake of the financial crisis when mortgage investors and
GSEs began conducting aggressive audits for possible R&W violations on every defaulted loan.
In the case of low-equity and underwater loans targeted by HARP the risk of default was
considered to be particularly high. As a result, mortgage originators that securitized their loans
through GSEs could have regarded R&W as a major liability.
Policymakers recognized this issue and HARP lessened the underwriting requirements and the
attendant R&W on loans refinanced through the program. However, this relief from put back risk
on refinanced loans was granted asymmetrically, favoring lenders that were already servicing
mortgages prior to their being refinanced through HARP.
10
Such lenders faced few underwriting
requirements and little exposure to this risk. In contrast, lenders that were refinancing mortgages
that they did not already service had to face stringent R&W treatment.
11
Finally, HARP rules
were also asymmetric in servicer treatment since the program required less onerous underwriting
if performed through a borrower’s existing servicer rather than through a different servicer.
12
II.C Post-HARP 2.0 developments
In January 2013, FHFA addressed concerns about the open-ended nature of R&W violation
reviews. This took on two forms: (1) FHFA clarified a sunset provision for R&W reviews,
setting the time frame over which such reviews could be done at 1-year for HARP transactions;
8
The estimates on the number of potentially eligible borrowers vary. Based on Treasury and FHFA estimates up to
8 million of borrowers could have been eligible for the program (with an estimated 4-5 million borrowers having the
opportunity to refinance under HARP 1.0 with up to additional 2-3 million borrowers becoming eligible due to the
removal of LTV eligibility limit under HARP 2.0). See FHFA, “HARP: A Mid Program Assessment” (2013).
9
In every transaction, the mortgage originator certifies the truthfulness of information collected as part of the
origination process, such as borrower income, assets, and house value. This certification is known as R&W.
10
Mortgage originators that securitized loans through GSE typically retained servicing rights on those mortgages. In
their role as a servicer, they collected payments, advanced them to the MBS trustee, and engaged in a variety of loss
mitigating actions on delinquent loans. We use the terms “servicer” and “lender” interchangeably.
11
This difference in treatment may result in higher expected origination costs for would-be competitors of existing
lenders. Additionally, this market power to existing servicers may have been more consequential in the case of high
LTV borrowers since such borrowers would be associated with greater default risk, and hence higher put back risk.
12
For instance, under HARP the lender had to verify that eligible borrowers missed at most one payment on their
existing loan during the previous 12 months. This information was already available with the incumbent lender.
9
(2) FHFA clarified which violations were subject to this sunset and which were severe enough
(e.g. fraud) to be subject to life-of-the-loan timeframe. These changes went into effect in January
2013. The clarification of the R&W process may have had a direct effect on the competitive
advantage of same-servicer HARP refinances. Before the sunset provisions, a new servicer was
taking on an indefinite (or at least ambiguous) R&W risk. However, with the provision in place,
this risk was limited to a 1-year window for a pre-specified set of violations.
III. Data and Empirical Setting:
III.A Data
The main data used in this study comes from a large proprietary database of conforming
mortgages securitized by a large secondary market participant. The conforming loans are
mortgages that satisfy the underwriting guidelines of GSEs (such as Fannie Mae and Freddie
Mac). These mortgages are usually made to borrowers with relatively high credit scores, low
initial LTV ratios, and fully documented incomes and assets. In addition, these mortgages must
meet the conforming loan limit. Recall that only conforming mortgages were eligible for
refinancing under HARP. Thus, our data, which covers more than 50% of conforming loans --
amounting to more than 20 million of outstanding residential mortgages as of the program
implementation in March 2009 -- is well suited to study the program.
This loan-level monthly panel data has detailed dynamic information on rich array of loan,
property, and borrower characteristics (e.g., interest rates, location of the property and current
borrower credit scores and LTV ratios) and monthly payment history (e.g., delinquent or not).
Importantly, as this data contains unique Social Security Numbers (SSN) for each borrower, we
can track the refinancing history of each borrower, the servicer responsible for prior and current
mortgage of the borrower, and whether refinancing was done under HARP. This matched data
allows us to obtain all the present and prior mortgage terms including all relevant information on
fees applied during the refinancing process including GSE g-fees.
This detailed panel data on the refinancing history of each borrower constitutes a considerable
advantage over commercially available products, which do not provide information on the entire
sequence of transactions at the borrower level. This rich data allows us to account for detailed
borrower level characteristics as well as conduct within borrower analysis to assess how terms of
loans obtained under HARP relate to the terms of their previous transactions.
The data provider has merged the mortgage data with each borrower’s consumer credit bureau
records by using unique borrower identifier. These merged data allow us to observe the current
credit history of mortgage holders in each month. Of particular importance to us is the auto debt
balance information, which allows us to construct empirical measures of new auto spending
10
patterns at the borrower level. Our data ends in mid-2013, a period after which there were
relatively few HARP originations.
In our analysis we also employ the loan-level mortgage data collected by BlackBox Logic that
covers more than 90% of privately securitized mortgages that were not sold to GSEs. These data
were merged with borrower-level credit report information collected by Equifax using a
proprietary match algorithm giving us a similar set of variables as our main dataset.
13
We focus
on a set of borrowers with loans in this data that are similar on observables to those in our main
sample. However, as these loans were not sold to GSEs they were not eligible for the program.
Consequently, this group of loans serves as a counterfactual for the loans that could be
potentially refinanced under the program.
Finally, in our regional analysis we collect individual loan-level information from four databases.
The first source is the LPS database maintained by Black Knight Financial Services, which
provides dynamic information on the vast majority of loans in the United States. We complement
this dataset with the information from the BlackBox database, which yields almost complete
coverage of mortgage loans in the United States, allowing us to compute zip-code-level
characteristics for variables such as average borrower FICO credit scores, fraction of HARP
eligible loans among all mortgages in a zip code, average mortgage interest rates, as well as zip
code level foreclosure rates. The second dataset provided by the Office of the Comptroller of
Currency allows us to measure the quarterly credit card spending of borrowers in a particular zip
code. The third database comprises the auto sales data from R. L. Polk & Company (see Mian,
Rao, and Sufi 2013), which allows us to directly measure the car purchases in a zip code. Finally,
we also use zip code level house price indices from CoreLogic.
III.B Empirical Setting
Our empirical analysis consists of two main parts. In the first part we aim to quantify the impact
of HARP on mortgage refinancing and assess household spending and other economic outcomes
around the program implementation. In the second part, we investigate the role of intermediary
competition on the reach and effectiveness of the program.
We start our analysis by assessing the impact of the program on the mortgage refinancing rate.
We focus on fixed-rate mortgages, the predominant mortgage type in the U.S., which, unlike
adjustable-rate mortgages, cannot automatically benefit from lower market rates. To get an
estimate of the counterfactual level in the absence of the program, we exploit variation in
exposure of similar borrowers to the program. Specifically, high LTV loans sold to GSEs
(“conforming” loans) serve as the treatment group, while loans with observationally similar
13
BlackBox is a private company that provides a comprehensive, dynamic dataset with information on privately
securitized subprime, Alt-A, and prime loans originated after 1999. Equifax is a major credit reporting agency that
provides monthly data on consumer credit standing.
11
characteristics but issued without government guarantees (“non-agency” loans) – ineligible for
the program -- serve as a control group. Using a difference-in-differences specification we will
assess the differential change in the refinancing rate of the treatment group relative to the control
group around the program implementation date. The identification assumption behind this
comparison is that, in the absence of the program, the refinancing rates in the control and
treatment groups would follow similar patterns (up to a constant difference).
Next we quantify the extent of savings received by borrowers refinancing under HARP and shed
some light on consumer spending patterns around the refinancing activity under the program. For
this purpose, we exploit the richness of our data -- in particular the ability to track borrowers
across transactions matched to consumer credit bureau records using SSNs -- to construct
empirical proxies capturing consumer durable spending patterns. This data allows us to assess
the reduction in interest rates provided to borrowers who refinanced under HARP, as well as
track changes in their consumption activity around refinancing dates.
Finally, we conclude the first part of our analysis by assessing regional outcome variables such
as non-durable consumer spending, foreclosures, and house prices in regions more exposed to
the program. Here, we rely on zip code data, since we do not have more micro data for variables
like consumer credit card spending or house prices. The main challenge when attempting to infer
such a connection is that a national program such as HARP affects borrowers in all regions. We
address this challenge by exploiting regional heterogeneity in the share of loans that are eligible
for HARP. In particular, we obtain a measure of ex-ante exposure of a region to the program as
the regional (zip code) share of conforming mortgages with the high LTV ratios. We account for
general trends in outcomes during the program period by focusing on relative change in the
evolution of outcomes between regions with differential ex-ante exposure. This approach is
similar to one used in Mian and Sufi (2012) and Agarwal et al. (2012).
In the second part of our analysis, we investigate the role of intermediary competition on the
reach and effectiveness of HARP. The main obstacle in evaluating the potential role of limited
competition is to get an estimate of the counterfactual level in the absence of such frictions. We
circumvent this issue in three ways.
First, we construct the difference in interest rates on HARP refinances and regular conforming
refinances, both originated during the same period and made to borrowers of similar credit risk
(“HARP-conforming refi spread”). The regular conforming refinances represent creditworthy
borrowers with significant housing equity who could refinance outside of HARP. This group
serves as a natural counterfactual since the market for such loans was quite competitive and
remained fairly unobstructed throughout the period of study. In computing the spread, we also
take advantage of our detailed data that allows us to precisely account for variation in interest
rate spreads due to differences in loan credit risk by absorbing variation due to g-fees. In our
12
empirical tests we assess how the HARP-conforming refi spread varies with LTV of the loan and
across lenders, while accounting for the rich array of borrower, property, and loan
characteristics. Higher LTV loans should see higher spreads since they may be subject to more
limited competition due to a stronger incumbent advantage. Similarly, loans made by larger
lenders, who may operate as monopolists in several markets, might see larger spreads.
While potentially suggestive, the first set of tests may not fully address the concerns that such
differences may reflect other factors besides competitive frictions. To address such concerns, in
our second set of tests, we exploit variation within HARP borrowers. In particular, we relate the
terms of refinanced mortgages to the empirical measure of their bargaining power. As we discuss
in detail in Section V.C, in the presence of competitive frictions, all else equal, borrowers with
higher legacy rates -- i.e., rates on mortgages prior to HARP refinancing – should face larger
markups on their refinanced loans.
Finally, in our key test, we exploit the change in the program rules from January 2013 onwards
that lowered the put back risk of new lenders for loans originated previously by other lenders. As
discussed in Section II.C, this change effectively alleviated barriers to competition in the HARP
refinancing market. To the extent these barriers were quantitatively important, we expect to see a
meaningful reduction of the HARP-conforming refi spread after the program change and an
increase in refinancing activity under the program. In particular, we exploit a difference-in-
differences setting, analyzing the differential change in HARP interest rate (intensive margin)
and refinancing activity (extensive margin) relative to mortgage rates and refinancing activity of
regular conforming loans around the program change date.
IV. Program Effect
IV.A Descriptive Statistics
We start our analysis by presenting the characteristics of loans that were eligible to be refinanced
under HARP and contrasting these with similar loans that were not eligible for the program. As
we discussed in Section III.B, the latter group serves as a counterfactual for the loans that can be
potentially refinanced under the program. In particular, the treatment group consists of all GSE
FRM loans that would have been HARP eligible (that is GSE loans with current LTV greater
than 80%) and the control group consists of all FRM loans that are similar on all other
dimensions (such as FICO, LTV, interest rates, and loan balances) except that these are non-GSE
loans and therefore are ineligible for HARP.
14
This results in a sample of about 92,000 loans
equally split between treatment and control groups. We track the refinancing patterns of these
loans from April 2008 to December 2012.
14
The matching is one-to-one done based on FICO, current LTV (as of March 2008), interest rates, and loan using a
sample of more than 1 million of program eligible FRM loans and more than 200 thousand of non-GSE FRM loans.
13
Table 1 discusses the summary statistics of loans in the treatment and control groups over the
period before the program (i.e., from April 2008 to February 2009). As can be seen loans in the
two groups consist of borrowers with similar FICO scores (728 in the control group versus 727
in the treatment group), LTV ratios (95.6 in the control group versus 95.5 in the treatment
group)
15
and interest rates (6.62 in the control group versus 6.60 in the treatment group), and
similar outstanding loan balances ($186,525 in the control group and $183,614 in the treatment
group). Notably, in unreported tests we also confirm that these differences remain similar
throughout the pre-program period.
IV.B Micro Analysis: Refinancing Activity
Figure 1A presents the first set of results related to the program. Here we plot the quarterly
HARP refinancing rate in the treatment group. As can be observed, not surprisingly, there is a
gradual increase in the refinancing activity done under the program once the program starts in
2009:Q1.
16
The estimates presented in the figure suggest that the refinances done in the treatment
group under the program – i.e., the fraction of treatment loans refinanced under HARP every
quarter -- is about 1.6%. In terms of cumulative effect over the sample period being depicted in
the figure (until December 2012), we get about 25% of the eligible loans being refinanced under
the program. As per the US Treasury, up to 8 million loans were broadly eligible for HARP (see
Section II.B). Hence, our estimates applied to the entire stock of potentially eligible mortgages
imply that about 2 million loans were refinanced under the program by end of 2012 and about 3
million loans by end of 2014. This compares well with the 2.16 million loans refinanced under
HARP by 2012 and the 3.27 million loans reported by the US Treasury by December 2014.
The above analysis focuses only on HARP refinancing activity done on the GSE loans that were
eligible under the program. However, the overall effect of the program on refinancing activity
among the eligible loans also needs to account for any changes in refinancing activity that are
induced on refinances done outside the program. To do so we estimate the differential change in
total refinancing activity – i.e., refinancing done under HARP or otherwise -- in the treatment
loans relative to the control loans. In particular, we estimate a difference-in-difference estimation
around the program start date, reporting the estimates on differential change in refinancing
activity after conditioning on borrower and loan characteristics. As can be observed from Figure
1B, the overall effects on refinancing activity are similar to the direct treatment effect implied by
the program refinances. Moreover, we do not observe significant differential changes in the
refinancing rate between the treatment and control groups prior to the program, which yields
further support to the validity of our empirical design.
15
Our main data includes monthly information on the current LTV ratio of the GSE loan. For non-GSE loans we
compute the current LTV in each month using information on the loan’s outstanding balance and the appraised
property value (computed with use of the zip code level CoreLogic house price indices)
16
It is worth noting that take up rate was initially on the slower side and picked up from December 2011 once “high
LTV” loans (i.e., loans with LTV of greater than 125) were made eligible under the program.
14
In Table 2A, we present the overall estimates of the program effect also accounting for various
controls. In particular, in Columns (1) to (4) we use whether or not a loan refinances in a given
quarter as the dependent variable and estimate the change in this variable for treatment loans
relative to the control sample, accounting for a rich set of loan, borrower, and regional
characteristics. The key explanatory variable is the HARP Eligible × After Q1 2009 that captures
the differential change in refinancing rate of HARP eligible loans relative to confirming
refinances after the program start date (after Q1 2009). The estimation is performed on quarterly
data. As is evident, on average, we find that the treatment loans see an increase in refinancing
activity by about 1.4-1.7% every quarter, the estimates that are in line with the ones implied by
the direct program effect.
Taken together this evidence suggests that the program induced a significant increase in
refinancing activity, although a sizeable proportion of eligible loans did not refinance under the
program. Moreover, our findings suggest the program did not lead to a significant substitution of
refinances performed outside of the program with ones done under HARP. This is not surprising
once we note that both the treatment and control groups experienced very low refinancing rates
prior to the program, due to virtual shutdown of the refinancing market for loans with high LTV
ratios in the period before the program.
The analysis so far has focused on the extensive margin (i.e., new refinancing activity). In Table
2B we turn our attention to the intensive margin to assess the extent of savings received by
borrowers refinancing under HARP. Columns (1) and (2) present the results for a sample of more
than three hundred thousand loans that refinanced under the program where the dependent
variable is the difference between the interest rate in a given quarter and initial interest rate. The
variable, After HARP, takes the value of one in the quarters following the HARP refinancing date
and is zero otherwise. The results suggest that borrowers refinancing under the program obtained
a reduction of roughly 1.4 percentage points (140 basis points) in their mortgage rate. These
results are robust to including MSA fixed effects as well as a variety of borrower and loan level
controls. This is an economically significant reduction since the average pre-program mortgage
rate among the eligible sample is 6.6%. As Columns (3)-(4) of Table 2A indicate this mortgage
interest rate reduction implies about $884 in savings to the borrowers per quarter, translating into
about $7,000 in cumulative savings over the two-year period following the HARP refinancing
date. We obtain very similar results when performing this analysis for the subset of HARP loans
belonging to matched sample described in Table 1.
Overall, the results in Section IV.B suggest that the program led to a significant increase in
refinancing activity among eligible loans and, conditional on refinancing under the program
there were significant savings received by the borrowers.
15
IV.C Micro Analysis: Consumer Spending
We now assess changes in consumer spending patterns around the refinancing activity under the
program. In particular, we use individual consumer credit bureau records merged with the
dynamic mortgage performance data using each borrower’s SSN. This data allows us to observe
the current credit history of mortgage holders during the months preceding and following the
HARP refinancing date. Using this data, we can identify new auto financing transactions within
each borrower (new purchases financed with auto debt or new car leases), since such transactions
are usually accompanied by a significant discontinuous increase in a borrower’s outstanding auto
debt. It also allows us to measure a net dollar increase in new auto consumption associated with
such new auto financing transactions (e.g., a difference between a new and prior auto debt level
when new auto financing happens). As the vast majority of auto purchases in the U.S. are
financed with debt (up to 90% according to reports by CNW Marketing Research), we will use
these variables as empirical proxies capturing consumer durable spending patterns.
17
We first investigate whether borrowers change their durable spending patterns after HARP
refinancing. For that purpose we estimate a specification where the dependent variable takes the
value of one if a new auto financing transaction takes place within a given borrower in a given
quarter and is zero otherwise. In this specification we include a set of controls capturing
borrower, loan, and regional characteristics. Again, the key control is the After HARP, which
takes the value of one in the quarters following the HARP refinancing date. The results are
presented in Columns (5) and (6) of Table 2B. The sample includes more than three hundred
thousand loans that refinanced under HARP for which we have reliable auto balance data.
On average, there is an increase in the quarterly probability of new car purchases associated with
new auto financing after the HARP refinancing by about 0.8% implying 6.4% absolute increase
during the two years following the refinancing date. This amounts to an increase of about 10%
relative to the mean level probability of new auto financing prior to the HARP refinancing.
Columns (7) and (8) present the analogous regressions using the net dollar increase in auto debt
associated with new auto financing transactions (the difference between new and prior auto debt
in the quarter of new car purchase) as the dependent variable instead. The estimates suggest a net
increase in the auto consumption on the order of $185-$198 per quarter after HARP refinancing
amounting to about $1,600 over the period of two years following the refinancing date.
18
Combining this effect with the estimated savings due to HARP refinancing from Columns (3)
and (4) of Table 2B suggests that the borrowers allocate about 20-22% of the extra liquidity
generated by rate reductions to new car consumption. These findings are consistent with
contemporaneous work by Amromin, Di Maggio, and Kermani (2015) who find that borrowers
17
We identify new auto financing transactions if the borrower auto balance increases in a given month by at least
$2,000. Our results are robust to perturbations around these thresholds (e.g., $3000 or $5000 thresholds).
18
We obtain similar results if we restrict the analysis to subset of HARP loans in the matched sample of Table 1.
16
increase their spending after HARP refinancing date and explore channels driving the borrower
response. Notably, this elasticity is also qualitatively and quantitatively similar to that found in
Keys et al. (2014) and Di Maggio et al. (2014) who study the effects of mortgage rate reductions
due to rate resets among borrowers with adjustable-rate mortgages.
Next, we assess the dynamics associated with these spending patterns among refinancing
borrowers. In particular, we use the same specification as above but include a set of quarterly
time dummies that capture the three quarters preceding the HARP refinancing and the eight
quarters following the HARP refinancing date (instead of the After HARP dummy). Appendix
A.1 shows the estimated quarterly time effects from this specification along with 99%
confidence intervals. As we observe, the borrowers do not display differential changes in the
quarterly probability of new auto financing or net dollar increase in new auto financing prior to
the HARP refinancing. After the HARP refinancing, however, we observe a significant increase
in both the probability and net dollar amount of new auto financing. Notably, although the
largest effect occurs during the second quarter after the refinancing date, we observe a persistent
increase in the probability of buying a new car and the associated net dollar increase in financing
even two years after the refinancing date.
19
Figure 2A shows the cumulative increase in the net dollar amount of new auto financing implied
by the estimated quarterly time effects we discussed above. As can be observed, borrowers
display a net increase in new car consumption after the HARP refinancing, with the cumulative
effect of about $1,600 over the two-year period. This economic effect is in line with the average
effect reported in Table 2B.
Our estimates based on the analysis above suggest that borrowers who refinanced under the
program significantly increased their spending on durables (new cars). An obvious concern with
taking these effects as being induced by refinancing is that the decision to refinance under the
program could be endogenously determined along with other consumer activity (such as
spending on automobiles). Note that the effects we document are not only centered at the quarter
of HARP refinancing but present across the board after the refinancing. This suggests that these
findings may not just be reflecting endogenous timing by borrowers. Nevertheless, it is possible
that the refinancing was initiated by borrowers anticipating a change in auto spending well into
the future. We therefore conduct two more tests to shed more light on this issue.
First, we isolate our analysis to subset of high LTV borrowers who refinanced their loans in the
vicinity of the program announcement. In other words, the decision to refinance for the bulk of
19
It is useful to contrast these findings with Mian and Sufi (2012) who analyze the 2009 CARS program consisting
of government payments to car dealers for every older less fuel efficient vehicle traded in by consumers for a fuel
efficient one. They find that almost all of the additional purchases under the program were pulled forward from the
very near future. Our contrasting results might reflect the fact that the nature of stimulus is different under HARP:
refinancing generates persistent interest savings that can amount to tens of thousands dollars over time.
17
these borrowers was essentially determined by the supply of refinancing – made possible once
the program was initiated. We find similar effects for such borrowers.
In the second test, we assess changes in consumer spending among eligible loans relative to a
control group of borrowers – i.e., loans in the non-GSE market with characteristics that are
similar to loans eligible for HARP as described in Table 1. We find a significant relative increase
in the probability of new auto financing (about 0.18% per quarter) and auto consumption (about
$43 per quarter) among the eligible loans relative to the control group after the program
implementation date (March 2009).
20
Moreover, these estimates are very consistent with the
magnitudes of the estimates from Table 2A-B. In particular, the results in Table 2A imply that
the program induced about one-fourth of eligible borrowers to refinance their loans by December
2012. Hence, we would expect the estimated magnitude of the differential increase in auto
consumption among eligible borrowers relative to the control group to be roughly one-fourth of
the estimated values presented in Table 2B provided that these values reflect the true treatment
effect of HARP refinancing on new auto consumption. This is precisely what we find.
Next we explore heterogeneity in these findings. In particular, we assess if these effects vary
depending on the creditworthiness and wealth of borrowers. We sort borrowers into two groups
based on their housing wealth, as proxied by their current LTV ratios (Panel B) and based on
their creditworthiness (Panel C), as proxied by their FICO scores. Figure 2B and 2C shows the
estimated cumulative increase in net dollar amount of new auto financing in these groups. As is
evident, there is a significant heterogeneity in this effect across borrowers. During the first years
after HARP refinancing, borrowers with lower housing wealth (above median LTV ratios)
experience about a 20% larger increase in new auto consumption relative to borrowers with
below median LTV ratios. Less creditworthy borrowers (below median FICO scores) experience
about a 50% larger increase in new auto financing after HARP refinancing relative to a
corresponding increase for more creditworthy borrowers (above median FICO score). These
differential patterns are highly statistically significant and we also verify that they were not
driven by the differences in mortgage balances or differences in mortgage rate reductions across
these groups. In fact, after scaling new auto financing data by initial mortgage payments or the
dollar amount of reductions due to HARP refinancing, we obtain similar relative differences in
the cumulative patterns among these groups.
21
As we will show in Section V, while high LTV
20
Note that if borrowers would increase their consumption regardless of their access to HARP refinancing we would
expect to find no relative increase in auto financing among eligible loans relative to the control group after the
program implementation since such effects would be differentiated out.
21
These findings are consistent with life-cycle household finance models (Zeldes 1989; Carroll and Kimball 1996;
Carroll 1997) that predict a larger increase in consumption due to a positive income shock among borrowers with
lower wealth levels. Similarly, we find that borrowers with below median credit score – those more likely to face
credit constraints from the market -- increase their durable spending more after HARP refinancing relative to
borrowers with above median credit scores.
18
and low FICO borrowers exhibit strongest responses to refinancing, they were also the group
most adversely affected by competitive distortions embedded in HARP program design.
We conclude this section by investigating whether borrowers experiencing a larger reduction in
interest rate show a larger increase in durable (auto) consumer spending. In particular, we
estimate a specification where the dependent variable is a change in new auto financing after
HARP refinancing (relative to its level prior to refinancing). We include a set of controls
capturing borrower, loan, and regional characteristics as well as a variable, Rate reduction,
which measures the extent of mortgage rate reduction due to HARP (in percentage points).
Appendix A.2 shows these estimates. We find that borrowers experiencing a larger interest rate
reduction due to HARP refinancing display a larger increase in durable spending. In particular, a
one percentage point increase in rate reduction due to HARP is associated with a 0.12 to 0.2
percentage point increase in the quarterly probability of new auto financing and a net increase of
about $37 to $55 in the quarterly new auto financing. We note that these estimates do not capture
the base effect of refinancing under HARP (on average borrowers receive a reduction of around
140 basis points). Rather, they just reflect the relation between the extent of rate reduction and
new auto financing among those that refinanced their loans.
22
IV.D Regional Analysis: Refinancing Activity, Consumer Spending, Foreclosures and House Prices
We end our analysis of program effects by using regional data to assess the regional outcome
variables such as consumer spending, foreclosures, and house prices. We rely on zip code data,
since we do not have more micro data for variables like consumer credit card spending on non-
durables or house prices.
As we noted in Section III.B, our analysis exploits regional heterogeneity in the share of loans
that are eligible for HARP. We obtain a measure of ex-ante exposure of a region to the program,
Eligible Share, as the regional (zip code) share of conforming mortgages with LTV ratios greater
than 80% prior to the program implementation date. As discussed earlier, these loans are broadly
eligible for the program. We account for general trends in economic outcomes over the time
period of the study by focusing on the relative change in the evolution of economic outcomes
during the program period. Our identification assumption is that in the absence of the program,
and controlling for a host of observable risk characteristics, the economic outcomes in regions
(zip codes) with a larger share of eligible loans would have a similar evolution as those with a
lower share, up to a constant difference.
We start with more than 10,000 zip codes for which we can compute the share of program
eligible loans. Figure 3 shows the distribution of these zip codes in the data. There is a
22
In unreported results, we address potential endogeneity of rate reduction by instrumenting for the extent of rate
reduction with a level of market interest rate prevailing at the time of origination of legacy loan. This approach is
similar to the one in Section V.C where we instrument legacy rate with the Treasury rate. The results are similar.
19
significant variation in the share of eligible loans across zip codes ranging from just few percent
of all mortgages to more than 70% of loans being program eligible. We further confine our
analysis to zip codes that have at least 250 mortgages and for which we have reliable data on
outcome variables. This leaves us with a sample of about 3,400 zip codes.
We first verify that, consistent with our loan level evidence, zip codes with a larger share of
HARP eligible loans are indeed more likely to experience more HARP refinances and
consequently a larger mortgage interest rate reduction due to the program. We find that a one
percentage point absolute increase in the ex-ante share of eligible loans for HARP is associated
with an increase of about 0.24 percentage points in the fraction of loans that refinance under the
program (see Appendix A.3). Moreover, the “first stage” results in Columns (1)-(2) of Table 3A
show that there is a strong association between the share of loans that are ex ante eligible for
HARP and the average interest rate reduction due to the program in a zip code. The effects are
economically meaningful since we find that a one percentage point absolute increase in the ex-
ante share of eligible loans for HARP is associated with a reduction of about 0.38 basis points in
the average zip code mortgage interest rate
.
We next turn to the association between the average mortgage interest rate reduction due to
HARP and household spending on non-durables and durables. As Table 3B shows, consistent
with our borrower-level results from Section IV.B, zip codes with larger rate reductions due to
HARP experienced a relative increase in durable and non-durable consumer spending. In
particular, the estimates in Table 3B suggest that a relative reduction of about0.15% in mortgage
interest rate payments due to HARP in a zip code is associated with a differential increase of
about 0.13% in the credit card spending growth (Column 2) and an increase of about 0.2% in
auto purchase growth (Column 4). Figure 4A and 4B plot the average growth in credit card
spending and auto sales, respectively in more (above median Eligible Share) and less exposed
(below median Eligible Share) zip codes to the program. Consistent with the results in Table 3B,
these figures show that we observe a significant relative increase in spending growth in more
exposed zip codes after the program implementation. Overall, these findings, similar to our
micro evidence, suggest that following mortgage refinancing under the program, borrowers
significantly increase their durable and non-durable consumption.
23
We next investigate foreclosures and house price patterns across zip codes. The results in Table
3B indicate that regions more exposed to HARP experienced a relative improvement in the
housing market. The estimates in Table 3B imply that a relative reduction of about 0.15% in
average mortgage interest payments in a zip code is associated with a decline of about 0.01% in
23
In unreported results we also find that the regions more exposed to HARP experienced a relative reduction in
consumer debt delinquency rate (a reduction of 10 basis points in the average zip code mortgage rate due to HARP
being associated with a 0.4% reduction in the quarterly consumer debt delinquency rate). This finding suggests that
borrowers use some of the extra liquidity generated by mortgage rate refinancing to service and repay their debts.
20
the foreclosure rate (Column 6) and an increase of 0.13% in the house price growth rate (Column
8). Figure 4C reinforces these findings by showing that the areas more exposed to the program
experienced relative improvement in house prices after the program implementation.
Overall, these findings support the view that policies aimed at reducing mortgage rates can have
a meaningful impact on consumer spending and house prices. This evidence is consistent with
Agarwal et al. (2012), who find that mortgage modification programs, when used with sufficient
intensity, may improve a range of economic outcomes. It is also consistent with Keys et al.
(2014) and Di Maggio et al. (2014) who show that a sizable decline in mortgage payments on
adjustable rate mortgages induces a significant increase in new financing of durable consumption
and an overall improvement in household credit standing.
Taken together, our evidence in this section suggests that borrowers and regions exposed to the
program experienced a sizable increase in consumption. However, as we already established in
Section IV.B, a significant proportion of eligible borrowers did not participate in the program. In
the next section we explore whether competitive frictions in the refinancing market inhibited
program participation and the pass-through of lower interest rates to households, thereby
adversely affecting the program impact on consumer spending among eligible borrowers.
V. Role of Competition in Inhibiting Program Effectiveness
V.A Descriptive Statistics
As noted in Section III.B, we start our analysis by constructing a simple measure that captures
the difference in interest rates on mortgages refinanced though HARP and on conforming ones
that are refinanced outside of the program. This measure allow us to quantify the extent of pass-
through of lower interest rates to borrowers on HARP refinances relative to conforming ones
with higher rates being associated with a lower pass-through. Notably, in the absence of
competitive pressure, lenders may have incentive to charge borrowers higher rates on HARP
refinances because such mortgages can generally be sold for more in the secondary market.
24
We define the HARP-conforming refi spread as the difference between the interest rates on a
given HARP loan and the interest rate on a randomly assigned conforming mortgage in the data
– i.e., those with LTV of 80% -- originated during the same calendar month, in the same location
(MSA), and for a borrower with a similar FICO credit score at the time of refinancing.
25
As
noted, the latter group represents conforming mortgage contracts of creditworthy borrowers
24
Note that as GSEs fully insure credit risk of conforming loans the consideration that higher rates may lead to more
defaults and losses is relatively unimportant for the investors in the secondary market.
25
The conforming loan has a borrower FICO credit score within 20 points of the corresponding HARP loan. To
avoid concerns about interest rate term premiums, we restrict our attention to fixed-rate 30-year mortgages. Note
that the vast majority of HARP refinances during our sample period consist of this mortgage type.
21
carrying significant housing equity that could be refinanced outside of HARP, and for whom the
refinancing process remained fairly unobstructed throughout the crisis period.
Importantly, in computing this spread we take advantage of the unique feature of the GSE loans,
namely the fact that the credit risk of these loans to investors is fully insured by the GSEs. As we
discussed in Section II, GSEs charge predetermined fees (g-fees) for insurance of credit risk and
these fees are reflected in mortgage rates charged to the borrower. We use our precise data on
actual g-fees charged on each loan and remove this fee from interest rates charged to borrowers
on HARP and the benchmark conforming loans while computing the HARP-conforming refi
spread.
26
Consequently, the HARP-conforming refi spread should not reflect the relative
difference in credit risk between HARP and conforming mortgages, which includes differences
in credit scores and LTV ratios.
We focus on the time period 2009-2012, which broadly corresponds to the first three years of
program implementation. We later extend this analysis through the first half of 2013, a period
that featured changes in program rules as discussed in Section II.C. Notably, the vast majority of
refinancing activities under HARP occurred until mid-2013 (more than 80%), and thus are
covered by our analysis.
The first column of Table 4A confirms the existence of a sizeable mortgage rate differential
between HARP and conforming loan refinances. Over the course of 2009-2012, the HARP-
confirming refi spread (g-fee adjusted) averaged about 16 basis points suggesting that there
might be support for market power driven pricing of HARP loans. This markup is substantial
relative to mean interest rate savings on HARP refinances in our sample (about 140 basis points).
Next, we assess if the spread is related to HARP specific features, by constructing an alternative
benchmark interest rate spread between conforming refinances in which existing servicers may
also have some market power and purchase mortgages in which there is likely no such
advantage. With refinancing transactions, an existing relationship might confer some competitive
advantage to the existing servicer, whether through lower (re-) origination costs or less costly
solicitation. In our analysis we compare conforming refinancing transactions with conforming
purchase ones, both of which have an LTV of 80 – since the market for conforming LTV 80
mortgages – both refinancing and purchase – are very liquid and quite competitive.
Columns (3) and (5) in Table 4A show the resulting conforming refi-purchase spread, computed
as the difference between average interest rates on conforming refinances and purchase
transactions originated in the same month. The data confirm that market competitiveness kept a
tight lid on whatever advantages the existing servicer might have had in the conforming
26
As discussed in Section II, HARP pricing surcharges (LLPAs) took the form of upfront fees. However, these fees
were typically converted into periodic interest rate charges. We use the actual conversion that was used to adjust the
observed interest rates on HARP contracts.
22
refinancing market. In particular, during the period preceding HARP 2005-2009, the average
conforming refi-purchase spread was virtually zero (-0.55 basis points in Column (5)). Notably,
the spread remained below 3 basis points even during the crisis period (2009-2012)
corresponding to the time of HARP implementation in our study (Column (3)). These results are
consistent with the idea that the conforming refinancing market operated with a healthier level of
lender competitiveness than the HARP market.
Table 4A provides further evidence on the central feature of HARP discussed at length in
Section II.B, namely, the preferential treatment of existing servicers. Such asymmetry in
treatment may lead to an unusually high share of new loans refinanced through the existing (or
“same”) servicers under HARP. Consistent with this view, as shown in Column (1), among
HARP transactions conducted during 2009-2012, 54% of loans were refinanced by the existing
servicer. On the other hand, as shown in Column (3), during the same period only about 33% of
regular conforming loans were refinanced with the same servicer. This number is even lower
during the period preceding the crisis (2005-2009) when only about 1 in 5 conforming loans in
our sample were refinanced with their existing lender (Column (5)).
Our discussion in Section II.B focused on the likely relationship between loan LTV and the
degree of pricing power afforded to the existing servicer under HARP. Table 4B breaks down the
key summary statistics of HARP refinances by four LTV categories: LTV ranging from 80 to 90,
90 to 105, 105 to 125 and greater than 125. We observe that the HARP-conforming refi spread
increases substantially with LTV despite the fact that in computing this spread we removed
adjustment by GSEs (g-fees) that accounts for differential mortgage credit risk due to higher
LTV ratios. In particular, the spread for loans with LTV greater than 125 is nearly thrice (33.7
basis points) that for loans with LTV between 80 and 90. However, even for the loans closest to
the regular conforming LTV levels (those in the 80-90 LTV category), the average spread
persists at a non-negligible level of 11 basis points. These differences exists despite that the fact
that the borrower and loan characteristics in this subsample of 80-90 LTV loans (Column (1) in
Table 4B) are quite similar to those for the subsample of 80 LTV conforming refinancing loans
(Column (3) in Table 4A). Moreover, these differences exist even though we account for
variation in interest rates due to differences in credit risk by removing g-fees in computing this
spread. Overall, it is unlikely that the significant and positive HARP-conforming refi spread is
driven by differences in credit risk between HARP and conforming loans.
Table 4B also reveals that the fraction of loans refinanced by the same servicer also substantially
increases with mortgage LTV ratios: about 51% of loans with LTV ranging from 80-90% were
refinanced by the same servicer compared with 78% of loans with LTV higher than 125%.
Overall, these patterns are consistent with the notion that, relative to conforming refinances,
HARP provided more power to lenders.
23
V.B Cross-Sectional Variation in HARP-Conforming Refi Spread
We build on the analysis in Table 4 by systematically evaluating the determinants of the HARP-
conforming refi spread by estimating the loan-level specifications of the following form:
,
,
,
.
(1)
In these specifications the dependent variable,
,
,
, is the HARP-conforming refi
spread for the HARP loan refinanced at t by borrower i.
27
Since we want to assess how this
spread is related to the observable characteristics of loans such as LTV, we include a vector of
controls
,
that consists of a set of borrower and loan level observable characteristics all
measured at t, as well as any remaining differences in these characteristics between the HARP
loan and a corresponding randomly assigned conforming loan.
As a first step, we compare the mean values of the spread across the four LTV categories,
treating loans with LTVs between 80 and 90 as the omitted group. The results, shown in Column
(1) of Table 5 confirm our earlier finding of uniformly positive fee-adjusted HARP-conforming
refi spreads that are monotonically increasing in LTV. As we observe in Column (2), after
accounting for characteristics such as borrower FICO scores, MSA fixed effects, year-quarter
fixed effects for timing of refinancing transactions, and servicer fixed effects, we still find that
HARP-conforming refi spreads is monotonically increasing in LTV. In particular, HARP loans
with the highest LTV (LTV > 125) carry rates that are about 15.7 basis points higher than HARP
loans in the excluded category (80 < LTV ≤ 90), which amounts to more than a 140% increase in
the rate spread relative to that group. Since we remove the exact g-fee adjustment that GSEs
charge for insuring credit risk, it is unlikely that sizeable positive spread among HARP loans
with high LTV ratios reflects greater default risk of these mortgages.
We next assess the robustness of the descriptive statistics related to refinancing by incumbent
bank for loans financed under HARP relative to those in the conforming market in Columns (3)
and (4) of Table 5. Consistent with our prior evidence, we find that HARP loans with higher
LTV ratios are much more likely to be refinanced by the same servicer compared with
conforming refinances that serve as an excluded category (33% of conforming loans are
refinanced by the same servicer). In particular, the coefficient for LTV>125 in Column (4) of
Table 5 shows that, even after accounting for a variety of observable characteristics, HARP loans
with LTV ratios greater than 125 are more than twice as likely to be refinanced by the same
servicer compared with conforming refinances (33%+39%=72% versus 33%).
27
Recall that this spread is computed as the difference between the interest rate (net of its g-fee) on the loan by
borrower i that was refinanced under HARP at time t,
,
, and
,
, which is the monthly interest rate (net of its
g-fee) of a randomly assigned conforming mortgage with LTV of 80%, originated during the same calendar month,
in the same location (MSA), and to a borrower with a similar FICO credit score at the time of refinancing as the
HARP loan of borrower i.
24
The evidence above suggests that the mortgage refinancing market under HARP may not have
been fully competitive. It is natural to ask if these effects are prevalent uniformly across lenders
in our sample. To the extent that our controls capture the relevant borrower, loan, and regional
characteristics one would expect no significant difference in the HARP spread across the lenders
if the refinancing market is fully competitive. Figure 5A plots the HARP-conforming refi spread
for different lenders in our sample. The coefficients (with 95% confidence intervals) correspond
to lender fixed effects from the specification that corresponds to column (2) of Table 5.
As can be seen, controlling for other observables, there is a sizable and statistically significant
variation in the HARP-conforming refi spread across the lenders, ranging from as low as -20
basis points to about 20 basis points. Moreover, as shown in Figure 5B we find a strong positive
relation between the magnitude of the spread charged by a lender and the lender’s size (proxied
by log assets) with the correlation between the two series being 56%. Overall, our evidence is
consistent with the notion that lenders with high market power extract surplus from borrowers
for their own benefit.
V.C Variation in HARP-Conforming Refi Spread: Using Legacy Interest Rates
While potentially suggestive, there is a natural concern that our evidence in Section V.B may
reflect other omitted factors. To address this concern, in our second set of tests, we exploit
variation within HARP borrowers. In particular, we develop and test the conjecture that the
legacy interest rate on the mortgage prior to refinancing is also systematically related to the
degree of pass-through under HARP. Specifically, we expect that, in the presence of competitive
frictions, between two HARP loans that are identical on every dimension except for the legacy
interest rate, the loan with the higher legacy rate would obtain a higher post-refinancing rate.
To see this argument further, consider the simple example where two similar HARP eligible
borrowers (A and B) want to refinance their loans. The interest rates on their original loans (i.e.,
the legacy interest rates) are R
A,0
and R
B,0
, such that R
A,0
> R
B,0
. Given the assumed similarity of
borrower’s risk characteristics, the difference in interest rates faced by the borrowers could
reflect time variation in benchmark risk-free rates pinned down by the timing of when these
borrowers obtained their original loans. We further assume that a borrower needs to obtain a
reduction of the interest rate of at least Δ per year in order to refinance a loan (see Agarwal,
Driscoll, and Laibson (2013) for an optimal rule for Δ). An existing lender requires an interest
rate of R
even
to break even on the new mortgage, given the borrower’s risk characteristics. Since
by assumption, borrowers A and B have identical risk profiles at the time of refinancing, they
face the same breakeven rate. In a perfectly competitive market – proxied in our benchmark
analysis by the conforming refinancing market -- the same R
even
would apply to all lenders.
However, suppose a new lender must charge a rate premium, δ, to compensate for higher
underwriting costs due to higher put back risk (Section II). In this setting, what interest rates
25
could be obtained by A and B on their respective loans? The diagram below sketches out the
case for borrower A.
The shaded area represents the region of rates that satisfy the participation constraints of both the
borrower and the lender. Although the existing lender can potentially charge R
A,0
– Δ, the
presence of the outside option offered by the new lender effectively constrains the maximum rate
offered by the existing lender to be R
even
+ δ. This results in the existing lender being able to
charge an interest rate above the expected cost and extracting some surplus from the borrower.
In the case of borrower B, the existing loan has a lower interest rate, although it is still higher
than the rate on the newly refinanced loan. In the diagram below, a new lender is unable to offer
a rate R
even
+δ as this rate does not satisfy the borrower’s participation constraint. However,
because (R
B,0
– Δ) < (R
even
+ δ), the existing lender can still realize a markup over its expected
cost of funding the loan, albeit smaller than in the case of Borrower A. This difference in the
interest rates obtained by A and B occurs despite the fact that these borrowers have the same risk
characteristics.
The discussion above assumes that the current lender (servicer) knows the borrower’s
participation constraint as well as the costs of other lenders. In reality the incumbent lender may
be imperfectly informed about both the borrower’s participation constraint as well as the
refinancing cost structure of its competition (i.e., the incumbent lender may only know
distributions of these factors). In such a scenario, the decision of the incumbent lender regarding
the rate to be offered will reflect the trade-off between the expected profit in the case that the
offer is accepted versus the risk of losing the borrower either due to violating the borrower
participation constraint (offering insufficient reduction in rate) or due to other lenders being able
to offer a lower rate (if their refinancing cost proves to be low enough).
28
28
It is not difficult to construct a simple market equilibrium model featuring such a tradeoff. In such equilibrium,
consistent with the above discussion, the incumbent lender will still offer higher refinancing rates to borrowers with
higher legacy rates and to borrowers for whom the expected cost of refinancing by other lenders is larger (and hence
there is less competition). Moreover, fewer borrowers will end up refinancing their loans relative to the case when
the incumbent servicer would offer its zero profit rates. These borrowers would be the ones with required reduction
Interest rate
R
B
,
0
R
B
,
0
-Δ
R
even
+δ
R
even
Interest rate
R
A
,
0
R
A
,
0
-Δ
R
even
+δ
R
even
26
To investigate the above conjecture of the positive relationship between the legacy interest rate
and the interest rate on HARP refinances we estimate the following loan-level specifications:
,
,
,
.
(2)
As in specification (1) the dependent variable,
,
,
, is the HARP-conforming refi
spread of the loan by borrower i refinanced under HARP at time t and
,
is a vector of
observable borrower and loan characteristics associated with the loan. Because we are interested
in assessing the relation between this spread and the legacy interest rate, we include an additional
control variable,
, which reflects the interest rate on the loan before HARP refinancing
by borrower i. The coefficient γ measures the association between the loan HARP-conforming
refi spread and the legacy rate.
Table 6A shows the relation between the HARP-conforming refi spread and the legacy interest
rate. The panel preserves the setup of Table 5, while adding the mortgage rate prior to
refinancing as a control variable. Column (1) shows that borrowers with higher legacy interest
rates indeed face higher post-refinancing rates. This inference persists in Columns (2) through
(4), with the magnitude of the estimated effect even strengthening once we control for a rich set
of risk characteristics. Overall, we estimate a post-refinancing markup of about 9.6 basis points
per 100 basis points in the higher legacy rate, holding the key borrower and loan characteristics
fixed. This effect implies that the 100 basis points higher legacy rate is associated with a more
than 50% increase in the markup relative to its mean level and about a 7% reduction in interest
rate savings compared with average savings on HARP refinances. It is also worth noting that
borrowers with higher LTVs continue to suffer from higher HARP-conforming refi spreads even
when conditioning on the legacy rate.
It is possible that, despite accounting for variety of a borrower, loan, and regional characteristics,
our results may still be driven by some unobservable factors correlated with higher legacy rates.
To address this issue we rely on Section II.A where we discussed that mortgage pricing is tightly
linked to the benchmark Treasury rates. In particular, we instrument the legacy interest rate on a
mortgage with the 10-year U.S. Treasury yield prevailing at the time of origination of the legacy
mortgage to obtain variation in the borrower’s legacy rate that is exogenous to individual and
regional characteristics.
Table 6B verifies the relevance of this instrument through a regression of the legacy mortgage
rate on the 10-year Treasury rate in the month of origination. As can be seen there is indeed a
in rates higher than the one offered by the incumbent lender but lower than what is implied by a break-even rate. In
addition, some loans will also be refinanced by other non-incumbent lenders (i.e., mortgages they can refinance at
relatively low refinancing costs).
27
strong association – a 1% increase in the 10-year Treasury rate is associated with a highly
statistically significant 0.61% increase in the mortgage rate.
Column (2) introduces other controls that contain categories for LTV range, FICO, FICO
squared, MSA fixed effects, servicer fixed effects and, importantly, for the quarter-year fixed
effects corresponding to the date of origination of the legacy loan. Consequently, in the
specification in Column (2) of Table 6A we effectively exploit within-quarter variation in the
relationship between Treasury rates and legacy mortgage rates. As we observe from Column (2)
of Table 6B, in this much stringent setting, we still find that the 10-year Treasury is strongly
related to the mortgage legacy rate (with a 1% increase in the 10-year Treasury rate being
associated with a highly statistically significant 0.55% increase in the mortgage rate). Moreover,
the high R
2
values of 0.45 in the baseline model (Column 1) and 0.50 in the full model (Column
2) indicates that mortgage rates indeed track the “risk free” rate quite closely.
In the second stage of our analysis, we investigate the relationship between the HARP-
conforming refi spread and the predicted legacy mortgage rate from the first stage regression.
Columns (3) and (4) present results where we instrument for the legacy rate using the
specifications in Columns (1) and (2). Focusing on the full specification in Column (4), we see
that the estimated coefficient on the instrumented rate (10.91) is similar in magnitude to the
estimated coefficient on the observed rate (9.61) obtained through the OLS model of Table 5A.
Overall, this analysis reinforces our earlier findings that unobserved borrower or regional level
variation are not likely impacting the strong relationship between the HARP-conforming refi
spread and the legacy rate.
V.D Evidence from “difference in difference” around the program change
In our final key test, we establish a direct connection between changes in the degree of
competition in refinancing market and the program interest and take-out rates. We take
advantage of the change in the program rules regarding the assumed legal risk of servicers with
respect to loans they were refinancing. In particular, as discussed in Section II.C, from January
2013 onward the program rules were changed significantly, limiting the legal risk of a lender
who refinances a loan originated by another lender. Accordingly, we expect this policy change to
result in a more competitive HARP market and thus possibly lead to a reduction in the HARP-
conforming refi spread and an increased program participation rate.
To investigate this we estimate the loan-level specifications of the following form:
,
,
,
2013
.
(3)
We follow the same structure as specification (2) with a few key changes. First, we focus on a
new and extended time period, mid 2012 through the end of our sample period (mid-2013).
28
Second, we add a dummy variable, 2013 HARP Refi, which equals one for loans refinanced
under the HARP program in the first half of 2013 and zero for those refinanced in 2012. As
before, the vector of the control variables X captures the borrower and loan characteristics
measured at the time of HARP refinancing as well as the legacy interest rate on the loan. The key
coefficient of interest, , measures the change in the HARP-conforming refi spread around the
program change in January 2013.
Before doing the formal analysis we explore how the two groups of loans in the difference-in-
difference analysis compare on various observables before the program change. First, we explore
the pre-program change evolution of FICO and LTV of borrowers in HARP refinances relative
to conforming refinances (see Appendix A.4). We find that the difference in LTV ratios across
borrowers in the two groups remains constant in the pre-program period. In particular, the
average LTV ratio for HARP refinances consistently remains about 30% above that for
conforming refinances, with little relative change over time (recall that, by construction,
conforming refinances have LTV ratios equal to 80%). Likewise, we do not observe any
differential change in the borrower credit scores between the two groups in the pre-program
change period. Thus, the two groups of loans seem well situated for us to conduct our analysis.
Table 7 presents the results of our formal analysis.
29
As can be observed across Column (1) and
(2), is negative and significant, implying that there is a substantial reduction in the HARP-
conforming refi spread after the program change. A borrower who refinances under HARP
during the first half of 2013 enjoys, on average, a discount of around a 9.03 basis points relative
to an otherwise similar borrower who secured a HARP refinance during the second half of 2012.
Moreover, the estimated size of this effect is stable, ranging from -8.11 to -9.03 basis points,
across specifications that account for a plethora of borrower, loan, and regional level as well as
servicer fixed effects.
To confirm that the change in the spread occurs precisely around the program change, we
explore the timing of the effects documented in Table 7. In particular, we replace the 2013 HARP
Refi dummy in specification (3) with monthly dummies corresponding to the month in which a
given loan was refinanced under HARP (the excluded category is loans made in June 2012). This
specification allows us to investigate the monthly changes in the HARP-conforming refi spread
around the program change in January 2013. We present the results in Figure 6A. Two facts are
worth discussing. First, the HARP-conforming refi spread remains at a stable level in 2012. Note
that what we have plotted are demeaned spreads. The average spread during this period is about
27 basis points. Second, and more important, there is a sharp reduction in the spread by about 10
29
Note that the specification above already removes g-fees that accounts for borrower credit risk. Moreover, since
our dependent variable is measured relative to conforming refinances, any movements in the refinancing market that
also affect conforming refinances are differenced out in our specifications.
29
basis points precisely in January 2013 (more than 30% reduction in HARP-conforming refi
spread). This difference persists until the end of our sample period.
Finally, we investigate the impact of the program change on the refinancing rate of eligible
borrowers. For that purpose, in Column (3) and (4) of Table 7, we estimate a similar
specification to (3) but where now the dependent variable takes the value of 1 if the loan
refinances in a given month and is zero otherwise. The excluded category is loans that are
eligible for conforming refinances (loans with current LTV as of June 2012 less than 80). The
key coefficient of interest is the 2013 × HARP, which captures the change in the refinancing rate
of HARP eligible loans relative to conforming loans. Column (4) of Table 7 shows that, after the
program rule change, we observe a 0.12 percentage point increase in the refinancing rate among
eligible loans (about a 6% relative increase). Figure 6B verifies that the timing of this effect
coincides with the change in the program rules. We also verify that almost all of this increase can
be accounted for by an increase in the HARP refinances among eligible loans following the
change in the program rules.
Overall, the evidence in this section strongly supports the view that the an increase in
competition between servicers resulted in a meaningful decline in interest rates on HARP
refinances (intensive margin) and significant increase in the HARP refinancing rate (extensive
margin). We next assess the market wide effects of the competitive frictions.
V.E Assessing Market Wide Effects of the Competitive Frictions
Our evidence presented in Sections V.A-V.D suggests that the limited competition had
meaningfully reduced the pass-through of lower interest rates to consumers. The estimates we
obtain are substantial. We find that, on the intensive margin, borrowers would receive a 10 to 20
percentage points higher reduction in mortgage payments if HARP refinances were priced as
competitively as conforming refinances (after accounting for g-fees). Moreover, on the extensive
margin -- after applying the estimates from Section V.D to the stock of all eligible loans as of the
program start date -- the refinancing rate among eligible borrowers would be 9.6 percentage
points larger by December 2012 if the HARP-conforming spread was zero. In addition, this
effect on take up rates due to elimination of the HARP interest rate markup is about twice as
large (19.8 percentage points) among eligible borrowers with high LTV ratios (LTV>125).
30
30
In computing these effects we note that the estimates in Table 7 imply that a decline of about 1 basis point in the
HARP-conforming refi spread is associated with a 0.013% increase in the monthly refinancing rate among eligible
borrowers. Assuming that this estimate is applicable over the initial 45 months of the program (until December
2012) and taking into account that average HARP-confirming refi was about 16 basis points, implies that reducing
this spread to zero would increase the refinancing rate by about 9.6%. As the spread for loans with LTV>125 is
about 33 basis points (Table 4B) performing the same computation for this subset implies that reducing the spread to
zero for these loans would increase the refinancing rate by about 20%.
30
Notably, this stronger extensive margin effect of markups could be particularly detrimental for
consumption response obtained from the program. The reason is that, as we showed in Section
IV.C, these highly indebted borrowers display (conditional on refinancing), a larger increase in
spending from savings they receive from refinancing. Thus, the competitive frictions operating
through both extensive and intensive margin, may have meaningfully reduced the program
impact on consumer spending among eligible borrowers (especially in its first few years).
To shed further light on the plausibility of these estimates we perform two further tests. First, we
focus on a representative sample of GSE loans (more than a million) that were eligible for the
program as of its implementation date in March 2009. Within this sample we perform a simple
quantitative assessment of the impact of the estimated HARP markups. For that purpose, we
assume that borrowers need to obtain an interest rate reduction of at least 100 basis points to
refinance their loans. This simple assumption is motivated by various factors, such as borrowers’
costs in refinancing their loans and the option value of waiting for further declines in interest
rates (see Agarwal et al. 2013). Performing this simple exercise we find that the refinancing rate
among eligible borrowers would be between 10% (for lower LTV loans) and 20% (for high LTV
loans) higher in the absence of HARP markups (see Appendix A.5). Notably the magnitude of
these effects is similar to the one implied by extensive margin estimates from Section V.D.
Second, we explore whether the program was less effective in regions with a larger concentration
of servicers with higher HARP-conforming refi spreads – i.e., those that charge borrowers higher
rates on loans refinanced under the program. Recall that our analysis in Section V.B indicates a
significant servicer level variation in the HARP-conforming refi spread that is not accounted for
by borrower, loan, and regional level characteristics. To conduct this analysis we classify the top
quartile of servicers with the highest estimated fixed effects displayed in Figure 5A. These high
cost servicers account for more than 60% of loans in our data. Consequently, we compute the zip
code level Eligible and High Cost Servicer Share as a fraction of loans in a zip code that both are
program eligible and are serviced by high cost lenders.
We find that the areas where a larger share of eligible loans is handled by high cost servicers do
experience significantly lower rate reduction due to HARP. In particular, we note that our
estimates (see Appendix A.3) suggest that on average the pass-through of lower interest rates to
consumer through program refinances would be about 35% lower in a zip code where all eligible
loans are serviced by high cost servicers compared with a zip code where all eligible loans are
serviced by low cost servicers – after controlling for the share of HARP eligible borrowers in a
zip code. Notably, both the intensive and extensive margin play an important role in explaining
this effect, since we also find that fewer HARP eligible borrowers (about 17% less) would
refinance their loans in the areas where all eligible loans were handled by high cost servicers.
VI. Conclusion
31
Our findings suggest that significant number of eligible borrowers did not take advantage of the
program. While certainly the borrower specific factors or other institutional frictions (e.g., like
servicer capacity constraints) may help account for this muted response, our paper finds that
limits to competition in refinancing market can also help explain part of this shortfall. Moreover,
by adversely altering refinancing activity – the take up rate as well as the pass through --
competitive frictions may have significantly reduced the program effect on consumption of
eligible households, especially indebted households who generally have a higher propensity to
spend from additional liquidity (see Mian, Rao and Sufi 2013).
31
Thus, our evidence suggests
that provisions limiting the competitive advantage of incumbent banks with respect to their
existing borrowers should be an active consideration when designing stabilization polices such as
HARP. This insight would also apply to other polices whose implementation depends on the
intermediaries that may have some incumbency advantage with respect to targeted agents.
Our results also speak to HARP’s impact on redistribution and the overall consumption response
in the economy. As Beraja, Fuster, Hurst and Vavra (2015) note, prior to HARP, refinancing was
only available to more creditworthy borrowers with lower LTV ratios, which could exacerbate
regional economic heterogeneity. Although we cannot quantify the overall GE effects of the
program that might include the impact of the program on profits of mortgage investors and their
consumption, our results suggest that less creditworthy and more indebted borrowers
significantly increased their spending following refinancing. To the extent that such borrowers
have the largest marginal propensity to consume, allowing them to refinance under the program
could increase overall consumption and alleviate the regional dispersion in economic outcomes
(see Auclert 2015 for redistributional effects of lower rates for aggregate consumption).
Our findings also have implications for the debate regarding optimal mortgage contract design
(see Eberly and Krishnamurthy 2014), highlighting potential benefits of adjustable rate
mortgages (ARMs). In particular, by automatically reducing mortgage rates when market rates
are low, ARMs can help alleviate frictions due to the limited competition in the loan refinancing
market. Moreover, as ARMs can allow quick refinancing of borrowers regardless of the extent of
their housing equity or creditworthiness, such contracts may reduce the need for large-scale
refinancing programs like HARP, which, as we show, can face implementation hurdles. There
are also additional benefits of ARMs that might be useful to discuss in our context.
32
Of course,
31
Because these indebted households face higher default risk and have larger propensity to spend from additional
liquidity (see Keys et al 2014), they are the key target of stabilization polices such as HARP.
32
By automatically reducing mortgage rates, ARMs may help alleviate the barriers to loan renegotiation due to
securitization (Piskorski et al. 2010; Agarwal et al. 2011) and lender concerns regarding borrowers’ strategic
behavior (Mayer et al. 2014). In addition, as ARM contracts do not require the active participation of borrowers in
the process of rate reduction, they can help alleviate the adverse effects of borrower inertia and inattention on
mortgage refinancing (see Keys, Pope, and Pope 2014 and Andersen et al. 2014 for the recent evidence on these
factors). See also Piskorski and Tchistyi (2010) who highlight the benefits of ARMs for less creditworthy borrowers
in an optimal dynamic contracting framework with costly default.
32
such benefits need to be carefully weighed against the potential adverse costs of ARMs. We
leave this issue for further research.
Finally, we note that our analysis using conforming market pricing as a benchmark does not
imply that the conforming refinancing market was fully competitive. In fact, recent evidence by
Scharfstein and Sunderam (2014) suggests that there are also significant frictions limiting
competition in the regular conforming refinancing market. Their findings suggest that our
estimates are, if anything, a lower bound on the overall effects of importance of competition for
program implementation.
References
Aaronson, Daniel, Sumit Agarwal, and Eric French, 2012, The Spending and Debt Response to
Minimum Wage Hikes, American Economic Review 102, 3111-39.
Agarwal, Sumit, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, Tomasz
Piskorski, and Amit Seru, 2012, Policy Intervention in Debt Renegotiation: Evidence from Home
Affordable Modification Program, NBER Working Paper 18311.
Agarwal, Sumit, John Driscoll, and David Laibson, 2013, Optimal Mortgage Refinancing: A
Closed Form Solution, Journal of Money, Credit, and Banking 45, 591-622.
Agarwal, Sumit, Richard J. Rosen, and Vincent Yao, 2014, Why Do Borrowers Make Mortgage
Refinancing Mistakes?, working paper, Federal Reserve Bank of Chicago.
Agarwal, Sumit, Chunlin Liu, and Nicholas Souleles, 2007, The Reaction of Consumption and
Debt to Tax Rebates: Evidence from the Consumer Credit Data, Journal of Political Economy
115, 986-1019.
Agarwal, Sumit, and Wenlan Qian, 2014, Consumption and Debt Response to Unanticipated
Income Shock: Evidence from a Natural Experiment in Singapore, American Economic Review
104: 4205-30.
Agarwal, Sumit, Souphala Chomsisengphet, Neil Mahoney, and Johannes Strobel, 2015, Do
Banks Pass Through Credit Expansions? The Marginal Profitability of Consumer Lending during
the Great Recession, working paper.
Amromin, Gene, Marco Di Maggio, and Amir Kermani, 2015, The Real Effects of HARP,
working paper.
Andersen, Steffen, John Y. Campbell, Kasper Meisner Nielsen, and Tarun Ramadorai, 2014,
Inattention and Inertia in Household Finance: Evidence from the Danish Mortgage Market,
working paper.
Auclert, Adrien, 2015, Monetary Policy and the Redistribution Channel, working paper.
Beraja, Martin, Andreas Fuster, Erik Hurst, Joseph Vavra, 2015, Regional Heterogeneity and
Monetary Policy, working paper.
Browning, Martin, and M. Dolores Collado, 2001, The Response of Expenditures to Anticipated
Income Changes: Panel Data Estimates, American Economic Review 91, 681-92.
33
Campbell, John Y., and João Cocco, 2003, Household Risk Management and Optimal Mortgage
Choice, Quarterly Journal of Economics 118: 1449–1494.
Carroll, Chris. D, and Miles S. Kimball, 1996, On the Concavity of the Consumption Function,
Econometrica 64, 981-992.
Carroll, Chris D., 1997, Buffer-Stock Saving and the Life Cycle/Permanent Income Hypothesis,
Quarterly Journal of Economics 112, 1-54.
Charles Kevin K., Erik Hurst, Matthew J. Notowidigdo, 2013, Manufacturing Decline, Housing
Booms, and Non-Employment, NBER Working Paper No. 18949.
Chen, Hui, Michael Michaux, and Nick Roussanov, 2014, Houses as ATMs? Mortgage
Refinancing and Macroeconomic Uncertainty, working paper.
Di Maggio, Marco, Amir Kermani, and Rodney Ramcharan, 2014, Monetary Pass-Through:
Household Consumption and Voluntary Deleveraging, Working Paper.
Eberly, Janice, and Arvind Krishnamurthy, 2014, Efficient Credit Polices in a Housing Crisis,
Fall 2014 Brookings Panel on Economic Activity.
Favilukis, Jack, Sydney Ludvingson, and Stijn Van Nieuwerburgh, 2013, The Macroeconomic
Effects of Housing Wealth, Housing Finance, and Limited Risk Sharing in General Equilibrium,
working paper.
Gelman, Michael, Shachar Kariv, Matthew D. Shapiro, Dan Silverman, and Steven Tadelis.
2014. “Harnessing Naturally Occurring Data to Measure Income and Spending Accurately in
Real Time.” Science 345(6193), 212-215
Gross, David B., and Nicholas S. Souleles, 2002. Do Liquidity Constraints and Interest Rates
Matter for Consumer Behavior? Evidence from Credit Card Data, Quarterly Journal of
Economics 117, 149-185.
Hubbard Glenn, and Christopher Mayer, 2009, The Mortgage Market Meltdown and House
Prices, The B.E. Journal of Economic Analysis & Policy 9, Issue 3 (Symposium), Article 8.
Hsu, Joanne W., David A. Matsa, and Brian T. Melzer, 2014, Positive Externalities of Social
Insurance: Unemployment Insurance and Consumer Credit, NBER Working Paper 20353.
Hurst, Erik and Frank Stafford, 2004, Home is Where the Equity is: Mortgage Refinancing and
Household Consumption, Journal of Money Credit and Banking 36, 985-1014.
Johnson David S., Jonathan A. Parker., Nicholas S. Souleles, 2006. Household Expenditure and
the Income Tax Rebates of 2001. American Economic Review 96: 1589-1610.
Jappelli, Tullio, Jorn-Steffen Pischke, and Nicholas S. Souleles, 1998, Testing for Liquidity
Constraints in Euler Equations with Complementary Data Sources, Review of Economics and
Statistics80,251‐262
Johnson, Eric, Stephan Meier, and Olivier Toubia, 2015, Leaving Money on the Kitchen Table:
Exploring sluggish mortgage refinancing using administrative data, surveys, and field
experiments, working paper.
Keys, Benjamin J., Tanmoy Mukherjee, Amit Seru, and Vikrant Vig, 2010, Did Securitization
Lead to Lax Screening: Evidence from Subprime Loans, Quarterly Journal of Economics 125,
307-362.
34
Keys, Benjamin J., Tomasz Piskorski, Amit Seru, and Vincent Yao, 2014, Mortgage Rates,
Household Balance Sheets, and Real Economy, NBER Working Paper No. 20561.
Keys, Benjamin J., Devin G. Pope, and Jaren C. Pope, 2014, Failure to Refinance, NBER
Working Paper No. 20401.
Koijen, Ralph, Otto van Hemert, and Stijn Van Nieuwerburgh, 2009, Mortgage Timing, Journal
of Financial Economics 93, 292-324.
Mayer, Christopher, Karen Pence, and Shane M. Sherlund. 2009, The Rise in Mortgage Defaults,
Journal of Economic Perspectives 23, 27-50
Mayer, Christopher, Edward Morrison, Tomasz Piskorski, and Arpit Gupta, 2014, Mortgage
Modification and Strategic Behavior: Evidence from a Legal Settlement with Countrywide,
American Economic Review 104, 2830-285.
Melzer, Brian, 2014, Mortgage Debt Overhang: Reduced Investment by Homeowners at Risk of
Default, Journal of Finance, forthcoming.
Mian, Atif, and Amir Sufi, 2012, The Effects of Fiscal Stimulus: Evidence from the 2009 'Cash
for Clunkers' Program, Quarterly Journal of Economics 127: 1107-1142.
Mian, Atif, Kamalesh Rao, and Amir Sufi, 2013, Household Balance Sheets, Consumption, and
the Economic Slump, forthcoming in the Quarterly Journal of Economics.
Neumark, D., Sharpe, S., 1992. Market Structure and the Nature of Price Rigidity: Evidence
from the Market for Consumer Deposits, Quarterly Journal of Economics 107, 657-80.
Parker, Jonathan A. 1999. “The Reaction of Household Consumption to Predictable Changes in
Social Security Taxes.” American Economic Review 89: 959-973
Parker, Jonathan A., Nicholas S. Souleles, David S. Johnson, and Robert McClelland. 2013,
Consumer Spending and the Economic Stimulus Payments of 2008, American Economic Review
103, 2530–2553.
Piskorski, Tomasz, and Alexei Tchistyi, 2010, Optimal Mortgage Design, Review of Financial
Studies 23, 3098-3140.
Rotemberg, Julio J, and Garth Saloner, 1987, The Relative Rigidity of Monopoly Pricing, American
Economic Review 7, 917-26.
Scharfstein, David, and Adi Sunderam, 2014, MarketPowerinMortgageLendingandthe
TransmissionofMonetaryPolicy,Working paper.
Shapiro, Matthew D., and Joel Slemrod, 1995, Consumer Response to the Timing of Income:
Evidence from a Change in Tax Withholding, American Economic Review, 85: 274-283.
Shapiro, M.D., and Slemrod, J., 2003, “Consumer Response to Tax Rebates,” American
Economic Review, 93, 381-396.
Souleles, Nicholas S., 1999, Response of Household Consumption to Income Tax
Refunds, American Economic Review 89, 947-958.
Stephens, Melvin Jr, 2008, The Consumption Response to Predictable Changes in Discretionary
Income: Evidence from the Repayment of Vehicle Loans, Review of Economics and Statistics
90, 241-52.
35
Stroebel Johannes, and Joseph Vavra, 2014, House Prices, Local Demand, and Retail Prices,
NBER Working Paper 20710.
Zeldes, Stephen P., 1989, Optimal Consumption with Stochastic Income: Deviations from
Certainty Equivalence, Quarterly Journal of Economics 107, 275-298.
36
Table 1:
Summary Statistics for HARP and Non-HARP Eligible Loans
This table presents summary statistics of key variables in the pre-HARP period (2008: Q2 to 2009: Q1) in the treatment and control groups. The treatment group
consists of GSE 30-year fixed-rate mortgages that have current LTV ratios greater than 80 as of March 2008 (one year prior to the program). The control group
consists of a sample of full documentation prime non-GSE 30-year fixed-rate mortgages (privately securitized) that have current LTV ratios greater than 80 as of
March 2008. Since, these loans were not sold to GSEs they are not eligible for HARP. These loans were further matched based on FICO credit scores of
borrowers, current LTV ratios, interest rates, and loan amounts. Data Source: Large secondary-market participant and BlackBox Logic.
Treatment
(HARP Eligible)
Control
(Non-HARP Eligible)
Mean
(1)
S.D
(2)
Mean
(3)
S.D
(4)
LTV 95.5 5.2 95.6 5.8
FICO 727.7 46.4 728.4 43.7
Interest Rate 6.60 0.64 6.62 0.66
Balance 183,614 91,718 186,525 110,104
Number of Loans 46,154 46,154
37
Table 2:
Borrower-Level Evidence: HARP, Refinancing Rate, Mortgage Payments, and Durable Spending (New Auto Financing)
Panel A of this table presents OLS estimates from regressions that track whether or not a loan refinances around the program implementation (Q2 2008 till Q4
2012). The dependent variable takes the value of one in the quarter a given loan refinances and is zero otherwise (the refinanced loans exit the estimation
sample). The variable, HARP Eligible, takes the value of one if a loan belongs to the treatment group as defined in Table 1 and is zero otherwise. The variable,
After Q1 2009, takes the value of one for the quarters after Q1 2009 and is zero otherwise. Column (1) presents the basic specification with no other controls but
a constant term, HARP Eligible dummy, After Q1 2009 dummy, and the interaction term of these two variables (HARP Eligible) × After Q1 2009). Column (2)
adds borrower controls that include variables such as FICO credit score, LTV, interest rates. Column (3) adds the fixed effects for the location (MSA) of the
property (MSA FEs). Column (4) clusters standard errors at the MSA level. The estimation sample consists of a matched set of treatment and control loans (as
defined in Table 1). The estimates in Column (1)-(4) of Panel A are in percentage terms and the estimation is performed on quarterly data. Panel B presents OLS
estimates from regression where the dependent variable is the current interest (Column 1 and 2), the quarterly mortgage interest rate payments (Column 3 and 4),
the variable that takes value of one if the new auto financing takes place in a given quarter and is zero otherwise (Column 5 and 6), and the net amount of new
auto financing in dollars (Colum 7 and 8) (the difference between the new and prior auto debt in the quarter in which new financing takes place). The variable,
After HARP, takes value of one in the quarters after HARP refinancing rate and is zero otherwise. The sample includes all the loans that refinanced under HARP
for which we have reliable auto balance data. In Column (2), (4), (6), (8) of Panel B the standard errors are clustered at MSA level and the estimates in Columns
(1)-(2) and (5)-(6) are expressed in percentage terms. The Year-Quarter FEs correspond to the quarter-year fixed effects for the date of HARP refinancing.
Standard errors are included in the parentheses.
Panel A: HARP and the refinancing rate
Panel B: Mortgage payments and durable spending (new auto financing) after HARP refinancing
Mortgage rate Mortgage payments
Probability of
new auto financing
Net amount of new
auto financing
(1) (2) (3) (4) (5) (6) (7) (8)
After HARP -1.40
(0.00)
-1.42
(0.07)
-884.11
(0.24)
-884.67
(37.93)
0.78
(0.03)
0.84
(0.04)
185.08
(5.63)
198.88
(9.38)
Borrower Controls No Yes No Yes No Yes No Yes
Year-Quarter FEs No Yes No Yes No Yes No Yes
MSA FEs No Yes No Yes No Yes No Yes
Observations 4,125,726 4,125,726 4,125,726 4,125,726 4,125,726 4,125,726 4,125,726 4,125,726
Adjusted R-Square 0.617 0.868 0.515 0.667 0.001 0.002 0.001 0.002
(1) (2) (3) (4)
(HARP Eligible) × After Q1 2009 1.73
(0.05)
1.35
(0.07)
1.37
(0.07)
1.37
(0.11)
Borrower Controls No Yes Yes Yes
MSA FEs No No Yes Yes
Observations 1,372,731 1,372,731
1,372,731
1,372,731
Adjusted R-Square 0.001 0.003
0.01
0.01
38
Table 3:
Regional Evidence: Consumer Spending, Foreclosures, and House Prices and Zip Code Exposure to HARP
This table examines the relation between regional (zip code level) consumer spending, foreclosures, and house prices and the average mortgage rate reduction in
a zip code due to HARP (in basis points) instrumented with the pre-program share of loans in a zip code that are eligible for HARP. The pre-program program
eligible share, Eligible Share, is the fraction of outstanding first-lien mortgage loans in a zip code that are conforming and have current LTV ratios greater than
80 prior to the program implementation. Column (1) presents the first stage specification without controls, in which the average mortgage interest rate reduction
due to HARP, Rate Reduction due to HARP, is instrumented with the share of program eligible loans. Column (2) repeats the first stage, but includes a series of
controls including the zip code average FICO credit score, LTV ratio, interest rate on mortgages along with the average zip code house price growth over the
prior five years and state fixed effects. Panel B shows the corresponding second stage estimates results for the change in the quarterly credit card spending
growth rate (Column 1 and 2), the auto purchase growth rate (Column 3 and 4), the foreclosure rate (Column 5 and 6), and the house price growth rate (Column 7
and 8), all computed as the average of the respective value during the program less its pre-program level. The analysis is based on a sample of 3,443 zip codes.
The estimation period is 2008:Q1 through 2013:Q2. Standard errors are included in parentheses.
Panel A: Average mortgage interest rate reduction due to HARP in a zip code (First Stage)
(1) (2)
Eligible Share
31.4
(0.007)
38.0
(0.010)
Zip Code Controls No Yes
State FEs No Yes
Adj. R-squared 0.47 0.71
Panel B: Instrumented “Interest rate reduction due to HARP” and consumer spending (credit card and auto), foreclosures, and house prices (Second Stage)
Credit card spending Auto purchase growth Foreclosure rate House price growth
(1) (2) (3) (4) (5) (6) (7) (8)
Rate Reduction due to HARP
0.119
(0.028)
0.129
(0.036)
0.394
(0.019)
0.201
(0.021)
-0.010
(0.002)
-0.025
(0.002)
0.353
(0.015)
0.134
(0.008)
Zip Code Controls No Yes No Yes No Yes No Yes
State FEs No Yes No Yes No Yes No Yes
Adj. R-squared 0.01 0.03 0.06 0.42 0.01 0.64 0.07 0.83
39
Table 4:
Summary Statistics for HARP and Conforming Refinances
This table presents the summary statistics for mortgage loans that were refinanced under the Home Affordable Refinance Program (HARP) during 2009-2012
period (Column 1 and 2) alongside with conforming refinances originated during 2009-2012 period (Column 3 and 4) and 2005-2009 period (Column 5 and 6).
The sample consists of more than 800,000 of HARP and conforming refinances. The variable HARP-conforming refi spread (in basis points) is computed as the
difference between the interest rate on a given HARP refinanced loan and the mean interest rate for conforming refinances with a loan-to-value ratio (LTV) equal
to 80 percent originated in the same month. This spread is guarantee-fee adjusted by subtracting guarantee fees from the HARP and conforming refinance rates
before computing the spread. The variable Conforming refi-purchase spread (in basis points) is computed as the difference between the interest rate on a given
conforming refinanced loan and the mean interest rate for purchase loans with an 80 percent LTV ratio originated in the same month. The Same servicer refi
dummy takes a value of one if the servicer is the same before and after a refinancing, otherwise it is zero. The table also present summary statistics for other key
variables, including the LTV ratio (in percentage terms) at the time of refinancing, FICO credit score of the borrower at the time of refinancing, interest rate (in
percentage terms) on a loan before refinancing (Previous rate), interest rate on a loan after refinancing (Rate after refinancing), and balance of a loan at the time
of refinancing (in thousands of dollars). Panel A presents the statistics of the full sample, while Panel B presents the mean values for HARP refinances separated
in the four LTV ranges (as of the time of refinancing).
Panel A: HARP and conforming refinances (All Sample)
HARP refinances
2009-2012
Conforming refinances
2009-2012
Conforming refinances
2005-2009
Mean
(1)
S.D
(2)
Mean
(3)
S.D
(4)
Mean
(5)
S.D
(6)
LTV 99.74 22.38 80 0 80 0
FICO 749.75 44.31 759.82 40.81 737.06 47.89
Balance 242.20 96.49 264.2 127.1 234.5 92.1
Previous rate 6.07 0.58 5.69 0.74 6.30 0.83
Rate after refinancing 4.67 0.52 4.55 0.51 5.75 0.69
HARP-conforming refi spread 16.07 37.54 - - - -
Conforming refi-purchase spread
- - 2.83 29.09 -0.55 32.11
Same servicer refi
0.54 0.50 0.33 0.47 0.28 0.45
40
Table 4 [continued]:
Panel B: HARP refinances by LTV ratio at the time of refinancing
80 < LTV ≤ 90 90 < LTV ≤ 105 105 < LTV ≤ 125 LTV > 125
(1) (2) (3) (4)
FICO 752.91 749.73 745.38 741.49
Balance 248.09 245.06 230.69 218.48
HARP-conforming refi spread 11.33 13.46 27.06 33.77
Same Servicer 0.51 0.50 0.62 0.78
41
Table 5:
HARP-Conforming Refi Spread, Same Servicer Refinances and the LTV Ratio
Column (1) and (2) of this table presents OLS regression results for a specification with the HARP-conforming refi spread (in basis points) as the dependent
variable. The set of control variables includes the three dummy variables that indicate the HARP loan LTV range of (90, 105], (105, 125] and >125. The loans
refinanced through HARP with the LTV range (80, 90] serve as the excluded category and have an average HARP-conforming refi spread of 11.33 basis points.
In Columns (3) and (4) the dependent variable is the dummy variable, Same Servicer, which takes the value of one if the loan is refinanced by the lender
servicing the legacy mortgage and is zero otherwise. The set of control variables includes the four dummy variables that indicate the HARP loan LTV range of
(80,90], (90, 105], (105, 125] and >125. The conforming refinances with LTV equal to 80 serve as the excluded category (33% of these loans are refinanced by
the same servicer). Column (2) and (4) add borrower controls including the current FICO score (and its square) of the borrower and the metropolitan statistical
area (MSA) fixed effects corresponding to the location of the property, time fixed effects capturing the quarter/year time during which the loan was refinanced
(Year-Quarter FEs), and the fixed effects corresponding to the identity of the lender refinancing the loan (Servicer FEs). Standard errors are included in the
parentheses.
Dependent variable:
HARP-conforming refi spread
Dependent variable:
Same servicer refinance
(1) (2) (3) (4)
80<LTV≤90 - - 0.19
(0.00)
0.18
(0.05)
90 < LTV ≤ 105 2.13
(0.13)
1.48
(0.44)
0.19
(0.00)
0.19
(0.06)
105 < LTV ≤ 125 15.73
(0.18)
10.98
(2.54)
0.31
(0.00)
0.29
(0.08)
LTV > 125 22.43
(0.22)
15.77
(3.06)
0.46
(0.00)
0.39
(0.10)
Borrower Controls
No
Yes No Yes
MSA FEs
No
Yes No Yes
Year-Quarter FEs
No
Yes No Yes
Servicer FEs
No
Yes No Yes
Observations 414,172 414,172 828,344 828,344
Adjusted R-squared 0.04 0.10 0.06 0.36
42
Table 6:
HARP-Conforming Refi Spread and the Previous Interest Rate
Panel A of this table presents OLS regression results for a specification with the guarantee fee adjusted HARP-conforming refi spread as the dependent variable and
the interest rate of a loan prior to refinancing as a control variable. Column (1) presents the basic specification with no controls but the previous interest rate. Column
(2) adds borrower controls including the current FICO score (and its square) of the borrower and the metropolitan statistical area (MSA) fixed effects corresponding to
the location of the property. Column (3) adds time fixed effects capturing the quarter/year time during which the loan was refinanced. Column (4) adds the fixed
effects corresponding to the identity of the lender refinancing the loan and clusters standard errors at the servicer level. Panel B of this table presents the results of a 2-
stage least squares regression in which the mortgage rate of a loan prior to refinancing, Previous rate, is instrumented with the average 10-year Treasury rate
corresponding to the month of origination of legacy mortgage. The first two columns show the first-stage results in which the previous mortgage rate is regressed on
the 10-year Treasury rate. Column (1) corresponds to a basic specification without additional controls. Column (2) introduces Other controls, including the dummy
variables for LTV ranges as in Tables 2 and 3, FICO, FICO squared, year-quarter fixed effects corresponding to the origination of legacy and HARP loan (Year-
Quarter FEs), MSA fixed effects, and the servicer fixed effects corresponding to the identity of the servicer handling the loan (Servicer FEs). Columns (3)-(4) present
the analogues second-stage results, whereby the dependent variable is the guarantee fee adjusted HARP-conforming refi spread and the control variable is the previous
mortgage rate instrumented with 10-year Treasury. In Panel B the standard errors are clustered at the quarter/year level corresponding to the origination time of the
legacy loan. Standard errors are included in the parentheses.
Panel A: HARP-conforming refi spread and the previous rate
(1) (2) (3) (4)
Previous rate 7.54
(0.10)
5.75
(0.10)
9.97
(0.11)
9.61
(0.78)
90 < LTV ≤ 105 1.69
(0.13)
-0.29
(0.13)
-0.10
(0.42)
105 < LTV ≤ 125 15.02
(0.19)
7.21
(0.20)
6.73
(1.71)
LTV > 125 21.42
(0.24)
10.20
(0.26)
9.31
(1.93)
Borrower Controls No Yes Yes Yes
MSA FEs No Yes Yes Yes
Year-Quarter FEs No No Yes Yes
Servicer FEs No No No Yes
Observations 414,172 414,172 414,172 414,172
Adjusted R-squared 0.01 0.05 0.10 0.11
43
Table 6 [continued]:
Panel B: Harp-conforming refi spread and previous interest rate instrumented with 10-year Treasury rate
(1)
Dependent variable:
Previous rate
(On the original loan)
(2)
Dependent variable:
Previous rate
(On the original loan)
(3)
Dependent Variable:
HARP-Conforming
Refi Spread
(4)
Dependent Variable:
HARP-Conforming
Refi Spread
10-year Treasury 0.61
(0.05)
0.55
(0.04)
- -
Previous rate
instrumented with
10-year Treasury
- -
4.30
(0.43)
10.91
(0.92)
Other controls No Yes No Yes
Observations 414,172 414,172 414,172 414,172
Adjusted R-squared 0.45 0.50 0.01 0.11
44
Table 7:
Difference-in-Difference around the Change in the Program Rules
Column (1) and (2) of this table presents the OLS regression results for the specification with the guarantee fee adjusted HARP-conforming refi spread as the
dependent variable and a dummy variable, 2013× HARP, equal to one if a HARP loan was refinanced in the first half of 2013 and equal to zero if it was
refinanced in the second half of 2012. The sample period consists of HARP loans originated from Q3 2012 till Q2 2013. Column (1) presents the estimation
results for the basic specification model with no additional controls. Column (2) introduces Other controls, including the dummy variables for LTV ranges as in
Tables 2 and 3, FICO, FICO squared, year-quarter fixed effects corresponding to the origination of legacy and HARP loan, MSA fixed effects, and the servicer
fixed effects corresponding to the identity of the servicer handling the loan. Column (3) and (4) present the OLS regression results (in the percentage terms) for
the specification with the dummy taking value of one if a loans refinances in given month and zero otherwise. Once the loan refinances it is dropped from the
estimation sample. The control variables include 2013 dummy that takes value of 1 if the loan is refinance in the first half of 2013 and equal to zero otherwise,
dummy variable HARP that takes value of 1 if a loan is HARP eligible (as of July 2012) and is zero otherwise, and the interaction of these two variables 2013×
HARP. In Column (3) and (4) the sample consists of loans eligible for HARP and conforming refinancing, respectively, tracked from July 2012 till June 2013.
The excluded category are mortgages that are eligible for confirming refinancing as of July 2012. Standard errors are included in the parentheses.
Dependent variable:
HARP-conforming
refi spread
Dependent variable:
Whether a loan refinances
in a given month
(1) (2) (3) (4)
2013 × HARP
-8.11
(0.19)
-9.03
(3.25)
0.10
(0.06)
0.12
(0.06)
2013 - -
0.10
(0.07)
0.21
(0.06)
HARP - -
-0.06
(0.04)
0.22
(0.05)
Other controls No Yes No Yes
Observations 164,144 146,144 1,181,839 1,181,839
Adjusted R-squared 0.01 0.10 0.01 0.01
45
Figure 1:
HARP and Refinancing Rate
Panel (a) of the figure shows the percentage of loans refinancing under HARP in the treatment group (eligible loans) in a given quarter. Panel (b) shows the
estimated coefficients of interaction terms between quarterly time dummies and the treatment indicator (HARP Eligible) along with 99% confidence intervals for
the specification where the dependent variable takes the value of one whether the loan refinances (through HARP or otherwise) in a given quarter and is zero
otherwise. The refinanced loans exit the estimation sample. The specification is similar to one in Column (4) of Table 2A but where we replace the After Q1 2009
dummy with a set of quarterly dummies (the excluded category includes observations from 2008:Q2). This specification allows us to investigate the quarter-by-
quarter changes in the refinancing rate between the treatment and control group (relative to the level in 2008:Q2). The estimation period is 2008:Q3 to 2012:Q4.
(a) Quarterly HARP refinancing rate in the treatment group (b) Change in the refinancing rate between treatment and control group
‐1%
0%
1%
2%
3%
4%
5%
6%
7%
8%
‐1%
0%
1%
2%
3%
4%
5%
6%
7%
8%
46
Figure 2:
Cumulative Change in the Durable Spending (New Auto Financing) after the HARP Refinancing Date
Panel (a) shows the estimated cumulative change in the net amount of new auto financing (in dollars) along with 99% confidence intervals in the eight quarters
following the HARP refinancing. These estimates are from a borrower level specification where the dependent variable takes the value of one if a new auto
financing transaction takes place in a given quarter and is zero otherwise. We include a set of controls capturing borrower, loan, and regional characteristics and a
set of quarterly time dummies that capture the three quarters preceding HARP refinancing and eight quarters following HARP refinancing date. Panel (b) plots
the estimates for less indebted borrowers with below median LTV (dashed line) and more indebted borrowers with above median LTV (solid line). Panel (c)
plots similar estimates for more creditworthy borrowers with above median FICO (dashed line) and less creditworthy borrowers with below median FICO (solid
line). The average differences across these groups in Panels (b) and (c) are statistically significant at 1%.
(a) Cumulative change in the net amount of new auto financing
(b) Less and more indebted (below/above median LTV) (c) Less and more creditworthy (below/above FICO)
47
Figure 3:
Geographical Distribution of Zip Codes and HARP Eligible Share
This figure presents the geographic distribution of zip codes in our overall sample across the United States. In addition, the figure displays the fraction of loans in
a zip code which are eligible for HARP (as of March 2009). As we observe, there is a significant variation in the HARP Eligible share across zip codes (ranging
from just few percent of loans being eligible for the HARP program to more than 70%).
48
Figure 4:
Regional Evidence: Credit Card Spending, Auto Purchase, and House Price Growth in High and Low HARP Exposed Areas
This figure shows the average credit card spending, auto purchase, and house price growth rates in the high HARP exposed (above median Eligible Share) and
low HARP exposed (below median Eligible Share) zip codes. The high HARP exposed group is displayed in solid line and low HARP exposed group is
displayed in dashed line. Zip code credit card spending growth is computed using proprietary data from U.S. Treasury. The auto purchase growth data come from
Mian and Sufi (2010) (based on R.L. Polk & Company data). The house price growth is computed using CoreLogic zip-code level price indices.
(a) Credit card spending growth (b) Auto purchase growth
(c) House price growth
49
Figure 5:
HARP-Conforming Spread across Lenders
Panel (a) of this figure plots the servicer fixed effects corresponding to the identity of HARP lender from the specification in Column (2) of Table 5, in which the
dependent variable is the HARP-conforming refi spread along with 95% confidence intervals. Lender names have been anonymized. Panel (b) shows the relation
between the lenders’ fixed effects (y-axis) and their log asset size as of 2009 (x-axis). This figure is plotted only for the lenders for which we have asset size data
collected from publicly available sources.
(a) Lenders’ fixed effects Lender’s fixed effects and log asset size
-25
-20
-15
-10
-5
0
5
10
15
20
25
8 9 10 11 12 13
50
Figure 6:
Change in the Program Rules and the HARP-Conforming Refi Spread and Harp-Conforming Refinancing Rate
Panel (a) of this figure plots the estimated coefficients (based on OLS) for monthly fixed effects along with 99% confidence intervals around these estimates from a
regression of HARP-conforming refi spread (in basis points) on a set of borrower and loan characteristics including current loan LTV ratios, borrower credit scores,
servicer fixed effects, MSA fixed effects, previous rate, and monthly time fixed effects. The excluded category corresponds to HARP refinances that occurred during
July 2012 so the plotted coefficients show the estimated change relative to the spread from this period. Panel (b) shows the corresponding results from the specification
where now the dependent variable takes the value of one if the loan refinances in a given month and is zero otherwise. The plotted coefficients are the estimated
interaction terms of time dummies with HARP eligible dummy. In panel (b) the base category are loans that are eligible for conforming refinances (loans with current
LTV as of June 2012 less than 80). The displayed coefficients in panel (b) show the estimated change in the difference between refinancing rates of HARP and
conforming loans (relative to the level in July 2012). The estimation period is from July 2012 till June 2013. As we observe the HARP spread and the difference
between refinancing rates of HARP eligible and conforming loans generally persists at a stable level prior to the change in program rules in January 2013. Once the
new rules are in place the spread declines sharply by about 10 basis points in 2013 and the HARP refinancing rate experiences a significant differential increase (by
about 0.12% per quarter).
(a) Change in the HARP-conforming spread (b) Change in the refinancing rate
51
On-Line Appendix
52
Appendix A.1:
Borrower-Level Evidence: Change in the Durable Spending (New Auto Financing) around the HARP Refinancing Date
Panel (a) of this figure plots the OLS estimates for quarterly time fixed effects (along with 99% confidence intervals) from the specification where the dependent
variable takes the value of one if a new auto financing transaction takes place in a given quarter and is zero otherwise. In this specification we include a set of
controls capturing borrower, loan, and regional characteristics and a set of (plotted) quarterly time dummies that capture the three quarters preceding HARP
refinancing and eight quarters following HARP refinancing date. Panel (b) shows the corresponding results for the specification with the net amount of new auto
financing (in dollars) as the dependent variable.
(a) Change in the probability of new auto financing (b) Change in the net amount of new auto financing
53
Appendix A.2:
Borrower-Level Evidence: Change in the Durable Spending (New Auto Financing) and the Rate Reduction due to HARP
This table presents OLS estimates from the specification where the dependent variable is the average quarterly probability of new auto financing in percentage
points (Column 1 and 2) and net dollar amount of new auto financing in dollars (Column 3 and 4) after the HARP refinancing date less the corresponding value
prior to the HARP refinancing. The control variable Rate Reduction captures the difference between the legacy interest rate and rate on refinanced HARP loan (in
percentage terms). Columns (1) and (3) present the estimation results for the basic specification with no additional controls. Columns (2) and (4) add borrower
controls that include variables such as FICO credit score, LTV, interest rates and the fixed effects for the location (MSA) of the property (MSA FEs). Column (2)
and (4) clusters standard errors at the MSA level. The estimation sample consists of a set of borrowers refinancing their loans thru HARP during Q2 2009 till Q4
2012 and tracked over the period of two years after their refinancing date. Standard errors are included in the parentheses.
Probability of
new auto financing
Net amount of new
auto financing
(1) (2) (3) (4)
Rate reduction 0.20 0.12 55.15 36.94
(0.03) (0.04) (11.25)
(72.13)
Borrower Controls
No Yes No Yes
MSA FEs
No Yes No Yes
Year Quarter FEs
No Yes No Yes
Servicer FEs
No Yes No Yes
Observations 357,507 357,507 357,507 357,507
Adjusted R-squared 0.001 0.001 0.001 0.001
54
Appendix A.3:
Regional Evidence: Mortgage Rate Reduction, Program Refinancing Rate and the Fraction of Eligible Loans and Eligible
Loans Serviced by High Cost Lenders in a Zip Code
Column (1) and (2) of this table investigates the relation between the reduction in the average mortgage rate in a zip code due to HARP (in basis points) during
first four years of the program and zip code level Eligible Share and Eligible and High Cost Servicer Share. Eligible Share is the fraction of loans in a zip code
that are GSE and have current LTV ratios greater than 80 prior to the program implementation. Eligible and High Cost Servicer Share is the fraction of loans in a
zip code that are GSE, have current LTV ratios greater than 80 and are serviced by high cost servicers prior to the program implementation. We also include a set
of controls including the zip code average FICO credit score, LTV ratio, interest rate on mortgages along with the average zip code house price growth over the
prior five years and state fixed effects. Column (1) presents the specification without Eligible and High Cost Servicer Share control, while Column (2) repeats
this analysis, but includes this variable in the set of controls. Columns (3) and (4) provide the corresponding analysis for the fraction of loans in a zip code
refinancing under HARP as the dependent variable. The servicer is classified as high cost if it is in the top quartile of servicers with the highest estimated fixed
effects as displayed in Figure 5A. These high cost servicers account for over 60% of loans in our data. The analysis is based on a sample of 3,443 zip codes.
Standard errors (based on the OLS estimates) are included in parentheses.
Dependent variable:
Reduction in the average mortgage
interest rate in a zip code due to HARP
(in basis points)
Dependent variable:
Fraction of loans in a zip code
refinancing under HARP
(1) (2) (3) (4)
Eligible Share
38.0
(0.01)
49.1
(0.03)
0.24
(0.01)
0.35
(0.01)
Eligible and High Cost Servicer Share -
-17.4
(0.05)
-
-0.16
(0.03)
Zip Code Controls Yes Yes Yes Yes
State FEs Yes Yes Yes Yes
Adj. R-squared 0.70 0.71 0.68 0.69
55
Appendix A.4:
Evolution of Observables among HARP and Conforming Refinances prior to the Program Change
This figure we track the evolution of average FICO credit score of borrowers (panel a) and LTV ratio (panel b) at the time of loan refinancing among HARP and
conforming refinance during the six months preceding the change in the program rules. The solid line represents HARP refinances while the dashed line shows
the corresponding means for conforming refinances. As we observe the average LTV ratio for HARP refinances consistently remain about 30% above that for
conforming refinances, with little relative change over time (by construction our benchmark conforming refinances have LTV ratios equal to 80 percent). We
also do not observe a substantial relative variation in the borrower credit scores between HARP and conforming refinances.
(a) FICO credit score
(b) LTV
350
400
450
500
550
600
650
700
750
800
850
2012‐Jul 2012‐Aug 2012‐Sep 2012‐Oct 2012‐Nov 2012‐Dec
0
20
40
60
80
100
120
140
2012‐Jul 2012‐Aug 2012‐Sep 2012‐Oct 2012‐Nov 2012‐Dec
56
Appendix A.5:
Simulation of the Program Effectiveness: Impact of Guarantee Fees and Servicer Markups on the Borrowers’ Savings
(Intensive Margin) and the Refinancing Rate (Extensive Margin)
In this appendix we perform a simple simulation to assess the impact of guarantee fees and servicer markups on the borrowers’ savings (intensive margin) and the
refinancing rate (extensive margin) induced by HARP. The figure plots the simulated interest rate savings of the borrowers (panel a) due to the HARP refinancing (in
percentage points) and the percentage of eligible borrowers refinancing their loans under the HARP (panel b) across the borrowers current LTV ratios as of March
2009 (the date of the program implementation). Top solid line shows the case in which borrowers can refinance (at no additional cost) to the benchmark conforming
rate (with no guarantee fees) as of March 2009. The middle dashed line shows the case in which borrowers face the HARP guarantee fees. The bottom dashed line
corresponds to the case when borrowers face both the HARP guarantee fees and the predicted servicer “monopolistic” markups when they refinance their loans. To
predict this markup we use the estimates from our specification in Column (4) of Table 5. In the calculation in panel (b) we assume that in order to refinance the
borrowers need to obtain an annual interest rate reduction of at least 1% (100 basis points per year). Source: a random sample of all borrowers with 30 year FRMs that
were eligible for HARP as of its implementation date in March 2009 based on their estimated LTV ratios of greater than 80 percent. This sample amounts to more than
1.1 million conforming mortgages from a large secondary market participant, which is more than 15% of the entire population of eligible loans.
(a) Interest rate savings on HARP refinances (in percentage points) (b) Percentage of eligible borrowers refinancing under HARP
Readers with comments may address them to:
Professor Amit Seru
5807 S. Woodlawn Avenue
Chicago, IL 60637
United States
The University of Chicago Law School
Kreisman Working Papers on Housing Law and Policy
For a listing of papers, please go to http://chicagounbound.uchicago.edu/housing_law_and_policy
1. Lee Anne Fennell and Eduardo M. Peñalver, Exactions Creep, December 2013
2. Lee Anne Fennell, Forcings, November 2013
3. Neil Bhutta and Benjamin J. Keys, Interest Rates and Equity Extraction during the
Housing Boom, January 2014
4. Christopher Mayer, Edward Morrison, Tomasz Piskorski, and Arpit Gupta, Mortgage
Modification and Strategic Behavior: Evidence from a Legal Settlement with
Countrywide, January 2014
5. Edward R. Morrison, Coasean Bargaining in Consumer Bankruptcy, January 2014
6. Atif Mian, Amir Sufi, and Francesco Trebbi, Foreclosures, House Prices, and the Real
Economy, January 2014
7. Sumit Agarwal, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, Tomasz
Piskorski, and Amit Seru, Policy Intervention in Debt Renegotiation: Evidence from the
Home Affordable Modification Program, August 2012
8. Sumit Agarwal, Effi Benmelech, Nittai Bergman, and Amit Seru, Did the Community
Reinvestment Act (CRA) Lead to Risky Lending? November 2012
9. Tomasz Piskorski, Amit Seru, and James Witkin, Asset Quality Misrepresentation by
Financial Intermediaries: Evidence from RMBS Market, February 2013
10. Umit G. Gurun, Gregor Matvos, and Amit Seru, Advertising Expensive Mortgages,
March 2013
11. Benjamin J. Keys, Devin G. Pope, and Jaren C. Pope, Failure to Refinance, August 2014
12. Adam B. Badawi and Anthony J. Casey, The Fannie and Freddie Bailouts through the
Corporate Lens, March 2014
13. Lee Anne Fennell, Property in Housing, March 2013
14. Lee Anne Fennell, Just Enough, August 2013
15. Yun-chien Chang and Lee Anne Fennell, Partition and Revelation, April 2014
16. Robert J. Chaskin and Mark L. Joseph, Contested Space: Design Principles and
Regulatory Regimes in Mixed-Income Communities Replacing Public Housing
Complexes in Chicago, October 2014
17. Lee Anne Fennell, Agglomerama, December 2014
18. Sebastien Gay and Nadia Nasser-Ghodsi, Guarding the Subjective Premium:
Condemnation Risk Discounts in the Housing Market, December 2014
19. Brian A. Jacob, Max Kapustin, and Jens Ludwig, Human Capital Effects of Anti-Poverty
Programs: Evidence from a Randomized Housing Voucher Lottery, December 2014
20. Sebastien Gay and Allen T. Zhang, Expertise Value Added in the Real Estate Market,
December 2014
21. Atif R. Mian and Amir Sufi, Fraudulent Income Overstatement on Mortgage
Applications during the Credit Expansion of 2002 to 2005, February 2015
22. Sebastien Gay, Investors Effect on Household Real Estate Affordability, May 2015
23. Omri Ben-Shahar and Kyle D. Logue, The Perverse Effects of Subsidized Weather
Insurance, May 2015
24. Lee Anne Fennell, Co-Location, Co-Location, Co-Location: Land Use and Housing
Priorities Reimagined, September 2015
25. Erik Hurst, Benjamin J. Keys, Amit Seru & Joseph S. Vavra, Regional Redistribution
Through the U.S. Mortgage Market, September 2015
26. Nicholas O. Stephanopoulos, Civil Rights in a Desegregating America, October 2015
27. Sumit Agarwal, Gene Amromin, Souphala Chomsisengphet, Tomasz Piskorski, Amit
Seru, & Vincent Yao, Mortgage Refinancing, Consumer Spending, and Competition:
Evidence from the Home Affordable Refinancing Program, October 2015
