EXPERIENCE OF PEOPLE OF COLOR, WOMEN, AND LOW-INCOME HOMEOWNERS IN THE HOME

advertisement
EXPERIENCE OF PEOPLE OF COLOR,
WOMEN, AND LOW-INCOME
HOMEOWNERS IN THE HOME
AFFORDABLE MODIFICATION
PROGRAM
Final Report
June 2012
Prepared for:
Open Society Foundations
Prepared by:
The Urban Institute
2100 M Street, NW ● Washington, DC 20037
EXPERIENCE OF PEOPLE OF COLOR, WOMEN, AND
LOW-INCOME HOMEOWNERS
IN THE
HOME AFFORDABLE MODIFICATION PROGRAM
Prepared by
Neil Mayer
Matt Piven
FINAL
June 2012
The Urban Institute
2100 M Street NW
Washington, D.C. 20037
TABLE OF CONTENTS
Executive Summary..................................................................................................................................... vii
1.
Introduction and Purpose ................................................................................................................ 1
2.
Data ................................................................................................................................................ 11
3.
Models and Methods ..................................................................................................................... 23
4.
HAMP Applicants Compared to Homeowners in Trouble ............................................................. 34
5.
Fair Housing Characteristics and HAMP Trial Loan Modifications................................................. 38
6.
Fair Housing Characteristics and HAMP Permanent Loan Modifications ...................................... 52
7.
Fair Housing Characteristics and Sustaining HAMP Permanent Modifications ............................. 65
8.
Summary of Findings for HAMP Direct Effects for Fair Housing Populations across Stages of
HAMP Activity ................................................................................................................................ 68
9.
Indirect Impact of HAMP by Fair Housing Populations.................................................................. 71
10.
Summary and Conclusions ............................................................................................................. 80
References .................................................................................................................................................. 83
Appendix ..................................................................................................................................................... 85
ii
LIST OF TABLES
Table ES-1. Summary of Findings for HAMP Direct Effects for Fair Housing Populations across Stages of
HAMP Activity. ...............................................................................................................................xiii
Table ES-2. Summary of Net (Direct and Indirect) Effects of Fair Housing Variables on Impacts of HAMP
for Various Outcomes. ...................................................................................................................xvi
Table 2-2. Comparison of Data Samples by Key Variables. ........................................................................ 16
Table 3-1. Trial Denial Reason Codes and their Relationship to Eligibility. ................................................ 25
Table 3-2. Variables for Models of Trial and Permanent Modifications. .................................................... 27
Table 4-2. Incomes of HAMP Applicants and of Homeowners with Seriously Delinquent and Foreclosed
Mortgages. ..................................................................................................................................... 36
Table 5-1. Fair Housing Populations in NANAs and Modified Loans. ......................................................... 40
Table 5-2. Analysis of Eligibility for Trial Modifications by Fair Housing Population. ................................. 41
Table 5-3. Comparison between Whites and Native Hawaiians and Pacific Islanders (NHPIs) in Effects of
Trial Modification Eligibility Standards. ......................................................................................... 42
Table 5-4. Predictive Accuracy and Summary Statistics for Trial Modification Multivariate Regression
Model. ............................................................................................................................................ 44
Table 5-5. Trial Modification Regression Results for Fair Housing Populations. ........................................ 44
Table 5-7. Seven Variables with Largest Negative Impact for Asians in Trial Modification Regression
Analysis. ......................................................................................................................................... 47
Table 5-8. Seven Variables with Largest Negative Impact for Native Hawaiians and Pacific Islanders in
Trial Modification Regression Analysis. ......................................................................................... 48
Table 5-9. Seven Variables with Largest Negative Impact for Low-Income Group in Trial Modification
Regression Analysis. ....................................................................................................................... 49
Table 6-2. Eligibility Analysis for Permanent Modifications, by Fair Housing Population. ......................... 57
Table 6-3. Predictive Accuracy and Summary Statistics for Multivariate Regression Model for Permanent
Modifications. ................................................................................................................................ 59
Table 6-5. Seven Variables with Largest Negative Impact for Asians in Permanent Modification
Regression Analysis. ....................................................................................................................... 61
Table 6-6. Seven Variables With Largest Negative Impact for Low-Income Group in Permanent
Modification Regression Analysis. ................................................................................................. 62
iii
Table 6-7. Fair Housing Population Results for Eligibility and Success in Obtaining HAMP Permanent
Modifications. ................................................................................................................................ 63
Table 6-8. Net Differential Rates in Obtaining Permanent Modifications, Combining Eligibility and
Approval among Eligible Trials, for Population Groups with Lower Rates of Eligibility, Approval of
Eligibles, or Both. ........................................................................................................................... 63
Table 7-1. Fair Housing Populations Compared to Counterparts in Sustaining of Permanent
Modifications. ................................................................................................................................ 66
Table 8-1. Summary of Findings for HAMP Direct Effects for Fair Housing Populations across Stages of
HAMP Activity. ............................................................................................................................... 69
Table 9-1. Summary of Net (Direct and Indirect) Effects of Fair Housing Variables on Impacts of HAMP for
Various Outcomes. ......................................................................................................................... 73
iv
LIST OF FIGURES
Figure 1-1. The HAMP Modification Process. ............................................................................................... 6
v
ACKNOWLEDGMENTS
The research for this report was supported by a grant from the Open Society Foundations. It
benefitted greatly from the issues raised early on and report comments from OSF’s Solomon Greene.
We also received valuable comments on our draft report from Geoff Walsh at the National Consumer
Law Center and from Lisa Rice and Debby Goldberg at the National Fair Housing Alliance.
Charles Calhoun of Calhoun Consulting undertook key elements of the analysis of net impacts of
HAMP. Ben Schmidt and an unnamed “webmaster” at the U.S. Treasury Department answered
numerous questions about the Treasury HAMP database. Peter Tatian at the Urban Institute provided
valuable editorial guidance.
vi
Executive Summary
Mortgage foreclosure has been a central component of the current deep recession, both as an
outcome of the economic crisis in general and housing market practices and trends in particular, and as
a cause of the recession’s depth and continuation as people struggle with high levels of debt and
depressed home values. The foreclosure problem has hit people of color—racial minorities and
Hispanics—particularly hard, with their high rates of subprime mortgages with elevated foreclosure
rates, large shares of recent home purchases, and low wealth levels heavily concentrated in housing
assets.
These differentials in mortgage costs and losses and their causes are themselves important
topics of fairness in housing and housing finance. Our focus here, however, is on how well federal policy
has responded to the problems, in particular in attempting to prevent foreclosures, in relation to the
scale of difficulties faced by important population groups. We concentrate specifically on distribution of
benefits of the Home Affordable Mortgage Program (HAMP).
A key question with respect to the public policy response to threatened additional foreclosures
is whether people of color receive assistance in proportion to their shares in the population of people in
mortgage difficulty and in danger of losing their homes. This study focuses on an expanded version of
that question: whether particularly vulnerable populations are being well served by the largest single
federal foreclosure prevention program. We broadened the question beyond communities of color to
include other populations that may face particular disadvantage in struggling with foreclosure: not only
racial minorities and Hispanics/Latinos, but also women and people at the low end of the income
spectrum. We refer to them below as fair housing populations, although in strict terms only race,
ethnicity, and gender fit within that formal designation. The groups appear to face many though not all
of the same issues in their home-buying experiences, and they overlap considerably—on such matters
as being the target of efforts to promote subprime mortgages, lacking substantial accumulated assets
with which to respond to economic contingencies, or sharing in concentration in the low end of the
income spectrum. We want to know whether public responses have been inclusive of all of them in their
reach and assistance.
The Home Affordable Modification Program (HAMP) is the largest federal program response to
the foreclosure crisis, in terms of effort to prevent foreclosures.1 By January, 2011, according to the data
accessible to us at the start of our study, over 2.7 million borrowers had entered the HAMP system since
its April, 2009 start. That number exceeded 3.4 million in August, 2011. When our research began, there
were 1.45 million modifications made under the program, of which about 580,000 had made it to
permanent status; and the numbers have grown since. The program provides federal financial incentives
to lenders/investors, servicers, and borrowers to modify existing mortgages, lowering payments and
providing means to deal with overdue payments to borrowers in order to keep homeowners in their
1
The actions of the Federal Reserve to keep interest rates low, especially in their impacts on adjustable rate
mortgages, may have impacted as many people but are not a foreclosure program response per se. Another large
federal program, the Neighborhood Stabilization Program, addresses impacts of foreclosures that have already
been carried to completion, leaving vacant homes, rather than seeking to prevent the foreclosures.
vii
homes. It is operated by the U.S. Treasury Department, in cooperation with the U.S. Department of
Housing and Urban Development, with funds allocated as part of Troubled Asset Relief Program (TARP)
appropriations.
Our central study question is: how well does HAMP serve racial minorities, Hispanics/Latinos,
women, and low-income people in seeking to modify loans in order to avoid foreclosures? We examine
first the direct impacts, in terms of those populations’ successful participation in HAMP itself, obtaining
HAMP mortgage loan modifications. Then we address indirect impacts, in which the creation and
implementation of HAMP may affect the same populations’ receipt of all loan modifications and other
mortgage default workout solutions. These may be HAMP-specific modifications, other proprietary2
modifications, or non-modification arrangements between lender/investors and borrowers like payment
plans or forbearances on overdue payments.3
Basics of the Home Affordable Modification Program
HAMP assistance is, during the period of our research4, restricted to owner-occupants of one to
four unit buildings, including fee simple and condominium ownership, for people who obtained their
loans before January 1, 2009. Borrowers must be either already delinquent in their payments or face
imminent default. Only those with housing payments exceeding 31 percent of income can be
considered.
Lender/investors can choose whether to participate, but the financial incentives have
apparently been sufficient to attract most of those with major market share. Lenders pay for the costs of
reducing borrower payments to 38 percent of income, and the HAMP program pays half the cost of
further reduction to a 31 percent level. There are financial incentives for borrowers, servicers, and
investors as well, dependent on modifications made and payments sustained afterward.
HAMP provides for a standardized reduction of payments to 31 percent of borrower incomes,
for approved modifications. Servicers move through a specified sequence of possible modifications to
loan terms to reach that level (the “waterfall”), beginning with lowering the interest rate, then
extending the term, and finally principal forbearance5. They are to approve modifications in which they
will receive a net financial benefit, based on the value of the mortgage unmodified with its risk of
default and the value of the loan modified and with its risk of redefault. Homeowners who apply for and
receive modification offers and choose to accept them receive trial loan modifications initially. They are
expected to make timely payments in the modified amounts. Those who continue to meet program
eligibility and other standards and make their payments then receive permanent modifications.
2
“Proprietary modifications” is language often currently used to refer to modifications not involving the federal
government but simply between borrowers and lender/investors.
3
We use a definition of modifications that produces reduced monthly payments through extensions of loan term
or lowering of interest rate and/or principal. See Mayer et al. (2010).
4
Newest changes allow investor owners as well, among other things.
5
A modified waterfall now applies to the Principal Reduction Alternative.
viii
Research Questions
This study addresses two sets of related questions about the extent of HAMP’s benefit for fair
housing populations compared to others. The first concerns the direct benefit to those homeowners
who are able to apply for and obtain HAMP mortgage modifications, with their lowered payments. The
second addresses the benefit that these and other homeowners may obtain insofar as HAMP’s
operation improves their access to HAMP modification, non-HAMP modifications and other mortgage
workouts combined.
Direct Impacts of HAMP for Fair Housing Populations
The structure of the HAMP program itself provides a natural shape for components of the
analysis of direct benefits. Homeowners with mortgage payment problems within our target populations
can become applicants, meet eligibility requirements, obtain trial modifications, receive permanent
modifications, and sustain those permanent modifications. These steps suggest five research questions
regarding direct HAMP benefits for the fair housing populations.
1. To what extent do fair housing populations enter the HAMP applications process in
proportion to their shares of the homeowner population facing mortgage difficulty?
2. What share of our target populations who enter the application system meet HAMP
eligibility requirements, compared to their counterpart majority populations?
3. Among eligible homeowners, how do fair housing populations perform in obtaining trial
modifications in proportion to their numbers?
4. Of homeowners who obtain trials, how do our focus populations fare in receiving
permanent mortgage modifications, compared to respective majority counterparts?
5. How well are fair housing populations able to sustain permanent modifications received, in
proportion to their numbers?
Indirect and Net Impacts of HAMP for Fair Housing Populations
HAMP may affect mortgage outcomes for homeowners beyond the direct effect of HAMP
modifications. Various observers have indicated that the structure of the HAMP program, once it was
created, may have been important in shaping lender/investors’ proprietary modification and other
workout decisions. Its standards for affordability such as the housing cost to income ratio of 0.31, its
sequenced waterfall of types of modification, its attention to and methods for comparing present value
of current and modified loans, and other program specifics have apparently had impact on the way
lenders and their servicers consider homeowners in search of assistance with their delinquent or
ix
potentially delinquent mortgages.6 The standardization HAMP helped to induce may have helped to lead
to increases in the numbers of loan modifications and in the extent to which loan amendments involve
actual reductions in mortgage payments, which many of the first so-called modifications did not. 7
Indeed, some observers have suggested these effects may be larger than the direct effects of the HAMP
program.
At the same time, HAMP can have effects on the broader picture for mortgage modifications in
an opposite direction. Lender/investors, servicers, and homeowners may enter into HAMP modifications
in cases in which, absent HAMP and its financial incentives, proprietary modifications or other workouts
would have been provided instead. In such cases, HAMP benefits to fair housing and other populations
are not net gains but substitutes. Thus in directions both positive and negative, measuring only HAMP
modifications themselves may misstate the extent of the program’s benefits for any population.
Comparing HAMP impacts for fair housing populations to majority counterparts in terms only of HAMPspecific modification, as a result, provides only a partial analysis of the possible total comparison of
effects.
Therefore, we conducted a second set of analyses that looks at the net effects of HAMP’s
initiation on a series of mortgage outcomes. It examines outcomes among a broad sample of
homeowners—including borrowers that sought and in some cases received HAMP modifications and
others that received non-HAMP modifications and other workouts or no assistance. It divides their
experience between the period prior to and following HAMP’s initiation in April, 2009, in order to isolate
HAMP’s effects. It focuses on whether fair housing populations received at least the same level of net
benefits, including direct and indirect effects, from HAMP in terms of the overall outcomes for
homeowners as did majority counterparts.
We examine how fair housing populations shared in the net impact of HAMP on overall
mortgage outcomes of the following kinds:
1. Size of modifications in terms of reduced monthly payments. Did HAMP contribute as much
to payment reduction for fair housing populations as for other groups?
2. Foreclosure stops. For loans that had already begun foreclosure process, did HAMP increase
the rate of stopping those foreclosures (leaving them still delinquent or current) as greatly
for fair housing populations as for others?
3. Serious mortgage delinquencies and foreclosures started that were cured to current status,
whether by modification or not. Did HAMP help increase cure rates as greatly for fair
housing populations as for others?
6
Mortgage counselors and lending industry participants have both suggested the importance of this kind of effect.
See Mayer and Piven (2011).
7
The majority of early “modifications” actually raised mortgage payments, often by capitalizing overdue payments
into mortgage principal (Mayer et al. 2010).
x
4. Cures sustained thereafter, avoiding redefault. Did HAMP increase the share of fair housing
homeowners able to sustain their modifications once completed as much as it did for
counterpart majority populations?
Neighborhood distribution of HAMP impact: a key question not yet addressable.
The discussion of research questions above focuses entirely on HAMP impacts, direct or indirect,
on mortgage modifications and other mortgage outcomes for individual homeowner households. It does
not address the question of whether HAMP serves neighborhoods with high percentages of households
from among racial minorities, Hispanics, women, and low-income people in proportion to those
neighborhoods’ levels of mortgage difficulties. Unfortunately the Treasury’s publicly released database
for HAMP does not include geographic information disaggregated below the level of MSAs, although the
Department does possess address-level information. Therefore we have as yet been unable to
determine the program’s geographic reach at the neighborhood or even city or county level. It would be
valuable to know, for instance, if neighborhoods with high concentrations of people of color were
receiving HAMP benefits in relation to their share of problem mortgages, in addition to whether FHV
populations as a whole are doing so; but that study is not feasible with the data that are publicly
available. Data and Methods
For studying HAMP’s direct impacts, we rely heavily on the loan-level database which Treasury
has made available to the public beginning in January, 2011. It covers all homeowners who enter into
the HAMP application process, although it has a more extensive set of measures available for those who
were found eligible and were then individually analyzed for possible modifications. To consider whether
fair housing populations obtained trial modifications in proportion to their share of all applications, and
obtained permanent modifications in proportion to their share of all trials, we used both simple
tabulations and multivariate logistic regression techniques. The latter are particularly useful in isolating
any separate impact of fair housing characteristics after controlling for many household, property,
original loan, and modification factors. Data and resource constraints limited us to tabulations only in
analyzing the role of fair housing characteristics of borrowers in proportions of applications compared to
all borrowers in trouble and sustained modifications (no redefault) compared to all permanent
modifications.
For studying HAMP’s indirect and net impacts, we rely on adaptation of a series of models of
mortgage outcomes for all borrowers (not just HAMP applicants) developed in a previous piece of
research by the authors and colleagues (Mayer et al. 2011). These are multivariate regressions focused
on the interaction between the start of the HAMP program and the impacts if any of fair housing
characteristics. For analysis of loan modification size, we employ ordinary least squares regression. For
the other mortgage outcomes enumerated in our research questions, we employed logistics regressions.
In both cases we treat HAMP as an intervention beginning in April, 2009 and compare experiences
before and after it began. The data are drawn from LPS Analytic’s proprietary database for 60 percent of
U.S. mortgages, the National Foreclosure Mitigation Counseling Program database, HMDA, and the U.S.
Census.
xi
Findings
Our consistent conclusion from these multiple analyses is that race, ethnicity, gender, and
income have very little impact on homeowners’ direct participation in HAMP and on the net benefits of
HAMP for homeowners’ overall mortgage outcomes.
In terms of direct benefits, HAMP serves most racial minorities, Hispanics, women and lowincome people at least in the same proportion as they do their respective reference populations of
whites, non-Hispanics, men, and higher income people. This is true at every step in the program: for
program applicants compared to all households with mortgage problems, for eligible homeowners
compared to all applicants, for trial modification recipients compared to all eligible households, for
permanent modification recipients compared to trial recipients, and for sustained modifications
compared to all permanent modifications. Most fair housing populations enter the program and
advance through the program steps in at least the proportion of their share of candidates for
advancement.
The direct benefit results for specific populations are summarized in Table ES-1. The table
indicates the direction of benefits for a fair housing population (e.g. African Americans) relative to those
for its counterpart majority population (in this example, whites): positive if the FHV group is more likely
than its counterpart to progress to a given stage, negative if less likely, neutral if no statistically
significant difference. For each group, it reviews their progress at each stage of the program:
proportions of applicants relative to proportions of troubled homeowners, eligibles for trials relative to
applicants, successful recipients of trials relative to eligibles, eligibles for permanent modifications
relative to trial recipients, successful recipients of permanent modifications relative to eligibles for
them, and those able to sustain their permanent modifications relative to those with such modifications.
xii
Table ES-1. Summary of Findings for HAMP Direct Effects for Fair Housing Populations across Stages of HAMP Activity.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: Positive means that the fair housing population fares better than its counterpart. In the left column only, the racial categories include only
non-Hispanics, whereas in the other analyses, those groups include Hispanics.
xiii
Among racial groups, African Americans’ share of HAMP activity exceeds its share of candidate
borrowers at every stage but the last. For them, the positive differentials they obtain at application and
trial and permanent modification stages outweigh the single small negative differential in terms of
sustaining modifications.8 Asians’ positive or neutral shares relative to whites at every stage produce an
overall positive position. Among the smaller racial groups, American Indians and Alaska Natives have
positive or neutral experience compared to whites at each stage except review of eligibility for trials,
and that single disadvantage is fully offset by just their positive position in trial approvals among
eligibles. People of more than one race come very close to balancing between positives and negatives.
Native Hawaiians and Pacific Islanders are the one racial group which benefits from HAMP less
than in proportion to their entrance into it. We lack separate data on their share of troubled loans. But
among those who start applications, we know they lag in gaining trials, both in eligibility and in
approvals among eligibles; and their advantage in eligibility at the permanent modification stage does
not fully offset those lags.
Hispanics present an only slightly more mixed picture than do populations by race. Among
HAMP-eligible homeowners, their percentages of trial and permanent modifications approved are
modestly smaller than for non-Hispanics. But they are more likely to be eligible, and they enter the
application system more than in proportion to their share of troubled loans. Overall Hispanics have a
very slender positive margin in HAMP participation through its multiple steps, compared to nonHispanics.
Women obtain and sustain HAMP modifications at least as successfully as men, with a positive
or neutral share of modifications out of candidates for each stage for which we have data. An important
gap is that, as with smaller racial groups, we do not have data for women’s share of troubled mortgage
situations, to which to compare their modification success.
