Recourse Mortgage Law and Asset Substitution: Evidence from the Housing Bubble * Tong Yob Nam † Seungjoon Oh ‡ August, 2013 Abstract In a state with non-recourse mortgage law, borrowers have limited liability on their mortgage loan. This paper tests whether state-level variation in recourse mortgage laws affects housing prices and mortgage lending. Using the difference-in-difference approach with the mortgage market collapse in 2007 as an exogenous shock, we find that non- recourse law results in larger bubbles in housing prices. To identify the causal effects of recourse law on housing prices, we further provide evidence by comparing housing prices in contiguous border county-pairs in the United States and examining discontinuities at state borders. We then demonstrate speculative investment behaviors of households in non-recourse states by showing their higher leverage decision and more asset allocation on the housing assets. We also explore whether mortgage lending behavior differs in non-recourse states owing to anticipation of additional risk. We find evidence of higher mortgage interest rate and higher loan denial rate in non-recourse states, which suggests that lenders are aware of additional risk in non-recourse loan. However, we find that the emergence of the originated-to-distribute (OTD) market on the housing markets enables lenders to effectively shift the risks to other investors and, thus, the mortgage lending behavior does not fully reflect the higher risk. The higher ratio of sub-prime loan in non-recourse states and its positive interaction effect with non-recourse law on housing price growth support our two-stage risk-shifting mechanism in the recent housing bubble. Keywords: Recourse mortgage, State border discontinuity, Housing bubble, Sub- prime crisis JEL Classification: E44, G11, G21, G28, K11, R20 * The authors are grateful for comments from John Bound, Amy Dittmar, Jing Zhang, Margaret C. Levenstein, Amiyatosh Purnanandam, Nejat Seyhun, Jeffrey Smith, and Jagadeesh Sivadasan. This research was supported in part by an NIA training grant to the Population Studies Center at the University of Michigan (T32 AG000221). Nam acknowledges the research grant support from the Michigan Institute for Teaching and Research in Economics (MITRE). All errors are our own. † Department of Economics, University of Michigan, e-mail: [email protected]‡ Stephen M. Ross School of Business, University of Michigan, e-mail: [email protected]
33
Embed
Recourse Mortgage Law and Asset Substitution: Evidence ... · PDF filesure requirement, which is related to foreclosure costs from a lender’s perspective, ... remaining mortgage
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Recourse Mortgage Law and Asset Substitution:
Evidence from the Housing Bubble∗
Tong Yob Nam† Seungjoon Oh‡
August, 2013
Abstract
In a state with non-recourse mortgage law, borrowers have limited liability on theirmortgage loan. This paper tests whether state-level variation in recourse mortgage lawsaffects housing prices and mortgage lending. Using the difference-in-difference approachwith the mortgage market collapse in 2007 as an exogenous shock, we find that non-recourse law results in larger bubbles in housing prices. To identify the causal effects ofrecourse law on housing prices, we further provide evidence by comparing housing pricesin contiguous border county-pairs in the United States and examining discontinuities atstate borders. We then demonstrate speculative investment behaviors of households innon-recourse states by showing their higher leverage decision and more asset allocationon the housing assets. We also explore whether mortgage lending behavior differs innon-recourse states owing to anticipation of additional risk. We find evidence of highermortgage interest rate and higher loan denial rate in non-recourse states, which suggeststhat lenders are aware of additional risk in non-recourse loan. However, we find thatthe emergence of the originated-to-distribute (OTD) market on the housing marketsenables lenders to effectively shift the risks to other investors and, thus, the mortgagelending behavior does not fully reflect the higher risk. The higher ratio of sub-primeloan in non-recourse states and its positive interaction effect with non-recourse lawon housing price growth support our two-stage risk-shifting mechanism in the recenthousing bubble.
Keywords: Recourse mortgage, State border discontinuity, Housing bubble, Sub-prime crisis
JEL Classification: E44, G11, G21, G28, K11, R20
∗The authors are grateful for comments from John Bound, Amy Dittmar, Jing Zhang, Margaret C.Levenstein, Amiyatosh Purnanandam, Nejat Seyhun, Jeffrey Smith, and Jagadeesh Sivadasan. This researchwas supported in part by an NIA training grant to the Population Studies Center at the University ofMichigan (T32 AG000221). Nam acknowledges the research grant support from the Michigan Institute forTeaching and Research in Economics (MITRE). All errors are our own.
†Department of Economics, University of Michigan, e-mail: [email protected]‡Stephen M. Ross School of Business, University of Michigan, e-mail: [email protected]
1 Introduction
A housing market bubble fueled by credit expansion is a fundamental source of crises
(Herring and Wachter (2003); Reinhart and Rogoff (2008); Reinhart and Rogoff (2009);
Mayer, Pence, and Sherlund (2009); and Makarov and Plantin (2013)). The bubble
bursts at some point and triggers collapses in asset prices that spread to the banking
system and the real economy. Understanding the mechanisms that create bubbles
in the housing market is thus a central challenge that both financial economists and
policy makers are facing. This paper builds on the previous literature by showing that
mortgage law plays an important role in housing market bubbles. In particular, it sheds
light on the effect of recourse mortgage law on housing prices and mortgage lending.
