Prepayment Penalties: Efficiency and Predation
October 2011
Morgan J. Rose
University of Maryland, Baltimore County
Department of Economics
1000 Hilltop Circle
Baltimore, MD 21250
Office of the Comptroller of the Currency
Visiting Scholar
250 E Street, SW
Washington, DC 20219
Abstract
This paper presents evidence that reductions in mortgage interest rates associated with
prepayment penalties are greater for riskier borrowers, as measured by mortgage type, credit
scores, and local incomes and education levels. This is consistent with an efficiency view that,
by reducing the reclassification risk faced by lenders, prepayment penalties can be welfare-
improving. Additional findings indicate that prepayment penalties are also used as a predatory
lending tool, but that the efficiency view dominates the predatory view in most circumstances.
State anti-predatory lending laws restricting the duration and amount of prepayment penalties
appear to curb the predatory use of prepayment penalties.
JEL Classifications: G21, G28, G01, D18, L85
Key words
: prepayment penalties; predatory lending; financial regulation; mortgage crisis;
reclassification risk
The views expressed herein do not reflect those of the Office of the Comptroller of the Currency
or the Department of the Treasury. All errors are my own.
1. Introduction
A prepayment penalty requires a borrower to pay a substantial fee if he or she repays a
mortgage loan within a specified time period following the origination of the loan. Although
prepayment penalties are rare among prime mortgages, they are prevalent in the subprime
market. They are also quite controversial. Critics consider prepayment penalties to be predatory
loan features that trap borrowers in high-cost loans, stripping borrowers of wealth and making
mortgage defaults more likely, especially among more vulnerable groups of borrowers.
Supporters argue that prepayment penalties are a means of protecting lenders from risks
associated with borrowers repaying mortgages early, and so allow lenders to offer more
affordable loans with lower interest rates, particularly for the riskiest borrowers.
These two views, which I term the “predatory view” and the “efficiency view” and
discuss more fully below, imply several empirically testable hypotheses regarding the
relationship between prepayment penalties and loan interest rates across different types of loans.
This paper uses a sample of nearly 200,000 subprime loans originated over 2003-2006 to test the
hypotheses across subsets of loans in the years leading up to the subprime mortgage crisis. In
general, the findings are more consistent with the efficiency view than the predatory view, with
riskier or more vulnerable borrowers receiving greater reductions in loan interest rates in
exchange for accepting loans with prepayment penalties. However, findings associated with
certain types of loans and the effects of state anti-predatory lending (APL) law provisions that
limit the use of prepayment penalties support the predatory view. Taken together, the results
lend credence to both views, but suggest that in most circumstances the efficiency view
dominates the predatory view in terms of overall effects on the pricing of prepayment penalties.
1
Advocates of the predatory view argue that prepayment penalties are abusive loan
features that strip wealth from borrowers by trapping them in expensive loans, requiring
borrowers to either continue making high monthly payments, pay a substantial prepayment fee,
or default. Borrowers whose credit improves such that they could qualify for lower interest rate
loans may be unable to afford a refinancing if it requires a large prepayment penalty. For
financially distressed borrowers who cannot afford their current monthly mortgage payments,
prepayment penalties that make refinancing or selling the house prohibitively expensive will
drive them toward default. Quercia et al. (2007), Danis and Pennington-Cross (2008), Rose
(2008), Pennington-Cross and Ho (2010), and Demyanyk and Van Hemert (2011) all find that
prepayment penalties are associated with greater probabilities of default, although in Rose (2008)
and Pennington-Cross and Ho (2010) this result is somewhat dependent on the specification and
type of loan used.
Prepayment penalties are also related to yield spread premiums, a form of mortgage
originator compensation that increases with the difference between the loan interest rate and a
benchmark rate for loans with similar characteristics set by the lending institution. When a
lending institution offers a third-party originator higher compensation for originating a loan with
a higher interest rate, a requirement that that the loan carry a prepayment penalty deters the
borrower from quickly refinancing into a less expensive loan. With a prepayment penalty in
effect, the purchasing institution can recoup the higher originator compensation through the
collection of either the higher monthly payments or the prepayment fee. This type of originator
compensation provides originators with an incentive to steer borrowers into more expensive
loans.
1
Berndt et al. (2010) and Ernst et al. (2008) provide examples of “rate sheets” that
1
For analyses of originator compensation via yield spread premiums, see Jackson and Burlingame (2007),
Woodward (2008), and Woodward and Hall (2010).
2
explicitly link the yield spread premium a lending institution is willing to pay originators to the
presence of a prepayment penalty. To the extent that yield spread premiums are conditional on
loans having prepayment penalties, prepayment penalties may be associated with increases in
loan interest rates.
Under the predatory view, borrowers receive no substantial benefits from prepayment
penalties. Lenders are therefore expected to originate more loans with prepayment penalties to
groups of borrowers that are more vulnerable to predatory lending practices, and the relationship
between prepayment penalties and loan interest rates is expected to be less favorable for those
groups as well. Such groups include less financially sophisticated borrowers who may be
unaware of either the range of loans for which they could qualify or the implications of their loan
terms, and financially constrained borrowers who may be more likely to accept the terms given
by a particular originator rather than shop around.
2
Rose (2011) finds that prepayment penalties
are more prevalent among subprime loans originated in locales with lower education levels,
household incomes, and resident ages. Woodward (2008) presents evidence that the total
charges paid at origination are negatively related to the education levels in a borrower’s census
tract.
The efficiency view is most explicitly laid out in a theoretical model by Mayer et al.
(2010). The model posits two reasons why lenders charge higher rates to riskier borrowers: (1)
riskier borrowers are more likely to default, and (2) riskier borrowers exhibit greater
“reclassification risk,” a term used here to denote the probability that borrowers who receive a
2
According to McCoy (2007), subprime borrowers are often required to pay originators hefty application and
appraisal fees prior to learning the interest rates and terms of a mortgage, which can make comparison shopping
prohibitively expensive for financially constrained borrowers. The same is not true in the prime mortgage market.
3
positive credit shock will prepay by refinancing into lower interest rate loans.
3,4
Lenders
anticipate that over time, riskier mortgage pools will see more borrowers refinance out of them
as positive credit shocks occur, causing lenders to charge ex ante higher loan rates to riskier
borrowers. Prepayment penalties impede refinancing and therefore reduce the reclassification
risk faced by lenders, allowing lenders to offer lower interest rates. Reclassification risk is
greater for riskier borrowers, and so the riskiest borrowers should see the largest reductions in
loan interest rates associated with prepayment penalties. The lower interest rates available due to
these “prepayment penalty discounts” make mortgages more affordable, which should both
expand credit availability and reduce the likelihood of default, with the greatest benefits accruing
to the riskiest borrowers.
Most studies of the pricing of prepayment penalties, including DeMong and Burroughs
(2005), Ernst (2005), Elliehausen et al. (2008), LaCour-Little and Holmes (2008), Mayer et al.
(2010), and Rose (2011), have found that loan interest rates are significantly lower for loans with
prepayment penalties, although there have been exceptions for certain types of loans. Ernst
(2005) finds that prepayment penalties are associated with higher loan rates for purchase
subprime fixed-rate mortgages (FRMs), but are not related to refinance FRM loan rates. Rose
(2011) finds that while the initial interest rates on subprime adjustable-rate mortgages (ARMs)
are lower for loan with prepayment penalties, the interest rates on those loans are subsequently
adjusted to greater margins above the prevailing market rates to which ARMs are indexed.
3
Avery et al. (2005) state that “Borrowers in the higher-priced segment of the home-loan market have higher
prepayment rates than others because many of them improve their credit profiles over time as they make regular
payments, and this improvement in turn allows them to qualify for a lower rate loan…. For a higher-priced loan, a
small improvement in the borrower’s credit history score may translate into a substantial reduction in interest rates
and may encourage prepayment,” (page 369).
4
The term “reclassification risk” is usually used in the context of life and health insurance markets, in which the
revelation of negative information about a person’s health can result in increased premiums. Hendel and Lizzeri
(2003) find that commitments to long-term life insurance contracts reduce this reclassification risk and improve
welfare relative to short-term contracts.
4
Mayer et al. (2010) find that while prepayment penalties are associated with lower loan rates for
most subprime FRM borrowers, they are associated with higher rates for subprime FRM
borrowers with high FICO scores. This is consistent with their model’s prediction that loan rate
reductions associated with prepayment penalties should be greater for riskier borrowers, but their
model provides no explanation for why prepayment penalties would be associated with higher
loan rates for some borrowers.
As discussed in the next section, the efficiency view and predation view imply several
contradictory predictions concerning the changes in loan interest rates associated with
prepayment penalties for different groups of borrowers. This is because the borrowers who are
likely to receive the greatest benefits from prepayment penalties under the efficiency view are
often the borrowers most likely to be harmed by prepayment penalties under the predatory view.
Borrowers with binding financial constraints or poor credit histories are the most likely to qualify
for better loans in the event of a positive credit shock, and so present lenders with the most
reclassification risk. The same borrowers are also the most likely to have fewer competing
sources of mortgage credit available, have less financial ability to shop around for favorable
mortgage terms, and, if credit history and financial constraints are linked to financial
sophistication, be less able to fully understand their loan terms.
The research design of this paper is to examine variations in prepayment penalty
discounts based on measures that capture borrower reclassification risk, credit histories, and
financial constraints or sophistication, in order to determine which view’s predictions are most
consistent with the empirical evidence. Mayer et al. (2010) and DeMong and Burroughs (2005)
both examine how prepayment penalty discounts vary based on borrower credit scores, but do so
using a single-equation approach that does not address the potential endogeneity between loan
5
interest rates, loan-to-value (LTV) ratios, and prepayment penalties. This paper uses a multiple-
equation instrumental variables approach to account for that endogeneity, and examines the
variation of prepayment penalty discounts along a greater number of margins relevant to the
efficiency and predatory views. The findings are more consistent with the efficiency view, in
that riskier borrowers, as defined by several measures, receive larger prepayment penalty
discounts than safer borrowers. However, selected findings are supportive of the predatory view.
These results suggest that while the effects of prepayment penalties described by the efficiency
view are predominant in most circumstances, some predatory use of prepayment penalties does
occur.
This paper makes several contributions to the growing literature on the pricing of
prepayment penalties. First, by explicitly drawing out multiple empirical implications of the
predatory and efficiency views for different loan and borrower characteristics, this paper presents
more detailed and direct empirical testing of the two views than the previous literature. Second,
the sample includes subprime mortgages originated during 2003-2006, while previous papers
(with the exception of Rose (2011)) use originations from 2004 or earlier. Demyanyk and Van
Hemert (2011) document the deterioration of subprime mortgage credit quality in the years
leading to the recent mortgage crisis. This weakening of underwriting standards suggests
increases in both reclassification risk and the potential scope for predatory lending, which I am
the first to exploit by tracking the evolution of prepayment penalty discounts over the sample
period. The findings indicate that as the years progressed, prepayment penalty discounts became
larger, with the increase concentrated among loans with the most reclassification risk. Third, this
is the first paper to empirically examine the role of prepayment penalty duration in the pricing of
prepayment penalties, finding that controlling for the duration of prepayment penalties
6
dramatically changes the estimates of prepayment penalty discounts. Finally, because my
sample captures variation across states and over time in state APL law provisions regarding
prepayment penalties, I provide the first evidence that prepayment penalty discounts are larger in
the presence of such restrictions. As is discussed below, this finding suggests that some
predatory use of prepayment penalties does occur, and that consumers may benefit from APL
laws that restrict, but do not prohibit, prepayment penalties. This is particularly relevant given
that the 2010 Dodd-Frank Wall Street Reform and Consumer Protection Act prohibits
prepayment penalties from some types of mortgages while restricting their duration and amount
in others.
5
The remainder of this paper is structured as follows. Section 2 presents hypotheses, and
Section 3 describes the data and econometric methodology used. Section 4 provides the results
of the empirical analyses. Section 5 discusses conclusions to be drawn from the results.
2. Hypotheses
This paper tests several hypotheses derived from the efficiency and predatory views. In
some cases the two views yield identical predictions, while in others the predictions conflict.
Note that the two views are not mutually exclusive, so where the predictions conflict one should
interpret the empirical evidence as indicating whether one view is or is not dominant, not
whether one view is or is not correct.
H1: The discount associated with a prepayment penalty is greater for refinance loans
than for purchase loans. This hypothesis holds under the efficiency view and the predatory view.
5
Title XIV, Section 1414 of the Dodd-Frank Act prohibits prepayment penalties entirely on all ARMs and certain
high-priced FRMs. For other FRMs, prepayment penalties cannot be imposed beyond three years after origination.
The amount of a prepayment penalty is limited to 3 percent, 2 percent, and 1 percent of the outstanding loan balance
in the first, second, and third years, respectively, after origination.
7
For the efficiency view, the salient fact concerning loan purpose is that borrowers of refinance
loans have already demonstrated a willingness to prepay a mortgage, and so reclassification risk
is likely to be higher for refinance loans than for purchase loans. This implies larger prepayment
penalty discounts on refinance loans. For the predatory view, the salient fact is that on average,
purchase loan borrowers are less experienced mortgage market participants than refinance loan
borrowers – by definition, a refinance is not a borrower’s first mortgage. Less experienced
borrowers are less likely to fully understand the terms of their loans and are more easily taken
advantage of, resulting in smaller prepayment penalty discounts on purchase loans.
H2: If the efficiency view is dominant, then the discount associated with a prepayment
penalty is greater for loans to less creditworthy, more financially constrained borrowers. If the
predatory view is dominant, then the discount associated with a prepayment penalty is lower for
loans to less creditworthy, more financially constrained borrowers.
6
According to the efficiency
view, the least creditworthy, most financially constrained borrowers are the borrowers who are
most likely to prepay their loans upon receiving a positive credit shock, and so those borrowers
should receive the largest prepayment penalty discounts. According to the predatory view, the
least creditworthy, most financially constrained borrowers are the most vulnerable to predatory
lending practices, due to either a lack of access to alternative mortgage credit sources or a link
between those characteristics and financial sophistication. As such, those borrowers should
receive the smallest prepayment penalty discounts.
H3: If the efficiency view is dominant, then in the years leading up to subprime mortgage
crisis the discount associated with a prepayment penalty increased, especially for refinance
loans and loans to less creditworthy, more financially constrained borrowers. If the predatory
6
The term “creditworthy” is used in this paper to refer only to borrowers’ credit histories, and not to any inherent
characteristics of borrowers.
8
view is dominant, then in the years leading up to subprime mortgage crisis the discount
associated with a prepayment penalty decreased, especially for purchase loans and loans to less
creditworthy, more financially constrained borrowers. As noted above, Demyanyk and Van
Hemert (2011) document a decline in subprime mortgage underwriting standards over the years
included in this paper’s sample. Under the efficiency view, the extension of subprime mortgage
credit deeper into the pools of more marginal potential borrowers should be associated with
increasing reclassification risk and therefore larger prepayment penalty discounts, and the effect
should be most pronounced among those groups of loans already associated with greater
reclassification risk. Under the predatory view, prepayment penalty discounts should fall as
credit is extended to less experienced, less financially sophisticated, or otherwise more
vulnerable borrowers.
H4: If the efficiency view is dominant, then conditional on loans having prepayment
penalties, prepayment penalty period durations are longer for refinance loans than for purchase
loans. If the predatory view is dominant, then conditional on loans having prepayment penalties,
prepayment penalty period durations are shorter for refinance loans than for purchase loans.
Under the efficiency view, borrowers with the greatest ex ante reclassification risk receive the
greatest benefit from having a prepayment penalty versus not having one. The same rationale
indicates that the marginal benefit of a longer prepayment penalty period versus a shorter one is
greatest for borrowers with the greatest ex ante reclassification risk. The predatory view
suggests that less experienced borrowers should receive more abusive loan terms, which includes
longer prepayment penalty period durations.
H5: Conditional on loans having prepayment penalties, prepayment penalty period
durations are longer for loans to less creditworthy, more financially constrained borrowers.
9
This hypothesis holds under the efficiency view and the predatory view. The efficiency view
predicts this because less creditworthy, more financially constrained borrowers are associated
with greater reclassification risk, while the predatory view predicts this because such borrowers
are more vulnerable to predatory lending practices.
H6: If an APL provision restricting the duration of prepayment penalty periods or the
amounts of prepayment penalties is effective at curbing predatory lending, then the discount
associated with a prepayment penalty is greater for loans originated with the APL provision in
effect. If such an APL provision does not substantially curb predatory lending, then the discount
associated with a prepayment penalty is lower for loans originated with the APL provision in
effect. Restrictions on the durations or amounts of prepayment penalties limit the protection
against reclassification risk that prepayment penalties can offer, which the efficiency view
suggests should reduce prepayment penalty discounts. Restrictions on the use of prepayment
penalties also reduce the incentives for the predatory use of prepayment penalties, which should
increase prepayment penalty discounts. To the extent that APL restrictions of prepayment
penalties discourage the most abusive uses of prepayment penalties while still allowing
prepayment penalties to provide lenders with substantial protection against reclassification risk,
then the net effect of the restrictions should be to increase prepayment penalty discounts. If APL
restrictions do not substantially reduce the predatory use of prepayment penalties, either because
the restrictions are ineffective or because such predatory use does not occur, then the net effect
of the restrictions should be to reduce prepayment penalty discounts.
3. Data and Methodology
10
The dataset for this paper consists of fixed-rate subprime mortgages for single family
residences originated during 2003-2006 in ten metropolitan statistical areas (MSAs) from the
LoanPerformance database from CoreLogic, Inc.
7
These are loans that were packaged into
subprime-grade private-label mortgage-backed securities. ARMs are excluded from the analysis.
The pricing of prepayment penalties for ARMs is complicated by the frequent use of low initial
“teaser” rates that remain fixed for some time after origination, after which loan rates become
indexed to a specified margin above a market interest rate.
8
Some borrowers take ARMs with
the intention of prepaying the loan before the first interest rate adjustment, further complicating
the pricing of prepayment penalties for ARMs.
Loans are taken from ten MSAs rather than a nationwide sample because the analysis
includes data on specific provisions of state APL laws, and an in-depth survey of the APL
provisions of all fifty states is beyond the scope of this paper. The selection of MSAs was based
on a report from RealtyTrac, Inc. (2008), providing 2007 foreclosure rates for the hundred
largest metropolitan areas in the United States. To ensure that the sample MSAs represent both a
substantial number of American households and a diverse range of mortgage market difficulties,
I divided the MSAs with populations over one million inhabitants into deciles based on the
reported foreclosure rates. From each decile I selected the MSA with the highest population,
with the condition that only one MSA from any state be included to ensure geographic diversity.
7
Mayer and Pence (2009) compare the LoanPerformance data’s coverage of subprime origination to the coverage of
two other sources, loans originated by lenders appearing on the list of subprime lenders maintained by HUD and
higher-priced loans identified since 2004 in data collected under the auspices of the Home Mortgage Disclosure Act.
The authors conclude that during the mid-2000s, the LoanPerformance data likely provide the most reliable
coverage of subprime originations.
8
Rose (2011) finds that prepayment penalties are associated with lower initial loan rates (14-21 basis points) for
refinance ARMs and higher initial loan rates (7 bp) for purchase ARMs. Prepayment penalties are associated with
46 and 73 bp increases in margins for refinance and purchase ARMs, respectively. This is consistent with findings
presented in the next section, in which the pricing of prepayment penalties is more favorable for FRM refinance
borrowers than to purchase FRM borrowers.
11
The ten MSAs are listed in Table 1.
9
To better control for loan terms and amortization schedules
that could affect the loan characteristics of interest, sample loans are limited to those with those
with maturities of fifteen or thirty years, and balloon and interest-only loans are excluded.
The LoanPerformance data contains loan-level information on whether a loan has a
prepayment penalty at origination (PP), the duration of the prepayment penalty period (if any),
the loan interest rate (InitialRate), LTV ratio (LTV), and borrower FICO score (FICO) at
origination, whether the loan was based on low- or-no documentation (LowNoDoc), whether the
loan is for an owner-occupied property (OwnerOcc), the loan purpose (refinance or purchase),
whether the borrower extracted cash out (RefiCash) or not (RefiNoCash) in a refinance loan, and
whether the loan term is thirty years (30Year) rather than fifteen years. RelLoanSize is calculated
as the loan origination amount divided by the average origination amount for all sample loans
with the same purpose and originated in the same MSA and year. The loan-level data was
merged with ZIP code level demographic information from the 2000 Census, monthly bank
prime interest rates from the Federal Reserve Bank of Saint Louis, and information on state APL
law provisions assembled by the author. Variables are defined in Table 2, with summary
statistics presented in Table 3.
To estimate the change in loan interest rate associated with a loan having a prepayment
penalty, I use an instrumental variables approach that addresses the endogeneity of loan interest
rates, LTV ratios, and prepayment penalties. This is the approach used in Rose (2011), and is
similar to those used by Elliehausen et al. (2008) and LaCour-Little and Holmes (2008). When
taking out a mortgage, borrowers frequently choose from a variety of combinations of interest
9
Population figures are from the July 1, 2007 estimates of the U.S. Census Bureau. The highest population MSA
from each decile included two California MSAs (Los Angeles and Riverside) and two MSAs covering parts of New
Jersey (New York City and Newark). In each case, the lower-population MSA (Riverside and Newark) was
replaced by the next most populous MSA in that decile (Miami and San Antonio, respectively).
12
rates and LTV ratios (with lower rates associated with lower LTV ratios), with a discrete
reduction in interest rates available for accepting a prepayment penalty. To address this
endogeneity in the choice of loan terms, I first use a probit model to estimate PP, and then use
the predicted values from that model in an equation-by-equation two stage least squares (2SLS)
model for estimating InitialRate and LTV.
10
A simultaneous equation 2SLS approach is more
efficient than an equation-by-equation one if all of the equations are specified correctly.
However, misspecification in one equation of a simultaneous equation system can cause
inconsistent coefficient estimates in the entire system, while in an equation-by-equation approach
this problem is confined to the equation in which the misspecification exists. The
LoanPerformance database contains little information on potentially relevant borrower
characteristics, raising a concern about misspecification and arguing for the more robust
equation-by-equation approach.
LTV and the predicted values of PP appear in the InitialRate equation, and InitialRate
appears in the LTV equation. Given the decision structure described above (selecting an interest
rate and LTV combination, then choosing whether or not to accept a prepayment penalty), LTV
and PP need not be determined simultaneously, and so PP does not appear in the LTV equation.
The other loan-level variables appear in all three equations. Each equation also includes
instruments specific to the dependent variable. The PP equation includes two variables,
%Refinance and %ShortTenure, designed to capture turnover in local home ownership, which
could affect expectations of how long a borrower will live in a particular house. High turnover
10
Elliehausen et al. (2008) and DeMong and Burroughs (2005) use APR, which captures the cost of initial points
and fees as their measure of loan prices. Information about points and fees is unavailable in LoanPerformance.
Those two studies generally find larger prepayment penalty discounts than have previous studies that use initial
interest rates as the measure of loan prices, including Rose (2011), Mayer et al. (2010), LaCour-Little and Holmes
(2008), and Ernst (2005). Another variable for which it would be useful to control in examinations of the pricing of
prepayment penalties is the amount of the required prepayment fee. To my knowledge no previous study has
incorporated data on prepayment fee amounts, and I also lack access to such data.
13
would lower borrowers’ preferences for prepayment penalties while raising the preferences of
lenders, leaving the expected net effect unclear. %Refinance indicates the percentage of
LoanPerformance subprime loans (FRMs and ARMs) originated in each ZIP code and year that
are refinances. %ShortTenure reflects the percentage of households in each ZIP code in which
the residents have lived in their houses for five years or less. APL_Dur indicates that an APL
provision restricting the duration of prepayment penalty periods is in effect at loan origination.
Such a provision can limit the protection against prepayment risk and reclassification risk that a
prepayment penalty offers a lender, and so make prepayment penalties less attractive to lenders
and more attractive to borrowers.
11
The instrument in the InitialRate equation is the bank prime
rate, which is mainly used to price business loans and proxies for the opportunity cost of
mortgage lending. Higher prime rates should be associated with higher loan interest rates. The
prime rate should not directly influence borrower’s choices regarding loan terms as it does not
generally change in response to changes in other market rates. The instruments in the LTV
equation are two sets of indicator variables describing the distribution of resident ages and house
values across ZIP codes. The premise is that older borrowers and borrowers buying higher-value
properties are on average wealthier, and that wealthier borrowers may prefer loans with lower
LTV ratios. All specifications include indicator variables for MSA and origination year.
Standard errors are clustered by the month of origination.
11
As a robustness check, I repeated the analyses reported in Tables 4 and 5 below after replacing APL_Dur with
APL_Amt, which indicates that an APL provision restricting the maximum amount that can be charged as a
prepayment penalty is in effect at loan origination. The changes in loan interest rates associated with a prepayment
penalty are generally larger in the results using APL_Amt than those reported below, but in most cases (those
involving College being the exception) show the same pattern of results with respect to the paper’s hypotheses.
Log-likelihood and pseudo-R
2
values are uniformly higher for the APL_Dur specifications than for the APL_Amt
ones. The two provisions are highly correlated, with correlation coefficient of 0.60, and including both APL_Dur
and APL_Amt in the same specification resulted in quite large but oppositely signed coefficient estimates for both.
All of these unreported results are available from the author.
14
4. Empirical Analyses
Panel A of Table 4 presents results from the probit and 2SLS specifications using the full
sample to estimate the loan interest rate discount associated with a prepayment penalty. The key
variable of interest is Pr(PP), the predicted values of PP derived from the probit model. The
change in loan interest rates associated with a loan having a prepayment penalty is calculated in
Panel B as the coefficient estimate for Pr(PP) in the 2SLS InitialRate equation multiplied by the
difference between the mean value of Pr(PP) for loans with prepayment penalties and the mean
value of Pr(PP) for loans without prepayment penalties. The results indicate that across the full
sample, a prepayment penalty is associated with a discount of 10.6 basis points (bp). This is
toward the middle of the range of previous papers’ findings for the change in loan rate associated
with a prepayment penalty, which runs from a decrease of 60 bp to an increase of 40-50 bp. The
following analyses split the sample along multiple lines to test the hypotheses presented above.
4.1 Testing Hypotheses 1 and 2
Table 5 presents results similar to those in Panel B of Table 4, showing the changes in
loan interest rates associated with prepayment penalties across sample splits designed to test
Hypotheses 1 and 2. (Complete results of the probit and 2SLS models on which Table 5’s
figures are based are in the Appendix of this paper.) According to Hypothesis 1, the prepayment
penalty discount should be greater for refinance loans that for purchase loans. The first two rows
of Table 5 support that hypothesis. Prepayment penalties are associated with a 16 bp decrease in
loan rates for refinances, but are associated with a 7 bp increase for purchase loans. Both the
efficiency and predatory views predict that the discount should be greater for refinances, but the
efficiency view does not offer an explanation for the increase in loan rates found for purchase
15
loans. The analyses below will provide evidence that in almost all subsets of loans, prepayment
penalties are associated with lower interest rates. Table 5’s results for purchase loans
empirically support the contention that while the effects described in the efficiency view may
dominate those described in the predatory view in most cases, the predatory view holds merit and
some amount of predatory lending utilizing prepayment penalties does occur, at least for
purchase loans.