Finally, people of lower income are faring positively in direct HAMP program participation and
success. Among applicants, they are more than proportionally eligible at the trial and permanent stage;
they receive more than a proportionate share of trial modifications among eligibles; and they more
frequently sustain permanent modifications. An important limit to this conclusion is that we lack a fully
comparable measure of troubled loans among low-income borrowers to compare to their share of
entering applicants to HAMP. But our rough approximation indicates a significant advantage to people
at the lower end of the income scale. Given the positive results for other stages of HAMP consideration,
we can reasonably expect low-income people to do well in the program.
In terms of net benefits of HAMP—combined direct and indirect-- on total mortgage outcomes,
fair housing populations also do well.9 But the findings are not quite as consistent as for direct
participation in HAMP. While most of the population groups fare at least as well as their counterparts, in
8
The numerical differences are reported in full in the body of the report. The statements about the net outcomes
balancing eligibility and approval outcomes that sometimes are opposite in direction are based on those exact
figures.
9
Because the net impacts analysis was not undertaken by gender, we do not have outcomes for women.
xiv
most steps in the HAMP process, African Americans, Asians (to a lesser extent), and low-income people
do less well in one or more program steps. The results are summarized in Table ES-2 for each population
group relative to its counterpart, for each of the net outcomes. Note that in this table, in the last two
columns only, a “positive” outcome means more redefaults so that “positive” is actually
disadvantageous to the population group. In the other columns, positive means beneficial (as it does
throughout Table ES-1).
xv
Table ES-2. Summary of Net (Direct and Indirect) Effects of Fair Housing Variables on Impacts of HAMP for Various Outcomes.
Source: Author's computations from NFMC and LPS data.
Note: For the first four outcomes, “positive” results are beneficial to the homeowner and “negative” are disadvantageous. In the last two
outcomes on redefaults, “negative” results (fewer redefaults) are beneficial.
xvi
African Americans get more net help from HAMP than do whites in modification size (reduced
payments) and in curing delinquencies and foreclosure, but less help in halting foreclosure processes
once begun and in sustaining cures of defaults. African Americans do benefit from HAMP in all but one
of those cases, but the gains are smaller than for whites. Asians receive higher HAMP benefits than
whites in loan modification size and foreclosure stops, and equal benefits for three of the remaining
outcomes. Only in loan modification cures is their benefit from HAMP smaller than that of whites. Other
races combined fare equally with whites in terms of HAMP benefits on all outcomes except modification
size, in which they have an advantage.
Hispanics quite consistently reap more net impact from HAMP than do non-Hispanics. The
HAMP impacts are larger for Hispanics in terms of modification size and foreclosure stops and equal in
cures. HAMP contributes more greatly to decline in Hispanic redefaults from both modification and nonmodification cures.
Lower income homeowners consistently receive less net benefit in overall mortgage outcomes
from HAMP’s creation than do people of higher income. Higher earners get a larger boost in loan
modification size. They get a modestly higher benefit in terms of foreclosure stop frequency, and
likelihood of modification cures as a result of HAMP. They don’t differ in obtaining modification cures
but they do get modestly more non-modification cures. HAMP also gives higher income people greater
help in avoiding redefaults of both types of cures. Only the difference in modification payment reduction
is substantial in size. That assumes all other factors but income are held constant. A low income
homeowner with the housing payments, loan amount, and other factors at the same level as an average
borrower will get significantly less of a payment reduction benefit from HAMP than a household with
the same other characteristics but average income. However, typically the low-income owner will also
have other financial characteristics that differ from their higher income counterparts. These may well
moderate the observed differences in this modification size impact that HAMP appears to be having for
differing income groups.
In sum, while the HAMP program has experienced both successes and failures, it systematically
benefits people—both directly and on a net basis including directly and indirectly—in very much the
same way regardless of homeowners’ race, ethnicity, gender, and income. There is additional research
to be done to tighten these results, as appropriate data and resources are available. Especially
important is a look at differential impacts of HAMP across neighborhoods of varying racial and ethnic
composition. But our results about impacts on households are robust and consistent.
Measures that strengthen HAMP’s ability to produce good mortgage outcomes for all
homeowners would be highly desirable. Other analyses have identified areas for improvement in HAMP
structure and operation and our study does not refute (nor explicitly support) them. The central lesson
of our research is that, at least in terms of the distribution of household (as distinct from neighborhood)
outcomes, attention would best be focused on the changes and additional supports that benefit all
homeowners with mortgage difficulties. Emphasis can be placed on extending the program’s reach and
effects to all homeowners consistent with its equitable distribution to date, rather than on changing the
mix of beneficiaries in terms of fair housing populations.
xvii
xviii
1.
Introduction and Purpose
Mortgage foreclosure has been a central component of the current deep recession, both as an
outcome of the economic crisis in general and housing market practices and trends in particular, and as
a cause of the recession’s depth and continuation as people struggle with high levels of debt and
depressed home values. It is difficult to find precise estimates of the magnitude of the national
foreclosure crisis, but Mayer (2011) writes that at least three million foreclosures have been completed
since the crisis began. Furthermore, estimates from Bocian et al. (2011) suggest that in February, 2011,
at least an additional 3.6 million loans were at serious risk of foreclosure.10
The foreclosure problem has hit people of color—racial minorities and Hispanics—particularly
hard. Within these populations, homeowners were more likely to take on subprime mortgages than did
others.11 Subprime mortgages have, without question, had much higher rates of foreclosure.12 People of
color also made bigger shares of home purchases relatively recently than in the past, and they faced
worse deterioration in employment, less non-housing wealth to see them through crisis, and other
factors contributing to higher foreclosure rates.13 African Americans and Hispanics suffered staggering
losses in wealth, given the housing base of their assets, absolutely and relative to whites (Taylor et al.
2011).
These differentials in mortgage costs and losses and their causes are themselves important
topics of fairness in housing and housing finance. Our focus here, however, is on how well federal policy
has responded to the problems, in particular in attempting to prevent foreclosures, in relation to the
scale of difficulties faced by important population groups.
A key question with respect to the public policy response to threatened foreclosures is whether
people of color received assistance in proportion to their shares in the population of people in mortgage
difficulty and in danger of losing their homes. This study focuses on an expanded version of that
question: whether particularly vulnerable populations are being well served by the largest single federal
foreclosure prevention program.
We broadened the question beyond communities of color to include other populations that may
face particular disadvantage in struggling with foreclosure. We include not only racial minorities and
Hispanics/Latinos, but also women and people at the low end of the income spectrum. The groups
10
Bocian et al. (2011) estimate that 3.6 million loans which originated between 2004 and 2008 were at risk in
February 2011, so if one includes all origination years, the number is presumably higher.
11
Bocian et al. (2011) estimate that Asians, Hispanics, and African Americans were all at least 30% more likely than
non-Hispanic Whites to originate a loan with at least one risky feature during the 2004-2008 period. Risky features
include: hybrid or option ARMs, prepayment penalties, and interest rates which are significantly higher than
Treasury bonds of similar maturity. Asians have a disparity ratio of 1.3, compared to 1.6 for Hispanics and African
Americans.
12
Bocian, Li, and Ernst (2010) estimate that, in an analysis of 2007-2009 foreclosure rates for 2005-2008
originations, that subprime loans ended in foreclosure 16.6% of the time, compared to 5.7% for the entire sample.
13
According to Carr, Anacker, and Mulcahy (2011), people of color obtained about 34 percent of loans in 20052008, well above their historical share; and the homeownership rates for African Americans and Hispanics peaked
in 2004 and now stand at their lowest rate in 15 years following high foreclosure rates.
1
appear to face many though not all of the same issues in their home-buying experience, and they
overlap considerably. As illustration, observers suggest that all may have been targets of efforts to
promote subprime loans in the past; they historically have limited assets to use to prevent defaults
when they face employment distress, and people of color and women are over-represented in the lowincome population. We want to know whether public responses have been inclusive of all of them in
their reach and assistance.
The Home Affordable Modification Program (HAMP) is the largest federal program response to
the foreclosure crisis, in terms of effort to prevent foreclosures.14 By January, 2011, according to the
data accessible to us at the start of our study, over 2.7 million borrowers had entered the HAMP system
since its April, 2009 start. That number exceeded 3.4 million in August, 2011. According to the authors’
analysis of the HAMP public data base, as of January, 2011, there were 1.45 million modifications made
under the program, of which about 580,000 had made it to permanent status. The program provides
federal financial incentives to lenders/investors, servicers, and borrowers to modify existing mortgages,
lowering payments and providing means to deal with overdue payments to borrowers in order to keep
homeowners in their homes. It is operated by the U.S. