y i
We
ullerto
omona
state-level and Zip code-level data over the period 2001–2008, we find that the fraction of
piggyback originations is related to higher foreclosure and default rates in subsequent
years, and this relation is strongest for non-owner-occupied properties. The pattern, how-
others).
One particular type of lending that grew rapidly during
the recent housing boom is piggyback lending. Piggyback
loans, more technically referred to as simultaneous close
ers taking out a piggyback loan in 2006 (State of New York
City’s Housing and Neighborhoods Report, 2007). Similarly,
about 37.3% of California borrowers also used piggyback
loans to finance home purchase in 2006 (Fishbien, 2006).
The rise of piggyback lending during 2001–2006 may
also have contributed to the rise in default and foreclosure
rates. Some have argued that piggyback lending enables
households to take on too much debt via the purchase of
1051-1377/$ - see front matter � 2011 Published by Elsevier Inc.
⇑ Corresponding author.
E-mail addresses: mlacour-little@fullerton.edu (M. LaCour-Little),
charles_calhoun@cox.net (C.A. Calhoun), weiyu@csupomona.edu (W. Yu).
Journal of Housing Economics 20 (2011) 81–100
Contents lists available at ScienceDirect
Journal of Housi
w.el
doi:10.1016/j.jhe.2010.11.002
The current financial crisis had its origins in 2006 as
house prices began to fall and the mortgage market expe-
rienced a sharp increase in subprime mortgage defaults
and foreclosures. Numerous papers have studied the fac-
tors that contributed to the unprecedented increase in de-
fault and foreclosure rates. Irrational expectations
regarding future house price growth, a proliferation of
non-agency mortgage securitization, lax underwriting,
and changing economic conditions are among the cited
factors (Bajari et al., 2008; Doms et al., 2007; Keys et al.,
2010; Mian and Sufi, 2009; Coleman et al., 2008, among
chase. These are generally used by homebuyers to
finance more than 80% of the house value without paying
private mortgage insurance, at least if the first lien is
GSE-financed. Piggyback lending played an important role
in home sales, especially from 2004 to 2006, and was
involved in about 22% of the one-to-four family owner-
occupied home purchases in 2006 (Avery et al., 2007b). It
is particularly popular in high-cost housing areas. For
example, between 2004 and 2006, the number of piggy-
back loans issued in New York city more than tripled,
resulting in more than 30% of the home purchase borrow-
JEL Classification:
G01
G21
R31
Keywords:
Mortgage
Foreclosure
Housing
Piggyback lending
1. Introduction
ever, appears to be limited to the use of subprime piggybacks, rather than a more general
phenomenon. Using a topology-based housing supply elasticity measure as an instrument
for house price growth, we further confirm that the strong association of subprime piggy-
back origination with worse loan performance was not driven by the endogeneity of house
price appreciation.
� 2011 Published by Elsevier Inc.
seconds, are junior lien mortgage loans taken out concur-
rently with the first mortgage to finance the home pur-
Article history:
Received 29 April 2010
Available online 20 February 2011
We examine the use of simultaneous close junior lien lending (‘‘piggybacks’’) over the
course of the recent housing bubble and subsequent mortgage market collapse. Using both
What role did piggyback lending pla
and mortgage collapse?
Michael LaCour-Little a,⇑, Charles A. Calhoun b,
aCollege of Business and Economics, California State University at Fullerton, F
bCalhoun Consulting LLC, Annandale, VA 22003, USA
cCollege of Business Administration, California State Polytechnic University, P
a r t i c l e i n f o a b s t r a c t
journal homepage: ww
n the housing bubble
i Yu c
n, CA 92834-6848, USA
, CA 91768, USA
ng Economics
sevier .com/locate / jhec
inflated assets (WSJ, 2009), and therefore helped to further
inflate the housing bubble. Once the bubble burst, it made
highly leveraged households at greater risk of negative
non-owner-occupied piggybacks and high-LTV piggybacks
that we have been able to identify that directly addresses
the determinants of home equity borrowing is Salandro
as in most data on home equity lending, their data does
82 M. LaCour-Little et al. / Journal of Housing Economics 20 (2011) 81–100
have stronger impact on mortgage performance. Of partic-
ular interest, we distinguish among three forms of piggy-
back lending: (1) prime first lien and prime second lien,
(2) prime first lien and subprime second lien, and (3) sub-
prime first lien and subprime second lien, and further ex-
plore the difference in these three piggyback lending
patterns in explaining state-level and Zip code-level mort-
gage performance.
The plan of the paper is as follows. In the next section,
we review the limited research related to this topic. In
the third section, we describe our data and empirical meth-
odology, including our method for identifying piggyback
loans from HMDA data. In the fourth section we present re-
sults of our empirical results. The final section concludes.