U.S. mortgage law varies state to state. Recourse law governs lenders’ right of
deficiency judgment when borrowers default on mortgage loan payments. Borrowers in
recourse states have full liability for their mortgage loans because lenders, in the event
that foreclosure value is insufficient to meet the debt obligation, are able to claim other
assets. Lenders in non-recourse states are precluded from doing this and so bear some
costs. This limited liability gives rise to the classic asset substitution problem in Allen
and Gale (2000), whereby borrowers increase risky investments to the point of creating
a bubble by bidding up prices above fundamental values.
This paper examines whether recourse law results in bubbles of different magni-
tudes. Our key hypothesis is that non-recourse law causes larger bubbles in housing
prices. Mortgage borrowers in non-recourse states, because they can walk away when
house values fall below fundamental values (i.e., are “underwater”), have speculative
motives to increase their leverage and allocate more capital to risky assets in the hous-
ing market. If the non-recourse law causes larger bubbles during a period of economic
expansion, then during a crisis housing prices in non-recourse states are likely to expe-
rience a larger drop than in recourse states.
We empirically test the effect of recourse law on housing prices using a difference-
in-difference framework that focuses on counties that were disproportionately affected
by the crisis. To show a causal relation, we use a contiguous border county-pair
sample. This identification strategy enables us to estimate the effect of recourse law on
housing price after controlling for unobserved spatial heterogeneity. Using ZIP code-
level housing prices from Zillow Real Estate Research between 1999 and 2011, we find
evidence that housing prices in non-recourse states increase more during housing market
booms and drop more steeply during housing market recessions. The economic impact
of recourse law is large. Prior to the crisis, recourse states experienced 9% increases
and the crisis reduced the housing prices by 11%. But in contiguous border pair-county
with non-recourse law experience 13% growth, and the corresponding drop in housing
prices was 17%. Price discontinuities at state borders provide further evidence of the
2
impact of recourse law. Controlling for the distance from state borders using ZIP
codes, we find that housing prices during the crisis increase abruptly upon crossing
into recourse states.
We then identify the channels of larger housing bubble in non-recourse states by
examining leverage and asset allocation of households. During the housing market
expansion, we find that the average ratio of housing assets to total household wealth
is higher in non-recourse states by 7-8 percentage points. In addition, mortgage bor-
rowers in non-recourse states have higher debt-to-income ratio by 11percentage points.
The higher leverage and more asset allocation in housing market demonstrate the spec-
ulative motives of households in non-recourse states drive the higher housing growth
during the economic expansion.
Having established the impact of recourse law on housing prices, we examined its
effect on mortgage lending behavior. Lenders can control the risk of borrower’s default
by means of down payments, high interest rates, and stricter screening processes.
Used appropriately, these tools can forestall borrowers shifting risk and bidding up
prices above their fundamental values. The corresponding hypothesis is that mortgage
interest rates are higher in non-recourse than in recourse states to minimize the costs
that result from lack of deficiency judgments. For this analysis, we obtain mortgage
spread data from HMDA(The Home Mortgage Disclosure Act) at the ZIP code-level
and employ a state-border discontinuity design with the contiguous border pair-county
sample. We also test the denial rate of loan applications to examine whether lenders
have different screening intensity.
We find evidence of higher interest rates in non-recourse than in recourse states. In
addition, the denial rate is higher in non-recourse states. The results suggest that
lenders in non-recourse states are aware of the additional risks embedded in non-
recourse loan. Then, it is surprising that how the larger bubbles are created in non-
recourse states notwithstanding the evidence that lenders in non-recourse states charge
higher mortgage interest rates to compensate for potential costs associated with limited
borrower’s liability. We conjecture that the emergence of the originate-to-distribute
(OTD) market, together with credit expansion, enabled lenders to effectively shift
the risk to other investors and reduced the screening incentive ex-ante (Purnanandam
(2011)). Therefore, it will lead to insufficient lenders’ control of additional risk in
non-recourse loan.
This two-stage risk-shifting hypothesis predicts that more sub-prime mortgage loans
are originated in non-recourse than in recourse states and non-recourse states with more
sub-prime mortgage loans experience larger housing bubbles. We test this hypothesis
by estimating the effect of recourse law on the sub-prime mortgage ratio using HMDA
data for 2004-2008. We find lenders in non-recourse states to originate, on average,
3
17% more sub-prime loans than lenders in recourse states. Furthermore, non-recourse
states with high sub-prim loan ratio experience particularly large housing bubbles.
Taken together, these results suggest that the housing bubble is likely to be larger
in non-recourse states, because the OTD market and credit expansion dissuade lenders
from controlling the consequent risk. This paper offers novel contributions to a growing
literature on the sources of mortgage crises, and examines mortgage laws and their
impact on housing investment.
The paper’s primary contribution is its findings on the impact of mortgage laws on
housing prices. Impacts of the judicial foreclosure requirement on house prices and the
supply of mortgage loans have been examined by Pence (2006) and Mian, Sufi, and
Trebbi (2013), respectively. Pence (2006) finds that the judicial foreclosure requirement
reduces mortgage credit supply by imposing greater costs on lenders seeking foreclo-
sures on houses. In contrast, Mian, Sufi, and Trebbi (2013) highlight that non-judicial
foreclosure requirements have a large negative impact on house prices by increasing
the supply of houses from foreclosure. Recourse law, which is not emphasized in their
studies, clearly differs from the judicial foreclosure requirement. The judicial foreclo-
sure requirement, which is related to foreclosure costs from a lender’s perspective, does
not protect borrowers from unlimited liability. Our overall findings relate to a large
literature on how bubbles can arise (eg., Tirole (1982, 1985); Allen and Gorton (1993);
Allen, Morris, and Postlewaite (1993); Allen and Gale (2000); Abreu and Brunner-
meier (2003); Scheinkman and Xiong (2003); Brunnermeier and Nagel (2004), Hong,
Scheinkman, and Xiong (2008); and Brunnermeier (2003)).
Several papers have examined the effect of recourse law on the mortgage default
rate. Ghent and Kudlyak (2011) show that mortgage defaults are more frequent in non-
recourse states, and find no evidence that mortgage interest rates vary according to
state laws, and Pavlov and Wachter (2004, 2006) propose a model for the underpricing
equilibrium of the put option embedded in non-recourse mortgage lending. The present
paper differs from these studies in emphasizing the agency problem between mortgage
lenders and borrowers, and in employing refined econometric designs to analyze housing
bubbles under different recourse laws.
By providing some of the first evidence on the relation between recourse law and
the housing price bubble, this paper expands previous research on the recent mortgage
crisis. Combined with significant credit expansion from low interest rate policies, the
role of the housing market preceding the crisis is highlighted (Herring and Wachter
(2003); Reinhart and Rogoff (2008); Reinhart and Rogoff (2009); Mayer, Pence, and
Sherlund (2009); and Makarov and Plantin (2013)). Many studies have shown that sub-
prime mortgage expansion promoted the unsustainable growth that led to the collapse
of the market (Mian and Sufi (2009), Purnanandam (2011)). This paper further extends
4
previous research by showing how recourse law, through its influence on borrower risk-
taking behavior, accounts for variations in sub-prime mortgage expansion and the
impact of the mortgage crisis.
The rest of this paper is organized as follows. The origins of recourse law are
explored and hypotheses developed in Section 2. Sample data are described in Section
3. In Section 4, an empirical strategy is developed and the impact of recourse law on
housing prices is examined. Mortgage lending behavior is investigated in Section 5,
the impact of recourse law on sub-prime mortgage expansion is analyzed in Section 6.
Section 7 concludes.
2 Recourse Law and Hypothesis
2.1 Mortgage Recourse Law
U.S. mortgage law varies across states in many important ways. Depending on
lenders’ right of deficiency judgment when borrowers default on residential mortgage
loans, state-level mortgage law can be classified as recourse and non-recourse. Recourse
law permits lenders to claim, in other assets and salary, the difference between a
remaining mortgage amount and the foreclosure value of a house. Non-recourse law
allows lenders to seize only the collateralized house in the event of a mortgage default.1
There have been few changes in the state-level recourse law since the legislation
enacted during the Great Depression of the 1930s. During this economic recession,
foreclosure sales were intense and widespread enough to distort the housing market and
caused houses to be sold below their fundamental value. However, mortgage lenders
sold borrower’s property at deep discount and then claimed deficiency judgments for
the full amount of the debt, which amplified the depression. This led to the legislation
of anti-deficiency judgment in many states (Solomon and Minnes (2011)). Following
Ghent and Kudlyak (2011), Figure 1 presents recourse and non-recourse law states.
2.2 Hypothesis Development
We attempt to understand in this paper whether the magnitude of housing market
bubbles reflects differences between recourse and non-recourse laws. We hypothesize
that a larger bubble is created during a housing market boom, and a larger burst
experienced during a housing market recession, in non-recourse law states. The asset
substitution model by Allen and Gale (2000) provides the theoretical rationale for
borrowers with limited liability investing aggressively in risky assets and creating a
1Even though states are not strictly classified as recourse and non-recourse, it is widely accepted amongboth academics and practitioners that 11 states have non-recourse mortgage laws.
5
bubble by bidding up asset prices above their fundamental value.
Hypothesis 1: A state with non-recourse law creates a larger housing bubble
during an economic expansion, and experience a steeper decline in housing
prices during an economic recession.
Specifically, the micro foundation of this housing price patterns is likely to come
from households’ speculative behaviors. The channels through which recourse law
influences households’ investment behavior can be divided into leverage decision and
asset allocation decision. The limited borrower’s liability may drive households to
invest in their house with higher debt-to-income ratio. The high levered investment
will increase their returns without bearing additional downside risk.
Hypothesis 2: Households in a state with non-recourse law invest in hous-
ing assets with higher debt-to-income ratio during an economic expansion
than in a state with recourse law.
Additionally, households can have different asset allocations over the recourse law
of their states. We expect that households in non-recourse states may allocate more
their capitals into housing assets in the anticipation of higher return.
Hypothesis 3: Households in a state with non-recourse law allocate more
capital on housing assets during an economic expansion than in a state
with recourse law.
Having established a role for recourse law in housing market bubbles and the
microfoundation, we turn to the question of whether mortgage lender behavior differs
between recourse and non-recourse law states. The Allen and Gale model assumes
the lending market to be competitive with unlimited credit supply, and lenders to
not observe the riskiness of assets. In practice, however, mortgage lenders can be
constrained by market incompleteness, capital market frictions, and regulatory capital
requirements (Stein (2007)). Moreover, mortgage lenders can exercise some control
over the riskiness of lending through down payments, mortgage spreads, and screening
of borrowers. The corresponding hypothesis is that lenders in non-recourse states, to
minimize costs from the lack of deficiency judgment, charge a higher mortgage interest
rate and down payment, lower loan-to-value ratio and stricter loan screening than
lenders in recourse states.
6
H4: Mortgage lenders in non-recourse states charge a higher mortgage
interest rate and stricter loan screening than lenders in recourse states.
It will be surprising if we observe larger housing bubble in non-recourse states even
in the presence of mortgage lenders’ control on additional risk. Literature suggests
that the originate-to-distribute (OTD) market, by enabling originators to shift risk
to other investors by securitizing mortgage loans and reselling them to third parties,
mitigates constraints in credit supply and the ex-ante incentive to screen borrowers
(Purnanandam (2011)).2 It is likely that the origins of loans are concealed when
loans are securitized in a complex structure of financial derivatives. Piskorski, Seru,
and Witkin (2013) argue that the true quality of loans in mortgage backed security
(MBS) market has been misreported to investors frequently. They show that one
out of ten loans in RMBS market has misrepresentation in occupancy status of
borrowers or second liens information, which is not priced in the securities at their
issuance. To the extent that it does not reflect the embedded risk in non-recourse
mortgage loans, the OTD market promotes a disproportionately large increase in poor
quality loan originations in non-recourse states. This is consistent with the argument
that the OTD model induces excessively risky mortgage loan originations (Pennacchi
(1988); and Gorton and Pennacchi (1995)). The corresponding hypothesis is as follows.
H5: More sub-prime mortgage loans are originated in non-recourse than
in recourse states.
3 Data
3.1 Housing Market Data
Housing market data used in this study are from Zillow Real Estate Research
(www.zillow.com). The Zillow database, which is widely used in related literatures,
provides ZIP code-level housing price data at the monthly level from 1999-2011 3.
The 12,397 ZIP codes included represent 42% of U.S. ZIP codes. For each ZIP code,
we use as a measure of housing price the median of sale prices scaled by a home’s
square footage. Alternatively, we use the median of the total prices of homes sold. We
2Rapid expansion of this market was also accompanied by a relaxation of the regulation of mortgagelending.
3Our empirical results will be updated upon receipt of Census data from the American Housing Sur-vey (AHS) (which provides detailed, tract-level information about housing and household characteristicsincluding household-level panel data for each property), for which we have submitted a request.
7
calculate the rate of annual growth in housing price at time t based on the price in
January in period t and t+1.
3.2 Households Investment Behavior
First, we construct average debt-to-income ratio at ZIP code-level using HMDA
data as a proxy for leverage decision of households’ investment in housing market.
HMDA data contains individual mortgage amount and income level of borrowers at
the origination of mortgage.
Second, we construct state-level households’ asset allocation in housing assets to
examine households’ speculative investment motive. The Panel Study of Income Dy-
namics (PSID) provides a wide range of households’ portfolio data including total asset
value, income, expenditure, and household demographic information. The data set is
based on the survey that the PSID has conducted to more than 8000 households every
two years. We employ the PSID data from 1999 to 2009 and estimate the asset allo-
cation of households on housing assets by the fraction of home equity to total wealth.
The home equity is the value of house minus the first and the second mortgage on
the house. The total wealth is the sum of home equity, farm/business assets, check-
ing/saving accounts, stocks, vehicles, annuities, other assets and other real estate assets
minus total debt. As we examine the home equity share of each household in different
states, we can understand how households response to the variation of housing value
during the period of housing market bubble and burst.
3.3 Mortgage Interest Rate
We obtain mortgage market data from the Home Mortgage Disclosure Act (HMDA)
data set, which provides loan and application information by geographic area at the
Census tract-level. Loan information includes property type, owner-occupancy, mort-
gage spread, and loan type, amount, purpose, and resale information. Application
information includes ethnicity, race, gender, and annual income. Because the HMDA
data contains Census tract-level data, we construct ZIP code-level HMDA data by
matching Census tracts to ZIP codes using the U.S. Department of Housing and Ur-
ban Development’s (HUD’s) ZIP code and Census tract crosswalk file. Our analysis
focuses on conventional loans for home purchases (Loan Type=1, Loan Purpose =
1). 4 We use the ZIP-code level mortgage spread, the difference between a mortgage
loan’s annual percentage rate (APR) and the rate for Treasury securities of comparable
4Mortgage loans are classified by the loan type (1.Conventional; 2.Federal housing Administration-insured;3.Veterans Administration-guaranteed; and 4.Farm Service Agency or Rural Housing Service), and loanpurpose (1.Home purchase; 2.Home improvement; and 3.Refinancing)
8
maturity measured at the time of loan orgizination.
3.4 Denial Rate
The mortgage application can be denied by financial institution. The reasons for
denial are related to 1.Debt-to-income ratio; 2.Employment history; 3.Credit history;
7Among 11 non-recourse states, 3 states (Iowa, North Dakota and Wisconsin) have the judicial foreclosurerequirement
8For example, New York’s Erie County and Westchester County have similar populations of 0.75 million,but median household income levels of $47,533 and $77,006, respectively, whereas Connecticut’s FairfieldCounty is contiguous with, and has socioeconomic characteristics similar to those of, Westchester county.
where Denial Rateit is the average denial rate in ZIP code-level. The detailed
definition of denial rate is described in Section 3.5. As in the spread test, our main
hypothesis predicts a positive jump of denial rate at the border when one crosses into
the non-recourse states, corresponding to β1 > 0.9
9We also intend to examine the impact of recourse law on loan-to-value ratio using the same specification.An analysis of LTV data, which we are in the process of acquiring from LPS, will be added in the next version
16
6.1 Results
Table 5 estimates the impact of recourse law on mortgage lending behaviors in-
cluding mortgage interest rate and denial rate. This analysis is based on the pre-crisis
sample since the effect of mortgage market collapse on mortgage interest rate is un-
clear as government policy and decreased demand in housing market are confounded
in the result. The denial rate is also likely to be biased by different selections of loan
applicants.
In Model (1)-(3), we test the recourse law on mortgage spread (%) with the contigu-
ous state border county-pair sample. The estimates present insignificant coefficients on
Non-recourse dummy in Model (1) and (2). However, in our preferred Model (3) with
state-border discontinuity, we find that the coefficient on Non-recourse is positive and
statistically significant at 5% level, which implies a positive jump in mortgage spread
when one crosses into non-recourse states.
Model (4)-(6) estimate the impact of recourse law on denial rate of loan application
using the contiguous state border county-pair sample in the pre-crisis period. The
estimates are positive and statistically significant at 1-5% level in all models and the
economic magnitude of Non-recourse on denial rate is large at 2 and 3% in Model (4)
and (5), respectively. The estimates become larger at 7% in Model (6) where we use the
state border discontinuity design. This implies that loan applications in non-recourse
states are more likely to be denied by 7 percentage points. The size is economically
meaningful considering that the average denial rate is xx% in the aggregate economy.
The overall results suggest that mortgage lenders are aware of the additional risk
embedded in non-recourse mortgage loan, charging a higher interest rate and denying
more frequently. The next question of this paper is then, why we observed larger
housing bubbles in non-recourse states in spite of the lenders’ control for additional
risk.
7 Recourse Effects on Sub-prime Mortgage
We conjecture that the surprising finding that housing price experiences larger
bubbles in non-recourse states in the presence of lenders’ control is resulted from the
emergence of the OTD market, which enables lenders to effectively shift the risk of
those costs to other investors. Said differently, the mortgage lending behavior does
not fully reflect the higher risk in non-recourse states. This two-stage risk-shifting
hypothesis predicts that more sub-prime mortgage loans are originated in non-recourse
than in recourse states.
of this paper. We will focus on initial LTV at the time of home purchase in order to minimize the effect ofsubsequent changes in housing price.
17
Figure 3 presents the time-series behavior of the aggregate sub-prime ratio in both
recourse and non-recourse states. Similar to the time-series pattern of housing price
growth rate, these sub-prime ratios move in a similar pattern over time, but greater
volatility is observed in non-recourse states. Consistent with our hypothesis, the sub-
prime loan ratio in non-recourse states is higher during the pre-crisis period of 2002-
2006, but falls below that of recourse states during the recent crisis period from 2007.
To test this relation, we first test the recourse effect on the sub-prime mortgage
ratio, calculated by the number of sub-prime mortgage loans divided by the total
number of mortgage loans originated using the specification in below.
where the dependent variable is the growth rate of housing price per square foot in
ZIP code i at time t. The interaction term between Sub-primeit and Non-recoursei
is the main variable of interest. Our hypothesis predicts positive coefficient on this
coefficient, or β3 > 0.
7.1 Results
Table 6 presents the results of the regression estimations on sub-prime loan ratio.
In Model (1)-(3), we use the contiguous border county-pair sample in the pre-crisis
period. We include state-level control variables in Model (2) and add distance measures
in Model (3) to test state-border discontinuity. The results in Model (1)-(2) show that
larger fraction of sub-prime loans is originated in non-recourse states than in recourse
states. The coefficients on Non-recourse dummy variable are positive and statistically
significant at 5-10% level. The economic magnitude of this effect is 3-5%. We also
find the consistent results from the state border discontinuity test. The results show
lenders in non-recourse states to originate, on average, 8 percentage points more sub-
prime loans than lenders in recourse states. These results are consistent with our
hypothesis, which states that the OTD market encourages more risk shifting by lenders
in non-recourse states than by lenders in recourse states.
Model (4)-(6) in Table 6 present the regression results with the difference-in-
difference specification. Non-recourse remain positive and statistically significant at
1-5% level. The economic magnitude of coefficient is 3-9%, similar to the previous
specification. A negative and significant coefficient on the Crisis dummy variable in-
dicates that, on average, sub-prime loan ratio declined significantly following the crisis
in 2007. In Model (4)-(6), the coefficient accounts for 8 percentage points decrease in
sub-prime loan ratio relative to the pre-crisis period at the state border. The interac-
tion effect of Non-recourse and Crisis shows a negative coefficient but the statistical
significance is marginal.
In Table 7, we report the interaction effect of Non-recourse dummy variable and
Sub-primeit on the housing price growth. We employ the total sample from 2005-2008
in Model (1)-(3) and then focus on the contiguous border county-pair sample in Model
(4)-(6).
The coefficient on Sub-primeit is positive and significant. The economic significance
of Sub-primeit is 1.6 - 2.6 percentage points higher housing price growth when one
19
standard deviation increase of sub-prime loan ratio 10. More importantly, the coefficient
on the interaction term is positive and statistically significant at 1% level in consistent
with our hypothesis. The economic magnitude of this effect can be interpreted as 1.3-
2.6 percentage points higher housing price growth in Model (1)-(3). In Model (4)-(6),
the coefficient on the interaction term remains economically large at 1.1 - 1.6 percentage
points and statistically significant at 1% level.
It is important to note that the coefficient on Non-recourse dummy variable becomes
negative (Model 1), smaller (Model 2 and 3) or insignificant (Model 4 and 6) relative to
the previous results in Section 4. This result implies that, without the two-stage risk-
shifting by the OTD market, the impact of non-recourse law on households’ investment
behavior is better controlled by mortgage lenders.
The overall results demonstrate the underlying mechanism of the recent housing
bubbles and, in particular, why we observe larger bubbles in the non-recourse states.
8 Conclusion
In this paper, we investigate the role of state-level variation in mortgage recourse
law in the creation of a bubble in the housing market. We perform on contiguous state
border pair-counties a state border discontinuity test combined with a difference-in-
difference setting using the mortgage market collapse in 2007 as an exogenous shock.
The results, which are economically large and robust, show that states with non-
recourse law experience a larger bubble and burst in housing prices. Our evidence
supports the bubble mechanism in the asset substitution problem, as proposed by
Allen and Gale (2000). We also examine the effect of recourse law on lending behavior.
Although we find evidence that mortgage lenders are aware of the additional risk in
non-recourse loan, the higher sub-prime loan ratio in non-recourse states suggests that
the OTD market enables lenders to effectively shift risk to other investors.
Comparing benefits and costs of recourse law is left to future research. The bubble
and burst cycle in the housing market has been repeated and amplified in non-recourse
states. Further evaluation of mortgage recourse law is needed to prevent future housing
market crises and collapses.
10The standard deviation of sub-prime loan ratio is 0.13. The economic magnitude is calculated by0.13*0.20= 0.026 in Model (1) and 0.13*0.12 = 0.016 in Model (3).
20
References
Abreu, D., and M. K. Brunnermeier (2003): “Bubbles and crashes,” Economet-rica, 71(1), 173–204. 4
Allen, F., and D. Gale (2000): “Bubbles and crises,” The economic journal,110(460), 236–255. 2, 4, 5, 16, 20
Allen, F., and G. Gorton (1993): “Churning bubbles,” The Review of EconomicStudies, 60(4), 813–836. 4
Allen, F., S. Morris, and A. Postlewaite (1993): “Finite bubbles with shortsale constraints and asymmetric information,” Journal of Economic Theory, 61(2),206–229. 4
Brunnermeier, M., and S. Nagel (2004): “Hedge funds and the technology bub-ble,” The Journal of Finance, 59(5), 2013–2040. 4
Brunnermeier, M. K. (2003): Asset pricing under asymmetric information: Bubbles,crashes, technical analysis, and herding. OUP Oxford. 4
Dube, A., T. W. Lester, and M. Reich (2010): “Minimum wage effects acrossstate borders: Estimates using contiguous counties,” The review of economics andstatistics, 92(4), 945–964. 12
Ghent, A. C., and M. Kudlyak (2011): “Recourse and residential mortgage default:evidence from US states,” Review of Financial Studies, 24(9), 3139–3186. 4, 5
Gorton, G. B., and G. G. Pennacchi (1995): “Banks and loan sales marketingnonmarketable assets,” Journal of monetary Economics, 35(3), 389–411. 7
Herring, R., and S. Wachter (2003): “Bubbles in real estate markets,” Asset PriceBubbles: The Implications for Monetary, Regulatory, and International Policies, pp.217–27. 2, 4
Holmes, T. J. (1998): “The effect of state policies on the location of manufacturing:Evidence from state borders,” Journal of Political Economy, 106(4), 667–705. 12
Hong, H., J. Scheinkman, and W. Xiong (2008): “Advisors and asset prices: Amodel of the origins of bubbles,” Journal of Financial Economics, 89(2), 268–287. 4
Makarov, I., and G. Plantin (2013): “Equilibrium subprime lending,” The Journalof Finance. 2, 4
Mayer, C., K. Pence, and S. M. Sherlund (2009): “The rise in mortgage de-faults,” Journal of Economic Perspectives, 23(1), 27–50. 2, 4
Mian, A., and A. Sufi (2009): “The consequences of mortgage credit expansion:Evidence from the US mortgage default crisis,” The Quarterly Journal of Economics,124(4), 1449–1496. 4, 9
21
Mian, A., A. Sufi, and F. Trebbi (2013): “Foreclosures, house prices, and the realeconomy,” Discussion paper, National Bureau of Economic Research. 4, 12
Pavlov, A., and S. Wachter (2004): “Robbing the bank: non-recourse lending andasset prices,” The Journal of Real Estate Finance and Economics, 28(2-3), 147–160.4
Pavlov, A., and S. M. Wachter (2006): “The inevitability of marketwide under-pricing of mortgage default risk,” Real Estate Economics, 34(4), 479–496. 4
Pence, K. M. (2006): “Foreclosing on opportunity: State laws and mortgage credit,”Review of Economics and Statistics, 88(1), 177–182. 4, 12
Pennacchi, G. G. (1988): “Loan sales and the cost of bank capital,” The Journal ofFinance, 43(2), 375–396. 7
Piskorski, T., A. Seru, and J. Witkin (2013): “Asset Quality Misrepresentationby Financial Intermediaries: Evidence from RMBS Market,” . 7
Purnanandam, A. (2011): “Originate-to-distribute model and the subprime mort-gage crisis,” Review of Financial Studies, 24(6), 1881–1915. 3, 4, 7
Reinhart, C., and K. Rogoff (2008): “Banking Crises: An Equal OpportunityMenace,” NBER Working Paper Series, 14587. 2, 4
Reinhart, C. M., and K. S. Rogoff (2009): “The Aftermath of Financial Crises,”The American Economic Review, 99(2), 466–472. 2, 4
Saiz, A. (2010): “The geographic determinants of housing supply,” The QuarterlyJournal of Economics, 125(3), 1253–1296. 27
Scheinkman, J. A., and W. Xiong (2003): “Overconfidence and Speculative Bub-bles,” Journal of Political Economy, 111(6). 4
Solomon, D., and O. Minnes (2011): “Non-Recourse, No Down Payment and theMortgage Meltdown: Lessons from Undercapitalization,” Fordham J. Corp. & Fin.L., 16, 529. 5
Stein, J. C. (2007): “An Adverse-Selection Model of Bank Asset and Liability Man-agement with Implications for the Transmission of Monetary Policy,” RAND Journalof Economics, 29(3), 466–486. 6
Tirole, J. (1982): “On the possibility of speculation under rational expectations,”Econometrica: Journal of the Econometric Society, pp. 1163–1181. 4
(1985): “Asset bubbles and overlapping generations,” Econometrica: Journalof the Econometric Society, pp. 1499–1528. 4
22
Figure 1: State-level variation in mortgage Recourse lawThis figure shows the classification of mortgage recourse law. States with shaded inblue represents non-recourse states. The states with non-recourse law are Alaska, Arizona,California, Iowa, Minnesota, Montana, North Carolina (purchase mortgages), North Dakota,Oregon, Washington, and Wisconsin.
23
Figure 2: Recourse Law and Housing Price Growth RateThis figure presents time-series behavior of aggregate housing price growth rate per squarefoot over recourse law.
24
Figure 3: Recourse Law and Sub-Prime RatioThis figure presents time-series trend of average sub-prime ratio over recourse law.
25
Figure 4: Regression Discontinuity at State BorderThis figure plots the housing price growth rate (Sq.Ft) with the estimates of Kernelregression for ZIP codes that are near borders where the recourse law changes across states.The first graph is for the period from 2005-2007 and the second graph is for the periodfrom 2000-2006. We generate the graphs by regressing the housing price growth rate on1-mile band distance-to-the-border dummies where the dummies have negative values fornon-recourse states and plot the coefficients on the dummies. The border is at 0.
26
Table 1: Summary StatisticsThis table provides the summary statistics of main variables from 1999-2011 used in theanalysis. Housing Price Growth (Sq. Ft) is the percentage annual growth rate of the medianof sale prices scaled by the square footage of a home. Housing Price Growth representsthe percentage annual growth of the median of sale prices without scaling. Both measuresfor housing price are aggregated at ZIP code-level. Mortgage spread is the differencebetween the annual percentage rate (APR) on mortgage loan and the rate on Treasurysecurities of comparable maturity. Sub-prime loan ratio is the aggregate ratio of the numberof sub-prime mortgage loans to the total number of mortgage loans originated in ZIPcode-level. The other variables are all state-level statistics. GDP Growth rate is the annualpercentage growth rate of nominal GDP. Income Growth Per Capita is the growth rate of(total income/population). Unemployment rate is the annual unemployment rate. Housingsupply elasticity is provided by Saiz (2010). Mortgage Growth Per Capita is the growthrate of total mortgage origination/population.
Panel A. Total Sample Mean Std. Dev. 10th 50th 90th N
Mortgage Growth Per Capita 0.08 0.08 -0.03 0.08 0.17 612
27
Table 2: Univariate AnalysisThis table provides the comparisons of main variables between Recourse state and non-Recourse state. Panel A presents the statistics during the sample period from 1999-2006(Expansion). Panel B presents the statistics during the sample period from 2007-2011(Recession). State are classified as the Recourse states if the state allows lender to claimdeficiency judgment in case of mortgage default. We present the difference between theaverage value in recourse state and in non-recourse state. ***, ** and * are significant atthe 1%, 5% and 10% level, respectively.
Recourse Non-recourse Diff.
Panel A. Expansion (1999-2006) Mean SD p50 Mean SD p50 -
Table 3: Recourse Law and Housing Price GrowthThe table reports estimates and standard errors, in parentheses, of a regression of housingprice growth on non-recourse law indicators in 2005-2008 for the full sample and contiguousborder county-pair sample. The dependent variable is Housing Price Growth (Sq. Ft), thepercentage annual growth rate of the median of sale prices scaled by the square footage of ahome. This measure is aggregated at ZIP code-level. Crisis is a dummy variable equals tozero before and including 2006, and one after that. A state is classified as the Non-recoursestate (Non-recourse = 1) if the state does not allow lender to claim deficiency judgment incase of mortgage default. Distance is the shortest distance between the closest border andthe centroid of ZIP code. DistanceR represents the interaction of distance and an indicator(I(recourse)). DistanceNR represents the interaction of distance and an indicator (I(non-recourse)). The other control variables are all state-level statistics and defined in Table1. Coefficients marked ***, ** and * are significant at the 1%, 5% and 10% level, respectively.
Total Sample County-Pair Sample(1) (2) (3) (4) (5) (6)
Table 4: Recourse Law and Households’ Investment BehaviorThe table reports estimates and standard errors, in parentheses, of a regression of house-holds’ investment behavior for contiguous border county-pair sample in the pre-crisis periodfrom 2005-2006. In column (1)-(3), the dependent variable is the average ratio of homeequity to total wealth at state level. In column (4)-(6), the dependent variable is theaverage debt-to-income ratio at ZIP code-level. A state is classified as the Non-recoursestate (Non-recourse = 1) if the state does not allow lender to claim deficiency judgment incase of mortgage default. Distance is the shortest distance between the closest border andthe centroid of ZIP code. DistanceR represents the interaction of distance and an indicator(I(recourse)). DistanceNR represents the interaction of distance and an indicator (I(non-recourse)). The other control variables are all state-level statistics and defined in Table1. Coefficients marked ***, ** and * are significant at the 1%, 5% and 10% level, respectively.
Table 5: Recourse Law and Mortgage Lending BehaviorThe table reports estimates and standard errors, in parentheses, of a regression of mortgagelending behavior for the contiguous border county-pair sample in the pre-crisis period.In column (1)-(3), the dependent variable is mortgage spread(%), the difference betweenthe annual percentage rate (APR) on mortgage loan and the rate on Treasury securitiesof comparable maturity. This measure is aggregated at ZIP code-level from Censustract-level data. In column (4)-(6), the dependent variable is denial rate of mortgage loanapplications aggregated at ZIP code-level. A state is classified as the Non-recourse state(Non-recourse = 1) if the state does not allow lender to claim deficiency judgment in caseof mortgage default. Distance is the shortest distance between the closest border and thecentroid of ZIP code. DistanceR represents the interaction of distance and an indicator(I(recourse)). DistanceNR represents the interaction of distance and an indicator (I(non-recourse)). The other control variables are all state-level statistics and defined in Table1. Coefficients marked ***, ** and * are significant at the 1%, 5% and 10% level, respectively.
Table 6: Recourse Law and Sub-prime Loan RatioThe table reports estimates and standard errors, in parentheses, of a regression of sub-primemortgage loan ratio for the contiguous border county-pair sample in 2005-2008. Model(1)-(3) employ the sample in the pre-crisis period from 2005-2006 and Model (4)-(6)employ the sample from 2005-2008 using the difference-in-difference approach. Crisisis a dummy variable equals to zero before and including 2006, and one after that. Thedependent variable is the aggregate ratio of the number of sub-prime mortgage loans tothe total number of mortgage loans originated in ZIP code-level. A state is classified asthe Non-recourse state (Non-recourse = 1) if the state does not allow lender to claimdeficiency judgment in case of mortgage default. Distance is the shortest distance betweenthe closest border and the centroid of ZIP code. DistanceR represents the interaction ofdistance and an indicator (I(recourse)). DistanceNR represents the interaction of distanceand an indicator (I(non-recourse)).The other variables are all state-level statistics. Theother variables are all state-level statistics and defined in Table 1. Coefficients marked ***,** and * are significant at the 1%, 5% and 10% level, respectively.
Table 7: Interaction of Recourse Law and Sub-prime Loan Ratio on Housing Price GrowthThe table reports estimates and standard errors, in parentheses, of a regression of housingprice growth on the interaction term of non-recourse law indicators and sub-prime loanratio in 2005-2008 for the full sample and contiguous border county-pair sample. Thedependent variable is Housing Price Growth (Sq. Ft), the percentage annual growth rateof the median of sale prices scaled by the square footage of a home. This measure isaggregated at ZIP code-level. The sub-prime loan ratio is the aggregate ratio of the numberof sub-prime mortgage loans to the total number of mortgage loans originated in ZIPcode-level. A state is classified as the Non-recourse state (Non-recourse = 1) if the statedoes not allow lender to claim deficiency judgment in case of mortgage default. Distance isthe shortest distance between the closest border and the centroid of ZIP code. DistanceR
represents the interaction of distance and an indicator (I(recourse)). DistanceNR representsthe interaction of distance and an indicator (I(non-recourse)).The other variables are allstate-level statistics. The other variables are all state-level statistics and defined in Table1. Coefficients marked ***, ** and * are significant at the 1%, 5% and 10% level, respectively.
Total Sample County-Pair Sample(1) (2) (3) (4) (5) (6)