The remaining rows of Table 5 split the sample based on FICO, College, and Income,
three variables intended to capture borrower creditworthiness and financial constraints as called
for in Hypothesis 2.
12
The results indicate that prepayment penalty discounts are higher for
borrowers with lower FICO scores and who reside in locales with lower levels of education and
household income. Assuming that these variables are reasonable proxies for borrower
creditworthiness and financial constraints, these results are consistent with the efficiency view
and contradict the predatory view.
4.2 Testing Hypothesis 3
Table 6 provides results based on analyses similar to those from Tables 4 and 5, but
performed separately for loans originated in each year of the sample period.
13
The top set of
results, based on the full sample, show that prepayment penalty discounts increased as credit
quality declined leading up to the subprime mortgage crisis, more than doubling from 8 bp for
2003 originations to 20 bp for 2006 originations. The results for the split samples indicate that
the changes in prepayment penalty discounts occurred primarily among refinance loans, loans to
12
It should be noted that College and Income are based on ZIP code level data and so are only proxies for borrower
education and income.
13
Results from the probit and 2SLS models on which Table 6’s figures are based are not presented for the sake of
brevity, but are available from the author.
16
borrowers with lower credit scores, and loans to borrowers residing in locales with lower
education levels and incomes. For purchase loans, the changes in loan rates associated with
prepayment penalties are positive and stable but the coefficient estimates for Pr(PP) are not
statistically significant. Prepayment penalty discounts hardly changed at all during the sample
period for borrowers with higher FICO scores, and only changed at the tail end of the sample
period for borrowers in better-educated, higher income ZIP codes. Prepayment penalty discounts
increased earlier and increased more for refinance loans and loans to less creditworthy, more
financially constrained borrowers. This is consistent with the efficiency view and contradicts the
predatory view.
4.3 Testing Hypotheses 4 and 5
Panel A of Table 7 shows the average duration of prepayment penalty periods,
conditional on a loan having a prepayment penalty. For the full sample, the average duration is
just under 35 months. The modal duration (not shown in the table) is 36 months, representing
seventy percent of all loans with a prepayment penalty. Eighteen percent of prepayment penalty
loans have shorter durations (mostly 12 months), and twelve percent have longer ones (almost all
60 months). Turning to the split samples, average durations are longer for refinances than for
purchases, which according the Hypothesis 4 is consistent with the efficiency view and
contradicts the predatory view. Durations are also longer for less creditworthy, more financially
constrained borrowers, which according to Hypothesis 5 is consistent with both views.
Panel B of Table 7 presents results from repeating the analyses of Tables 4 and 5 while
constraining the sample to include only loans that do not have prepayment penalties and loans
that have prepayment penalties of 36 months, thereby controlling for durations in estimating the
17
pricing of prepayment penalties.
14
The changes in InitialRate associated with prepayment
penalties are more extreme here than in Tables 4 and 5, with several exceeding 100 bp, but the
pattern of results is largely the same. Discounts are greater for refinance loans and for loans with
lower values of FICO, College, and Income. Prepayment penalties continue to be associated
with increases in loan rates for purchase loans, but now the same holds for loans to borrowers
with high FICO scores.
15
Clearly the artificial constraint of possible durations of prepayment
penalties imposed here raises selection bias concerns, and so the estimates in Panel B need to be
viewed with caution. Nonetheless, two points may perhaps be taken from them. First, the fact
that the pattern of results with regard to Hypotheses 1-3 matches those in the previous tables
provides some assurance that differences in durations do not drive the results in the previous
tables. Second, the dramatic changes in prepayment penalty discounts between previous tables
and Table 7 suggest that the results of the previous tables and of previous studies on the pricing
of prepayment penalties may mask substantial heterogeneity of prepayment penalty discounts
based on durations. Investigating that heterogeneity while appropriately addressing selection
issues is beyond the scope of this paper.
4.4 Testing Hypothesis 6
Table 8 provides estimates of the change in loan interest rates associated with
prepayment penalties with the sample split by APL_Dur and APL_Amt. These variables indicate
state APL provisions in effect at the time of a loan’s origination that place greater restrictions on
the duration and amount, respectively, of prepayment penalties than the federal Home Ownership
14
Results from the probit and 2SLS models on which the figures in Panel B of Table 7 are based are not presented
for the sake of brevity, but are available from the author.
15
This is consistent with Mayer et al. (2010), who find that prepayment penalties are negatively associated with loan
interest rates for loans to borrowers with FICO scores under 680, but positively associated with loan interest rates
for loans to borrowers with FICO scores over 680.
18
and Equity Protection Act (HOEPA). Because HOEPA applies nationwide, only state APL
provisions that are more restrictive than the corresponding provisions in HOEPA should affect
mortgage lending practices. HOEPA restricts the duration of a prepayment penalty period on a
covered mortgage to sixty months after origination, so APL_Dur takes a value of one when and
where a state’s APL law prohibits prepayment penalties on covered loans prior to sixty months
after origination, and zero otherwise. HOEPA does not restrict the amounts of prepayment
penalties, so APL_Amt takes a value of one when and where a state’s APL law places any
restriction on prepayment penalty amounts.
The results indicate that prepayment penalty discounts are several times greater for loans
originated with APL provisions restricting the use of prepayment penalties in effect. Based on
Hypothesis 6, this is consistent with the APL provisions reducing the use of prepayment
penalties as a predatory loan feature while preserving the usefulness of prepayment penalties as a
way for lenders to protect themselves against reclassification risk. Stated differently, in general
the effects described in the efficiency view appear to dominate the effects described in the
predatory view, but where APL provisions restrict (without prohibiting) the use of prepayment
penalties, the reduction in predatory lending via prepayment penalties makes the efficiency view
even more dominant.
16
5. Conclusions
The findings in the preceding section provide evidence that prepayment penalty discounts
are greater for refinance (versus purchase) loans, loans to borrowers with lower credit scores, and
16
It should be noted that there are endogeneity concerns regarding these findings, as a state’s previous mortgage or
housing market conditions could affect both a state’s APL laws and subsequent loan pricing characteristics. A
complete examination of the efficacy of state APL laws would need to address the determinants of those laws and
their specific provisions, but such an analysis is beyond the scope of this paper.
19
loans originated to residents of locales with lower incomes and education levels. These results
are consistent with the efficiency view, in which the riskiest borrowers receive the greatest
benefit from prepayment penalties. Although the efficiency view appears to dominate the
predatory view, particular results that cannot be explained by the efficiency view lend credence
to the predatory view. Prepayment penalties are associated with higher loan interest rates for
purchase loans, which could indicate predatory lenders taking advantage of borrowers with less
mortgage market experience (relative to refinance borrowers). APL provisions restricting the
use of prepayment penalties are associated with larger prepayment penalty discounts, suggesting
that such provisions reduce predatory lending via prepayment penalties. This implies both that
the predatory view has real merit, and that APL provisions that restrict, without prohibiting,
prepayment penalties can protect some borrowers from the harmful effects that the predatory
view predicts.
The APL provision results also suggest that the prepayment penalty restrictions on non-
high-priced FRMs under the Dodd-Frank Act may benefit consumers by curbing some predatory
lending practices and encouraging larger prepayment penalty discounts. However, the
prohibition of prepayment penalties on high-priced FRMs may reduce the welfare of those least
creditworthy, most financially constrained borrowers who could benefit the most from the
availability of prepayment penalties. The results also suggest that the prohibition of prepayment
penalties on ARMs may be detrimental to consumer welfare, but this conclusion must be
regarded as speculative as the analyses here include only FRMs.
Overall, these findings inform, but do not definitively answer, the question of whether
prepayment penalties are on the whole beneficial or harmful to consumers. A definitive answer
would require examinations of, among other things, probabilities of default and prepayment, the
20
benefits of access to credit, and the costs of default. While acknowledging that the net benefit of
prepayment penalties may be positive or negative, the results presented in this paper do strongly
suggest that the net benefit is greater for riskier borrowers. For those concerned with the impacts
of loan features particularly on more vulnerable groups of consumers, this conclusion remains
highly relevant.
Appendix
Tables A1-A4 provide the complete regression results on which Table 5 is based. Tables
A5-A6 do the same for Table 8. The regression results underlying Table 6 and Panel B of Table
7 are not included here for the sake of brevity, but are available from the author.
21
References
Avery, R.B., G.B. Canner, and R.E. Cook. 2005. New Information Reported under HMDA and
Its Application in Fair Lending Enforcement. Federal Reserve Bulletin 91 (Summer):344-394.
Berndt, A., B. Hollifield, and P. Sandas. 2010. The Role of Mortgage Brokers in the Subprime
Crisis. NBER Working Paper 16175.
Danis, M.A., and A. Pennington-Cross. 2008. The Delinquency of Subprime Mortgages.
Journal of Economics and Business 60 (1-2):67-90.
DeMong, R.F. and J.E. Burroughs. 2005. Prepayment Fees Lead to Lower Interest Rates.
Equity Magazine November/December 2005:19-21.
Demyanyk, Y., and O. Van Hemert. 2011. Understanding the Subprime Mortgage Crisis. The
Review of Financial Studies 24 (6):1848-1880.
Elliehausen, G., M.E. Staten, and J. Steinbuks. 2008. The Effect of Prepayment Penalties on the
Pricing of Subprime Mortgages. Journal of Economics and Business 60 (1-2):33-46.
Ernst, K.S. 2005. Borrowers Gain No Interest Rate Benefits from Prepayment Penalties on
Subprime Mortgages. Center for Responsible Lending research report.
Ernst, K.S., D. Bocian, and W. Li. 2008. Steered Wrong: Brokers, Borrowers, and Subprime
Loans. Center for Responsible Lending research report.
Hendel, I.E., and A. Lizzeri. 2003. The Role of Commitment in Dynamic Contracts: Evidence
from Life Insurance. Quarterly Journal of Economics 118 (1):299-327.
Jackson, H.E., and L. Burlingame. 2007. Kickbacks or Compensation: The Case of Yield
Spread Premiums. Stanford Journal of Law, Business, and Finance 12 (2):289-361.
LaCour-Little, M., and C. Holmes. 2008. Prepayment Penalties in Residential Mortgage
Contracts: A Cost-Benefit Analysis. Housing Policy Debate 19 (4):1-43.
Mayer, C.J., and K. Pence. 2009. Subprime Mortgages: What, Where, and To Whom? E.
Glaeser and J. Quigley, editors. Housing Markets and the Economy: Risk, Regulations, and
Policy: Essays in Honor of Karl E. Case Lincoln Land Institute: Cambridge, MA.
Mayer, C.J., T. Piskorski, and A. Tchistyi. 2010. The Inefficiency of Refinancing: Why
Prepayment Penalties Are Good for Risky Borrowers. NBER Working Paper 16586.
McCoy, P.A. 2007. Rethinking Disclosure in a World of Risk-based Pricing. Harvard Journal
on Legislation 44:123-166.
22
23
Pennington-Cross, A., and G. Ho. 2010. The Termination of Subprime Hybrid and Fixed Rate
Mortgages. Real Estate Economics 38 (3):399-426.
Quercia, R.G., M.A. Stegman, and W.R. Davis. 2007. The Impact of Predatory Loan Terms on
Subprime Foreclosures: The Special Case of Prepayment Penalties and Balloon Payments.
Housing Policy Debate 18 (2):311-346.
RealtyTrac Inc. 2008. Detroit, Stockton, Las Vegas Post Highest 2007 Metro Foreclosure
Rates. Press release, February 13, 2008.
Rose, M.J. 2011. Origination Channel, Prepayment Penalties, and Default. Real Estate
Economics, forthcoming.
Rose, M.J. 2008. Predatory Lending Practices and Subprime Foreclosures: Distinguishing
Impacts by Loan Category. Journal of Economics and Business 60 (1-2):13-32.
Woodward, S.E. 2008. A Study of Closing Costs for FHA Mortgages. U.S. Department of
Housing and Urban Development, Office of Policy Development and Research.
Woodward, S.E., and R.E. Hall. 2010. Consumer Confusion in the Mortgage Market: Evidence
of Less than a Perfectly Transparent and Competitive Market. American Economic Review 100
(2):511-515.
Table 1: Sample metropolitan statistical areas
MSA foreclosure rates at year-end 2007 are from RealtyTrac, Inc. (2008), which defines the
foreclosure rate as the percentage of total households entering some stage of foreclosure during
the year 2007. Population estimates as of July 1, 2007, are from the US Census Bureau.
Foreclosure Sample loans Population
MSA State(s) rate Number Percent Number Percent
Miami FL 2.7 20,030 10.1% 2,382,961 4.6%
Atlanta GA 2.5 17,848 9.0% 5,261,296 10.2%
Phoenix AZ 1.9 15,190 7.6% 4,165,921 8.1%
Chicago IL 1.6 23,483 11.8% 7,929,840 15.4%
Los Angeles CA 1.4 55,686 28.0% 9,807,870 19.1%
San Antonio TX 1.1 10,530 5.3% 1,984,921 3.9%
Minneapolis MN-WI 0.8 9,315 4.7% 3,197,620 6.2%
Baltimore MD 0.7 13,602 6.8% 2,663,805 5.2%
New York NJ-NY 0.5 22,680 11.4% 11,627,931 22.6%
Pittsburgh PA 0.4 10,494 5.3% 2,354,159 4.6%
Total 198,858 51,376,324
24
Table 2: Variable definitions
Bank prime loan rates are from the Federal Reserve Bank of Saint Louis. Resident education, income, age, tenure and house value
data are from the 2000 Census. Information on state anti-predatory lending laws is from the author’s analysis of the relevant state
legislation and regulations. All other variables are from the LoanPerformance dataset from CoreLogic.
PP
Equals 1 if the loan features a prepayment penalty; 0 otherwise
InitialRate
Initial loan interest rate at origination
LTV
Loan-to-value ratio at origination
FICO
Borrower’s FICO score at origination
LowNoDoc
Equals 1 if the loan is based on reduced documentation; 0 otherwise
RelLoanSize
Ratio of loan origination amount to the average origination amount of all sample loans of the same purpose (purchase
or refinance) originated in the same MSA and year
OwnerOcc
Equals 1 if the loan is for an owner-occupied property; 0 otherwise
RefiCash
Equals 1 if the loan is a cashout refinance; 0 otherwise
RefiNoCash
Equals 1 if the loan is a non-cashout refinance; 0 otherwise
30Year
Equals 1 if the loan is a 30-year loan; 0 if it is a 15-year loan
%Refinance
% of LoanPerformance subprime loans by ZIP code and origination year that are refinances
%ShortTenure
% of owner-occupied households in the borrower’s ZIP code in which the residents have lived in their current houses
for five years or less
APL_Dur
Equals 1 if a state’s APL law’s prohibition against prepayment penalties on covered loans takes effect sooner than
five years after loan origination, 0 otherwise
APL_Amt
Equals 1 if a state’s APL law restricts the maximum amount that can be charged as a prepayment penalty on a
covered loan, 0 otherwise
PrimeRate
Monthly bank prime loan rate at origination
%Age18-34
% of residents in the borrower’s ZIP code between the ages of 18 and 34
%Age35-44
% of residents in the borrower’s ZIP code between the ages of 35 and 44
%Age45-59
% of residents in the borrower’s ZIP code between the ages of 45 and 59
%Age60+
% of residents in the borrower’s ZIP code 60 years old or older
%Value$1-$2
% of specified owner-occupied housing units in the borrower’s ZIP code valued between $100,000 and $200,000
%Value$2-$3
% of specified owner-occupied housing units in the borrower’s ZIP code valued between $200,000 and $300,000
%Value$3-$5
% of specified owner-occupied housing units in the borrower’s ZIP code valued between $300,000 and $500,000
%Value$5+
% of specified owner-occupied housing units in the borrower’s ZIP code valued above $500,000
College
Percentage of residents 25 years old or older with at least a Bachelor’s degree in borrower’s ZIP code
Income
Median household income (in thousands) in borrower’s ZIP code
25
Table 3: Summary statistics
Variable Mean Median St. Dev.
PP
0.69 1.00 0.46
InitialRate
7.33 7.10 1.23
LTV
74.64 79.89 16.01
FICO
630.33 628.00 62.56
LowNoDoc
0.31 0.00 0.46
RelLoanSize
0.99 0.88 0.51
OwnerOcc
0.94 1.00 0.24
RefiCash
0.75 1.00 0.43
RefiNoCash
0.11 0.00 0.31
30Year
0.92 1.00 0.26
%Refinance
0.70 0.71 0.11
%ShortTenure
0.34 0.32 0.11
APL_Dur
0.82 1.00 0.38
APL_Amt
0.62 1.00 0.48
PrimeRate
5.34 4.43 1.53
%Age18-34
24.10 23.88 4.79
%Age35-44
16.18 15.93 2.16
%Age45-59
17.24 17.27 3.10
%Age60+
14.21 13.71 5.58
%Value$1-$2
47.09 48.14 24.50
%Value$2-$3
14.01 7.94 14.48
%Value$3-$5
5.96 1.52 10.05
%Value$5+
2.18 0.39 6.79
College
20.85 18.40 12.75
Income
46.13 43.75 15.62
26
Table 4: Prepayment penalties and loan rates – full sample
Panel A presents the results of probit and 2SLS regressions using loan-level data for subprime
fixed-rate mortgages originated during 2003-2006. Pr(PP) is defined as the predicted values
from the probit model. Other variables are defined in Table 2. In Panel B, the change in loan
interest rate associated with a loan having a prepayment penalty is calculated as the coefficient
estimate for Pr(PP) multiplied by the difference between the mean value of Pr(PP) for loans
with prepayment penalties and the mean value of Pr(PP) for loans without prepayment penalties.
Vintage year indicators, MSA indicators, and a constant term are included in all specifications.
Standard errors are clustered by month of origination. Levels of significance are indicated by *,
**, and *** for 10%, 5%, and 1%, respectively.
Panel A: Determinants of PP, InitialRate, and LTV
Model: Probit 2SLS 2SLS
Dependent variable:
PP InitialRate LTV
Pr(PP)
-0.242***
[0.0291]
LTV
0.0122***
[0.000372]
InitialRate
-7.034***
[0.399]
FICO
-7.13E-05 -0.00851*** -0.0411***
[0.000148] [3.39e-05] [0.00338]
LowNoDoc
-0.111*** 0.379*** 1.065***
[0.0138] [0.00442] [0.167]
RelLoanSize
0.0642*** -0.378*** 12.10***
[0.0172] [0.00477] [0.111]
OwnerOcc
-0.0803*** -0.504*** -3.799***
[0.0260] [0.00832] [0.240]
RefiCash
0.0147 -0.244*** -11.71***
[0.0252] [0.00695] [0.177]
RefiNoCash
0.0248 -0.342*** -11.44***
[0.0168] [0.00870] [0.227]
30Year
0.241*** 0.154*** 9.976***
[0.0152] [0.00846] [0.165]
%Refinance
0.234***
[0.0668]
%ShortTenure
0.115*
[0.0615]
APL_Dur
1.169***
[0.123]
PrimeRate
0.250***
[0.00531]
%Age18-34
-0.153***
[0.0121]
%Age35-44
0.245***
[0.0247]
%Age45-59
-0.153***
27
28
[0.0211]
%Age60+
-0.0295***
[0.0102]
%Value$1-$2
-0.155***
[0.00266]
%Value$2-$3
-0.276***
[0.00466]
%Value$3-$5
-0.363***
[0.00668]
%Value$5+
-0.564***
[0.00728]
Observations 194,194 198,858 198,858
R
2
0.377 0.500 0.097
Panel B: Change in InitialRate associated with a prepayment penalty
Means of Pr(PP)
Estimate for Pr(PP) PP = 1 PP = 0 Change in InitialRate
-0.242*** 0.829 0.390 -0.106
Table 5: Prepayment penalties and loan rates – sample splits
This table presents results based on probit and 2SLS regressions using loan-level data for
subprime fixed-rate mortgages originated during 2003-2006 with the sample split based on loan
purpose and the sample medians of FICO, College, and Income. Complete regression results
appear in Tables A1-A4 of the appendix to this paper. Levels of significance for coefficient
estimates of Pr(PP) in the 2SLS regressions are indicated by *, **, and *** for 10%, 5%, and
1%, respectively.
Means of Pr(PP)
Estimate for Pr(PP) PP = 1 PP = 0 Change in InitialRate
Refinances -0.338*** 0.833 0.376 -0.155
Purchases 0.205** 0.789 0.429 0.074
FICO < median -0.487*** 0.860 0.417 -0.216
FICO >= median -0.134*** 0.840 0.448 -0.053
College < median -0.339*** 0.840 0.402 -0.148
College >= median -0.219*** 0.824 0.398 -0.094
Income < median -0.570*** 0.835 0.391 -0.253
Income >= median -0.188*** 0.830 0.407 -0.080
29
Table 6: Prepayment penalties and loan rates by origination year
This table presents results from analyses similar to those in Tables 4 and 5 run separately for
loans in each origination year. Complete regression results are available from the author. Levels
of significance for coefficient estimates of Pr(PP) in the 2SLS regressions are indicated by *, **,
and *** for 10%, 5%, and 1%, respectively.
Full sample
Estimate Means of Pr(PP) Change in
for Pr(PP) PP = 1 PP = 0
InitialRate
2003 -0.188*** 0.838 0.391 -0.084
2004 -0.192*** 0.852 0.400 -0.087
2005 -0.246*** 0.841 0.363 -0.118
2006 -0.396*** 0.802 0.300 -0.199
Refinances Purchases
Estimate Means of Pr(PP) Change in Estimate Means of Pr(PP) Change in
for Pr(PP) PP = 1 PP = 0
InitialRate
for Pr(PP) PP = 1 PP = 0
InitialRate
2003 -0.253*** 0.845 0.379 -0.118 0.025 0.765 0.399 0.009
2004 -0.243*** 0.857 0.393 -0.112 0.022 0.830 0.436 0.009
2005 -0.286*** 0.841 0.348 -0.141 0.093 0.810 0.398 0.038
2006 -0.425*** 0.803 0.289 -0.218 0.018 0.774 0.330 0.008
FICO below median FICO at or above median
Estimate Means of Pr(PP) Change in Estimate Means of Pr(PP) Change in
for Pr(PP) PP = 1 PP = 0
InitialRate
for Pr(PP) PP = 1 PP = 0
InitialRate
2003 -0.295*** 0.880 0.424 -0.135 -0.194*** 0.842 0.433 -0.079
2004 -0.335*** 0.874 0.427 -0.150 -0.089** 0.863 0.452 -0.036
2005 -0.310*** 0.857 0.389 -0.145 -0.154*** 0.853 0.406 -0.069
2006 -0.500*** 0.832 0.314 -0.259 -0.130* 0.816 0.376 -0.057
College below median College at or above median
Estimate Means of Pr(PP) Change in Estimate Means of Pr(PP) Change in
for Pr(PP) PP = 1 PP = 0
InitialRate
for Pr(PP) PP = 1 PP = 0
InitialRate
2003 -0.149* 0.856 0.407 -0.067 -0.205*** 0.827 0.392 -0.089
2004 -0.268*** 0.858 0.401 -0.122 -0.156*** 0.849 0.411 -0.068
2005 -0.350*** 0.844 0.356 -0.171 -0.195*** 0.844 0.382 -0.090
2006 -0.475*** 0.807 0.302 -0.240 -0.365*** 0.800 0.304 -0.181
Income below median Income at or above median
Estimate Means of Pr(PP) Change in Estimate Means of Pr(PP) Change in
for Pr(PP) PP = 1 PP = 0
InitialRate
for Pr(PP) PP = 1 PP = 0
InitialRate
2003 -0.198*** 0.850 0.398 -0.090 -0.197*** 0.835 0.401 -0.086
2004 -0.392*** 0.854 0.395 -0.180 -0.139*** 0.855 0.418 -0.061
2005 -0.418*** 0.841 0.346 -0.207 -0.180*** 0.846 0.387 -0.083
2006 -0.454*** 0.800 0.289 -0.232 -0.380*** 0.814 0.321 -0.187
30
Table 7: Prepayment penalty durations
Panel A presents the average number of months after origination that a prepayment penalty is in effect for those loans that have
prepayment penalties. T-statistics from difference in means tests all indicate significance at the 1% level. Panel B presents results
from analyses similar to those in Tables 4 and 5 including only loans with prepayment penalty durations of 36 months and loans
without prepayment penalties. Complete regression results are available from the author. Levels of significance for coefficient
estimates of Pr(PP) in the 2SLS regressions are indicated by *, **, and *** for 10%, 5%, and 1%, respectively.
Panel A
Panel B
Mean Means of Pr(PP)
duration T-statistic Estimate for Pr(PP) PP = 1 PP = 0 Change in InitialRate
All loans 34.98 -1.761*** 0.821 0.280 -0.952
Refinances 35.15 12.79 -2.076*** 0.826 0.269 -1.158
Purchases 33.88 1.555*** 0.776 0.322 0.707
FICO < median 35.56 17.38 -2.042*** 0.857 0.309 -1.119
FICO >= median 34.41 0.744*** 0.830 0.331 0.370
College < median 36.21 36.95 -1.277*** 0.830 0.283 -0.700
College >= median 33.76 -0.957*** 0.813 0.288 -0.503
Income < median 36.48 44.88 -1.867*** 0.825 0.272 -1.032
Income >= median 33.50 -0.259 0.823 0.298 -0.136
31
Table 8: Prepayment penalties and loan rates – samples split by APL provisions
This table presents results based on probit and 2SLS regressions using loan-level data for
subprime fixed-rate mortgages originated during 2003-2006 with the sample split based on
APL_Dur and APL_Amt. Complete regression results appear in Tables A5 and A6 of the
appendix to this paper. Levels of significance for coefficient estimates of Pr(PP) in the 2SLS
regressions are indicated by *, **, and *** for 10%, 5%, and 1%, respectively..
Means of Pr(PP)
Estimate for Pr(PP) PP = 1 PP = 0 Change in InitialRate
APL_Dur = 1 -2.016*** 0.848 0.421 -0.861
APL_Dur = 0 -0.886*** 0.720 0.576 -0.127
APL_Amt = 1 -2.512*** 0.857 0.427 -1.080
APL_Amt = 0 -0.621*** 0.824 0.638 -0.115
32
Tab
le A1: Prepayment penalties and loan rates – sample split by loan purpose
This table presents the results of probit and 2SLS regressions using loan-level data for subprime fixed-rate
mortgages originated during 2003-2006 with the sample split between refinance and purchase loans. Specifications
are identical to those in Table 4 except RefiCash is omitted for refinances and both RefiCash and RefiNoCash are
omitted for purchases. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%, respectively.
Model: Probit 2SLS 2SLS
Dependent variable: PP InitialRate LTV
Refinances Purchases Refinances Purchases Refinances Purchases
Pr(PP) -0.338*** 0.205**
[0.0296] [0.104]
LTV 0.0103*** 0.0371***
[0.000366] [0.00200]
InitialRate -7.619*** -2.923***
[0.443] [0.683]
FICO 1.44E-05 -0.000768*** -0.00866*** -0.00786*** -0.0491*** 0.000325
[0.000169] [0.000208] [3.58e-05] [0.000106] [0.00384] [0.00504]
LowNoDoc -0.112*** -0.151*** 0.370*** 0.457*** 1.271*** -0.580*
[0.0136] [0.0283] [0.00472] [0.0128] [0.183] [0.308]
RelLoanSize 0.0745*** 0.0814*** -0.356*** -0.435*** 13.55*** 3.804***
[0.0182] [0.0279] [0.00518] [0.0115] [0.120] [0.231]
OwnerOcc -0.243*** 0.152*** -0.510*** -0.471*** -3.422*** -3.340***
[0.0324] [0.0420] [0.00970] [0.0187] [0.270] [0.407]
RefiCash -0.0353 0.0980*** -0.0719
[0.0238] [0.00635] [0.122]
30Year 0.201*** 0.331*** 0.165*** 0.0953* 9.831*** 10.81***
[0.0177] [0.0593] [0.00838] [0.0494] [0.174] [0.651]
%Refinance 0.304*** 0.264**
[0.0752] [0.121]
%ShortTenure 0.142** -0.0595
[0.0654] [0.106]
APL_Dur 1.237*** 0.869***
[0.126] [0.113]
PrimeRate 0.240*** 0.327***
[0.00558] [0.0162]
%Age18-34 -0.171*** -0.0288
[0.0133] [0.0249]
%Age35-44 0.328*** -0.247***
[0.0272] [0.0541]
%Age45-59
-0.168*** -0.0129
[0.0232] [0.0435]
%Age60+ -0.0315*** -0.0334
[0.0112] [0.0211]
%Value$1-$2 -0.166*** -0.0701***
[0.00287] [0.00542]
%Value$2-$3 -0.297*** -0.139***
[0.00507] [0.0102]
%Value$3-$5 -0.388*** -0.185***
[0.00720] [0.0152]
%Value$5+ -0.598*** -0.323***
[0.00790] [0.0162]
Observations 166,928 27,266 171,177 27,681 171,177 27,681
R
2
0.416 0.289 0.505 0.471 0.063 -0.012
33
Tab
le A2: Prepayment penalties and loan rates – sample split by borrower FICO score
This table presents the results of probit and 2SLS regressions using loan-level data for subprime fixed-rate
mortgages originated during 2003-2006 with the sample split by FICO. Specifications are identical to those in
Table 4. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%, respectively.
Model: Probit 2SLS 2SLS
Dependent variable: PP InitialRate LTV
FICO
below
median
FICO
at or above
median
FICO
below
median
FICO
at or above
median
FICO
below
median
FICO
at or above
median
Pr(PP) -0.487*** -0.134***
[0.0443] [0.0329]
LTV 0.0135*** 0.0155***
[0.000590] [0.000454]
InitialRate -5.597*** -6.692***
[0.473] [0.573]
FICO 0.00263*** -0.00147*** -0.0127*** -0.00507*** -0.0180*** -0.0602***
[0.000187] [0.000173] [9.95e-05] [6.22e-05] [0.00597] [0.00335]
LowNoDoc -0.0917*** -0.141*** 0.449*** 0.383*** 0.266 0.871***
[0.0121] [0.0192] [0.00732] [0.00514] [0.230] [0.234]
RelLoanSize 0.174*** -0.0128 -0.498*** -0.295*** 14.24*** 10.91***
[0.0165] [0.0190] [0.00881] [0.00508] [0.179] [0.118]
OwnerOcc -0.276*** 0.0187 -0.559*** -0.523*** -2.023*** -4.417***
[0.0388] [0.0281] [0.0161] [0.00877] [0.339] [0.343]
RefiCash -0.207*** 0.112*** -0.269*** -0.195*** -10.48*** -12.01***
[0.0361] [0.0205] [0.0124] [0.00794] [0.239]
[0.238]
RefiNoCash -0.0808** 0.0453** -0.363*** -0.295*** -9.007*** -12.61***
[0.0316] [0.0205] [0.0148] [0.0101] [0.299] [0.313]
30Year 0.112*** 0.338*** 0.126*** 0.158*** 8.587*** 10.75***
[0.0210] [0.0184] [0.0126] [0.0106] [0.205] [0.250]
%Refinance -0.0638 0.445***
[0.0699] [0.0877]
%ShortTenure 0.148* -0.0591
[0.0796] [0.0682]
APL_Dur 1.189*** 1.204***
[0.137] [0.113]
PrimeRate 0.263*** 0.257***
[0.00792] [0.00667]
%Age18-34 -0.187*** -0.118***
[0.0170] [0.0163]
%Age35-44
0.247*** 0.238***
[0.0341] [0.0336]
%Age45-59 -0.194*** -0.0799***
[0.0290] [0.0288]
%Age60+ -0.0471*** -0.0218
[0.0139] [0.0138]
%Value$1-$2 -0.159*** -0.144***
[0.00327] [0.00369]
%Value$2-$3 -0.281*** -0.266***
[0.00638] [0.00652]
%Value$3-$5 -0.377*** -0.348***
[0.0101] [0.00880]
%Value$5+ -0.505*** -0.570***
[0.0121] [0.00941]
Observations 96,153 98,041 98,575 100,283 98,575 100,283
R
2
0.429 0.342 0.457 0.441 0.170 0.127
34
Tab
le A3: Prepayment penalties and loan rates – sample split by education
This table presents the results of probit and 2SLS regressions using loan-level data for subprime fixed-rate
mortgages originated during 2003-2006 with the sample split by College. Specifications are identical to those in
Table 4. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%, respectively.
Model: Probit 2SLS 2SLS
Dependent variable: PP InitialRate LTV
College
below
median
College
at or above
median
College
below
median
College
at or above
median
College
below
median
College
at or above
median
Pr(PP) -0.339*** -0.219***
[0.0837] [0.0285]
LTV 0.0157*** 0.0102***
[0.000657] [0.000533]
InitialRate -8.104*** -4.445***
[0.553] [0.513]
FICO 1.42E-05 -0.000175 -0.00861*** -0.00838*** -0.0476*** -0.0233***
[0.000168] [0.000161] [5.01e-05] [4.62e-05] [0.00469] [0.00433]
LowNoDoc -0.0968*** -0.142*** 0.394*** 0.375*** 1.502*** 0.0725
[0.0183] [0.0156] [0.00638] [0.00626] [0.234] [0.216]
RelLoanSize 0.345*** 0.0189 -0.655*** -0.278*** 20.47*** 10.11***
[0.0240] [0.0151] [0.0158] [0.00595] [0.230] [0.117]
OwnerOcc -0.0576** -0.143*** -0.457*** -0.515*** -4.215*** -3.166***
[0.0270] [0.0380] [0.0110] [0.0128] [0.313] [0.332]
RefiCash 0.025 -0.00467 -0.272*** -0.201*** -13.18*** -10.14***
[0.0269] [0.0289] [0.0108] [0.00960] [0.279]
[0.204]
RefiNoCash 0.0815*** -0.0214 -0.372*** -0.298*** -12.59*** -9.939***
[0.0273] [0.0202] [0.0132] [0.0118] [0.348] [0.269]
30Year 0.163*** 0.288*** 0.169*** 0.162*** 8.850*** 10.01***
[0.0180] [0.0208] [0.0120] [0.0124] [0.232] [0.223]
%Refinance 0.260*** 0.155*
[0.0763] [0.0909]
%ShortTenure -0.384*** 0.127*
[0.0817] [0.0657]
APL_Dur 0.623*** 1.430***
[0.104] [0.124]
PrimeRate 0.261*** 0.254***
[0.00754] [0.00753]
%Age18-34 -0.420*** -0.234***
[0.0258] [0.0159]
%Age35-44
-0.0111 -0.0538
[0.0405] [0.0369]
%Age45-59 -0.736*** -0.198***
[0.0424] [0.0267]
%Age60+ -0.0476*** -0.183***
[0.0179] [0.0150]
%Value$1-$2 -0.187*** -0.135***
[0.00350] [0.00483]
%Value$2-$3 -0.305*** -0.286***
[0.0102] [0.00633]
%Value$3-$5 -0.520*** -0.326***
[0.0328] [0.00799]
%Value$5+ -0.464*** -0.526***
[0.0816] [0.00801]
Observations 97,029 97,165 99,462 99,396 99,462 99,396
R
2
0.428 0.337 0.509 0.496 0.072 0.225
35
Tab
le A4: Prepayment penalties and loan rates – sample split by household income
This table presents the results of probit and 2SLS regressions using loan-level data for subprime fixed-rate
mortgages originated during 2003-2006 with the sample split by Income. Specifications are identical to those in
Table 4. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%, respectively.
Model: Probit 2SLS 2SLS
Dependent variable: PP InitialRate LTV
Income
below
median
Income
at or above
median
Income
below
median
Income
at or above
median
Income
below
median
Income
at or above
median
Pr(PP) -0.570*** -0.188***
[0.0737] [0.0300]
LTV 0.0126*** 0.00895***
[0.000612] [0.000501]
InitialRate -8.355*** -4.524***
[0.574] [0.511]
FICO -6.75E-05 -0.000131 -0.00876*** -0.00819*** -0.0506*** -0.0231***
[0.000159] [0.000161] [5.04e-05] [4.57e-05] [0.00498] [0.00423]
LowNoDoc -0.0756*** -0.156*** 0.384*** 0.372*** 1.611*** 0.131
[0.0147] [0.0176] [0.00636] [0.00618] [0.242] [0.215]
RelLoanSize 0.247*** -0.0102 -0.513*** -0.254*** 15.38*** 11.09***
[0.0240] [0.0145] [0.0112] [0.00595] [0.225] [0.115]
OwnerOcc -0.0763*** -0.131*** -0.479*** -0.492*** -4.422*** -3.075***
[0.0267] [0.0450] [0.0107] [0.0135] [0.327] [0.335]
RefiCash -0.0352 0.0484* -0.322*** -0.195*** -13.58*** -9.692***
[0.0301] [0.0258] [0.0106] [0.00938] [0.296]
[0.196]
RefiNoCash 0.0722** -0.0128 -0.403*** -0.296*** -13.28*** -9.402***
[0.0283] [0.0202] [0.0132] [0.0115] [0.368] [0.261]
30Year 0.176*** 0.290*** 0.192*** 0.164*** 9.163*** 10.45***
[0.0170] [0.0220] [0.0118] [0.0123] [0.238] [0.223]
%Refinance 0.394*** 0.163
[0.0773] [0.103]
%ShortTenure -0.248** 0.0279
[0.0969] [0.0724]
APL_Dur 0.805*** 1.308***
[0.126] [0.119]
PrimeRate 0.248*** 0.251***
[0.00759] [0.00741]
%Age18-34 -0.257*** -0.0762***
[0.0202] [0.0167]
%Age35-44
-0.345*** 0.206***
[0.0466] [0.0357]
%Age45-59 -0.326*** -0.0107
[0.0395] [0.0270]
%Age60+ -0.0687*** -0.0870***
[0.0157] [0.0158]
%Value$1-$2 -0.180*** -0.167***
[0.00387] [0.00487]
%Value$2-$3 -0.294*** -0.282***
[0.00925] [0.00613]
%Value$3-$5 -0.462*** -0.350***
[0.0181] [0.00823]
%Value$5+ -0.452*** -0.571***
[0.0245] [0.00830]
Observations 96,920 97,274 99,289 99,569 99,289 99,569
R
2
0.410 0.349 0.511 0.486 0.024 0.229
36
Tab
le A5: Prepayment penalties and loan rates – sample split by APL_Dur
This table presents the results of probit and 2SLS regressions using loan-level data for subprime fixed-rate
mortgages originated during 2003-2006 with the sample split by APL_Dur. Specifications are identical to those in
Table 4 except APL_Dur is omitted. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%,
respectively.
Model: Probit 2SLS 2SLS
Dependent variable: PP InitialRate LTV
APL_Dur
equals 1
APL_Dur
equals 0
APL_Dur
equals 1
APL_Dur
equals 0
APL_Dur
equals 1
APL_Dur
equals 0
Pr(PP) -2.016*** -0.886***
[0.135] [0.208]
LTV 0.0123*** 0.0126***
[0.000382] [0.00124]
InitialRate -7.871*** -6.061***
[0.457] [0.819]
FICO -7.13E-05 -0.000395* -0.00836*** -0.00935*** -0.0481*** -0.0305***
[0.000171] [0.000235] [3.69e-05] [8.91e-05] [0.00381] [0.00752]
LowNoDoc -0.112*** -0.0647** 0.339*** 0.387*** 1.455*** 0.421
[0.0124] [0.0328] [0.00551] [0.0119] [0.189] [0.360]
RelLoanSize 0.0235 0.264*** -0.366*** -0.360*** 13.38*** 7.041***
[0.0174] [0.0485] [0.00532] [0.0135] [0.120] [0.286]
OwnerOcc -0.220*** 0.554*** -0.584*** -0.468*** -4.115*** -3.110***
[0.0301] [0.0343] [0.0114] [0.0341] [0.264] [0.574]
RefiCash -0.0516* 0.346*** -0.251*** -0.262*** -12.18*** -10.61***
[0.0302] [0.0341] [0.00783] [0.0233] [0.196]
[0.413]
RefiNoCash 0.00337 0.155*** -0.329*** -0.363*** -12.12*** -9.752***
[0.0231] [0.0456] [0.00968] [0.0216] [0.257] [0.490]
30Year 0.273*** 0.114*** 0.248*** 0.216*** 10.09*** 9.839***
[0.0189] [0.0383] [0.0125] [0.0197] [0.187] [0.365]
%Refinance -0.0261 0.202**
[0.0702] [0.0996]
%ShortTenure 0.238*** -0.637***
[0.0615] [0.159]
PrimeRate 0.246*** 0.279***
[0.00574] [0.0138]
%Age18-34 -0.205*** -0.0312
[0.0145] [0.0258]
%Age35-44 0.153*** 0.259***
[0.0279] [0.0637]
%Age45-59
-0.211*** 0.0236
[0.0258] [0.0433]
%Age60+ -0.118*** 0.0833***
[0.0124] [0.0217]
%Value$1-$2 -0.172*** -0.0995***
[0.00318] [0.00566]
%Value$2-$3 -0.284*** -0.228***
[0.00513] [0.0129]
%Value$3-$5 -0.380*** -0.320***
[0.00728] [0.0208]
%Value$5+ -0.583*** -0.445***
[0.00788] [0.0237]
Observations 159,116 35,078 163,070 35,788 163,070 35,788
R
2
0.393 0.459 0.509 0.445 0.079 -0.008
37
Tab
le A6: Prepayment penalties and loan rates – sample split by APL_Amt
This table presents the results of probit and 2SLS regressions using loan-level data for subprime fixed-rate
mortgages originated during 2003-2006 with the sample split by APL_Amt. Specifications are identical to those in
Table 4 except APL_Dur is omitted. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%,
respectively.
Model: Probit 2SLS 2SLS
Dependent variable: PP InitialRate LTV
APL_Amt
equals 1
APL_Amt
equals 0
APL_Amt
equals 1
APL_Amt
equals 0
APL_Amt
equals 1
APL_Amt
equals 0
Pr(PP) -2.512*** -0.621***
[0.127] [0.129]
LTV 0.0122*** 0.0117***
[0.000457] [0.000642]
InitialRate -6.056*** -9.695***
[0.454] [0.749]
FICO -1.77E-06 -0.000441*** -0.00831*** -0.00884*** -0.0321*** -0.0647***
[0.000178] [0.000171] [4.24e-05] [5.80e-05] [0.00378] [0.00654]
LowNoDoc -0.105*** -0.0977*** 0.319*** 0.389*** 0.624*** 2.204***
[0.0137] [0.0218] [0.00614] [0.00766] [0.188] [0.321]
RelLoanSize 0.0397** 0.154*** -0.349*** -0.369*** 12.31*** 11.48***
[0.0177] [0.0282] [0.00598] [0.00870] [0.129] [0.212]
OwnerOcc -0.276*** 0.419*** -0.658*** -0.502*** -3.130*** -5.998***
[0.0314] [0.0234] [0.0136] [0.0196] [0.262] [0.508]
RefiCash -0.147*** 0.302*** -0.284*** -0.296*** -10.23*** -14.62***
[0.0328] [0.0189] [0.0102] [0.0151] [0.184]
[0.387]
RefiNoCash -0.0449* 0.120*** -0.333*** -0.394*** -10.10*** -14.15***
[0.0260] [0.0240] [0.0111] [0.0152] [0.247] [0.463]
30Year 0.271*** 0.198*** 0.284*** 0.211*** 9.396*** 11.26***
[0.0198] [0.0233] [0.0136] [0.0143] [0.193] [0.317]
%Refinance -0.0124 0.561***
[0.0777] [0.0748]
%ShortTenure 0.313*** -0.556***
[0.0683] [0.113]
PrimeRate 0.253*** 0.246***
[0.00658] [0.00897]
%Age18-34 -0.179*** -0.116***
[0.0153] [0.0229]
%Age35-44 0.234*** 0.249***
[0.0297] [0.0520]
%Age45-59
-0.176*** -0.104***
[0.0279] [0.0380]
%Age60+ -0.0609*** 0.00034
[0.0147] [0.0177]
%Value$1-$2 -0.152*** -0.162***
[0.00335] [0.00499]
%Value$2-$3 -0.269*** -0.285***
[0.00527] [0.00945]
%Value$3-$5 -0.344*** -0.412***
[0.00762] [0.0139]
%Value$5+ -0.569*** -0.534***
[0.00845] [0.0145]
Observations 121,135 73,059 124,141 74,717 124,141 74,717
R
2
0.388 0.245 0.540 0.424 0.161 -0.087
38