2. Literature review
Compared to other research on mortgage markets, ju-
nior lien lending is a relatively unexplored arena. The still
narrower topic of piggyback lending has received even less
rigorous research.
Beginning with the broader research on junior lien debt,
Canner et al. (1988) describe the early stages and growth of
the home equity lending segment, following passage of the
1986 tax law changes which are generally acknowledged
to have spurred this segment of consumer lending.1 Canner
and Luckett (1994) and Canner et al. (1998) update those
findings, including Survey of Consumer Finance (SCF) data
showing home equity balances outstanding reached $110
billion by 1994. Weicher (1997) reviews the growth of the
home equity lending industry during the 1990s, describing
it as business based on recapitalizing borrowers with sub-
stantial housing equity, but impaired credit. The only paper
1 Prior to 1986 most interest on consumer debt was tax-deductible; after
1986 tax law changes, only debt secured by residential mortgage debt
remained generally deductible for those who itemize deductions.
equity and more vulnerable to default. The use of piggy-
back loans has been shown to be important in explaining
the magnitude of negative equity (LaCour-Little et al.,
2009). Piggyback loans also make the loan modification
process more complicated because first-lien and junior-
lien loans are packaged and sold to different portfolio
securitizations (Rosengren, 2008). Moreover, junior-lien
lenders, if different from first-lien lenders, usually have lit-
tle incentive in modifying the loan to avoid foreclosure if
there is no equity protecting them (The Washington Post,
2008).
In this paper, we study the relationship between the
mortgage performance (delinquency, foreclosure, and de-
fault rates) and homeowner piggyback borrowing patterns
at both the state level and the Zip code level. We ask
whether states and Zip codes with higher proportions of
piggyback loans issued during 2001–2006 are associated
with worse mortgage performance in later periods, espe-
cially during 2007 and 2008. We also examine whether
not contain information about the underlying first mort-
gage loans, since first and junior debt is often held by dif-
ferent lenders, a pattern that applies to piggyback lending
as well. LaCour-Little et al. (2009) report that roughly 80%
of Southern California borrowers facing foreclosure during
2006–2008 had at least one junior lien outstanding, though
information on the loans themselves is limited.
Calhoun (2006b) develops a method for identifying pig-
gyback loans from HMDA data. Calhoun (2006a) argues
that simultaneous-close or ‘‘piggyback’’ transactions sys-
tematically raise risk throughout the mortgage finance sys-
tem, yet presented no loan performance data. Bernstein
(2008), as well as others mentioned in the introduction,
document the increase in the use of piggyback lending over
the period we study. Using American Housing Survey data,
Bernstein (2008) reports that multiple-mortgage financing
packages as a percent of newly originated mortgages in-
creased from 14.8% in survey year 2001 to 21.5% in survey
year 2007, corroborating the growth in this category docu-
mented by others.
Clearly considerable additional research is necessary to
more completely understand this new market phenome-
non, its causes, and its effects. Our effort here addresses
this gap in the literature.
3. Data and empirical methodology
3.1. Data
To calculate the proportion of piggyback loans to total
home purchase loan originations for each state and Zip
code, we use Home Mortgage Disclosure Act (HMDA) data
from 2001 to 2008. We first identify piggyback loans2 using
each of the methods proposed by Avery et al. (2007a) and
Calhoun (2006b). To calculate the proportion of piggyback
originations at the state level for a given year, we aggregate
the number of piggybacks by state and year and divide it by
the total number of first-lien home purchase loan origina-
tions. For Zip code-level piggyback originations, we calculate
the piggyback originations at the census tract level first and
2 Details of the identification method are addressed later.
and Harrison (1997), who used 1989 and 1992 SCF data, well
before the dramatic increase in home equity lending
occurred.
In more recent work, LaCour-Little (2004) argues that a
borrower’s post-origination home equity borrowing di-
lutes their equity, increasing the risk of default on the se-
nior debt. Agarwal et al. (2006a) shows that patterns of
home equity line use are also related to borrower credit
quality, as measured by their FICO score. Extending that
analysis further, Agarwal et al. (2006b) examine the perfor-
mance of home equity lines and loans, finding considerable
difference in terms of default and prepayment risk. Mian
and Sufi (forthcoming) examine home equity-based bor-
rowing from 2002 to 2006 and find that it is associated
with high default rates from 2006 to 2008. Unfortunately,
then aggregate it to the Zip code using a database that
matches census tract numbers with Zip codes from Missouri
Census Data Center.3 Because a given census tract can corre-
spond to more than one Zip codes, we create a weight vari-
able based on the share of housing units in each census tract
that lie within a given Zip codes. Using this weight variable,
we can calculate the weighted average piggyback origina-
tions at the Zip code level.4
For state-level loan performance measures, we use the
state-level percentage of mortgage foreclosure inventory
from 2001 to 2008, obtained from the Mortgage Banker’s
Association (MBA), as our proxy for state foreclosure rates.
Since the MBA data are only available at the state level, we
use the Zip code proportion of noncurrent (delinquent5 and
defaulted6) mortgages and the proportion of mortgages in
default from 2001 to 2008, obtained from Equifax, as our
proxy for Zip code delinquency and default rates.
We supplement the data with additional economic vari-
ables that may also affect loan performance. First, our
state-level house price data come from Federal Housing Fi-
nance Agency (FHFA, formerly Office of Federal Housing
M. LaCour-Little et al. / Journal of Housing Economics 20 (2011) 81–100 83
Enterprise Oversight, or OFHEO) purchase-only House
Price Indexes from 2001 to 2008. We use this data to calcu-
late annual house price appreciation rates. At Zip code le-
vel, we use the CoreLogic7 Zip code Single Family
Detached House Price Index to calculate house price appre-
ciation rates. The CoreLogic House Price Index covers about
7600 Zip Codes in the United States. Second, we use both
state and MSA level per-capita income data from the Bureau
of Economic Analysis. The state per-capita income is used in
our state-level analysis, and the MSA per-capita income is
used as a proxy for Zip code income. Third, we obtained state
unemployment data from the Department of Labor Statis-
tics. We use the MSA data as a proxy for the Zip code income
and unemployment rate. Fourth, we supplement the Zip
code-level data with Zip code credit risk data obtained from
Equifax.
Lastly, subprime loans are identified in two ways: For
loans in HMDA from 2004 and later years we treat high-
3 We recognize that the HMDA data uses the 1990 census tract
definitions before 2003 and the 2000 census tract definitions starting
2003. Therefore, we use a database that matches 1990 census tract
numbers with Zip codes for data before 2003 and another database that
matches 2000 census tract numbers with Zip codes for data on and after
2003. Both databases are available from Missouri Census Data Center.
4 For example, census tract 1 has 500 housing units and 20% of the
housing units are within Zip code A. Census tract 2 has 1000 housing units
and 15% of the housing units are within Zip code A. Census tract 1 has 100
piggyback originations and census tract 2 has 200 piggyback originations.
For simplicity, Zip code A are solely composed of census tract 1 and 2. To
calculate the piggyback origination for Zip code A, we first create a weight
variable for census tract 1 and 2 based on the number of housing units of
each census tract that lie within Zip code A. The weight for census tract 1 is
40% (500 � 20%/(500 � 20% + 1000 � 15%)) and the weight for census tract
2 is 60% (1000 � 15%/(500 � 20% + 1000 � 15%)). We then calculate the
weighted average piggyback originations for Zip code A as
100 � 40% + 200 � 60% = 160.
5 A mortgage is coded as delinquent if it is more than 30 days and less
than 90 days past due.
6 A mortgage is coded as in default if it is more than 90 days past due, in
bankruptcy status or in severe derogatory status.
7 CoreLogic was spun off in an IPO from The First American Corporation
on June 1, 2010.
cost or spread-reportable loans as subprime loans. For
years prior to 2004 we use the Department of Housing
and Urban Development’s (HUD) list of subprime lenders
to identify subprime loans based on the lender IDs in
HMDA. We then aggregate the subprime loans by state
and Zip code and calculate the proportion of subprime loan
originations to total loan originations in a given year. Both
measures of subprime loans are imperfect. Use of the HUD
list is a somewhat crude proxy because all loans made by a
subprime lender will be classified as subprime loans and
all loans made by a prime lender will be classified as prime
loans. Reliance on the spread-reportable threshold in
HMDA may under- or over-estimate the prevalence of sub-
prime loans depending on current interest rate conditions,
and specifically the relationship between mortgage rates
and comparable maturity Treasury rates. Detailed defini-
tions of the key variables used in state-level and Zip
code-level regressions and their sources are listed in
Table 1.
3.2. Identification of piggyback loans
We use two methods to identify piggyback loans in the
HMDA data. The first method is based on Avery et al.
(2007a). Since information about lien status is only avail-
able starting in 2004, the identification process is a little
different before 2004 and after 2004. Before 2004, we sort
home purchase loans each year by state, county, census
tract number, lender ID, owner-occupancy status, bor-
rower income, race, and sex. If we find duplicate loan re-
cords according to this set of matching factors, then the
one with the smaller loan amount is identified as a piggy-
back loan. The basic assumption underlying this method is
that if two home purchase loans involve a property in the
same census tract and same owner-occupancy status, bor-
rowers with identical income, race and sex, and was issued
by the same lender, then most likely these two loans are
used for the purchase of the same home. After 2004, with
the addition of lien status in HMDA, we separate the home
purchase loans into two samples. The first sample includes
all junior-lien purchase loans and the second sample in-
cludes all first-lien purchase loans. We then match the sec-
ond sample to the first sample by census tract, lender ID,
owner-occupancy status, borrower income, race and eth-
nicity, and sex. If there is a match, then the matched
junior-lien loan is identified as piggyback loans. One limi-
tation of the Avery method is that it may underestimate
the number of piggyback loans because it can not identify
piggyback loans that are issued by lenders different from
the first-lien lenders.
The second method follows Calhoun (2006b). The Cal-
houn approach is similar to that of Avery, but recognizes
the following two potential problems: First, piggyback
loans may be issued by a different lender from the first-lien
lender. Second, the lien status is usually missing for loans
that are reported as secondary market purchases in HMDA.
Calhoun proposed a two-step matching procedure. The
first step follows Avery, by finding duplicates according
to a set of matching factors. In the second step, if the lien
status is not missing, then the data are sorted by state,
county, census tract, borrower income, lien status, and
Source
Mortgage Banker’s
Association
-lien purchase loan originations HMDA
l foreclosure procedures and 3-year lag of piggyback HMDA
Bureau of Economic
nation
rtgage
er of m
-lien
l forec
84 M. LaCour-Little et al. / Journal of Housing Economics 20 (2011) 81–100
Table 1
Variable definitions.
Variable Definition
Panel A: State-level regression
Foreclosure State foreclosure rate
Piggyback Proportion of piggyback loan originations to total first
Piggylag Two-year lag of piggyback for states with non-judicia
for states with judicial foreclosure procedures
Lnincome Log of state per-capita income
Unemploy State unemployment rate
Pctsubprimelag Two-year lag of the proportion of subprime loan origi
Hpigrowth House price appreciation rate
Hpigrowthlag One-year lag of house price appreciation rate
Panel B: Zip code-level regression
Noncurrent Proportion of noncurrent (delinquent and default) mo
given Zip code
Default Proportion of defaulted mortgage loans to total numb
Piggyback Proportion of piggyback loan originations to total first
Piggylag Two-year lag of piggyback for states with non-judicia
for states with judicial foreclosure proceduresa
Lnincome Log of MSA per-capita income
Unemploy MSA unemployment rateb
loan amount. Adjacent first and second liens with identical
values of borrower income are identified as piggyback
combinations and removed from the data. If the lien status
is missing, then the remaining data are sorted by state,
county, census tract, borrower income, and loan amount
and adjacent loans with identical values of borrower in-
come are matched. In the second step of this two-step
matching, the matching factors do not include lender ID.
The relaxation of the same lender assumption helps to
identify additional piggyback combinations that are issued
by different lenders. To further confirm a piggyback match,
Calhoun then calculated the ratio of the lower loan amount
to higher loan amount for the duplicates. The piggyback
combinations are further confirmed if the ratio of the smal-
ler loan amount to the larger loan amount falls into certain
ranges that are consistent with the piggyback loan struc-
tures (such as 80-20-0 structure, 80-10-10 structure, etc.).
To examine whether non-owner-occupied and highly-
leveraged piggybacks have a stronger effect on default
and foreclosure rates, we also identify piggyback loans
used for non-owner-occupied house purchases and piggy-
back loans with combined loan-to-value (CLTV) ratios
above 100%. Since information on the house value is not
provided by HMDA data, we assume the first lien is
approximately 80% of the house value. The combined
Pctsubprimelag Two-year lag of the proportion of subprime loan origination
Hpigrowth Zip code house price appreciation rate
Hpigrowthlag One-year lag of ho
本文档为【What role did piggyback lending play in the housing bubble】,请使用软件OFFICE或WPS软件打开。作品中的文字与图均可以修改和编辑,
图片更改请在作品中右键图片并更换,文字修改请直接点击文字进行修改,也可以新增和删除文档中的内容。
该文档来自用户分享,如有侵权行为请发邮件ishare@vip.sina.com联系网站客服,我们会及时删除。
[版权声明] 本站所有资料为用户分享产生,若发现您的权利被侵害,请联系客服邮件isharekefu@iask.cn,我们尽快处理。
本作品所展示的图片、画像、字体、音乐的版权可能需版权方额外授权,请谨慎使用。
网站提供的党政主题相关内容(国旗、国徽、党徽..)目的在于配合国家政策宣传,仅限个人学习分享使用,禁止用于任何广告和商用目的。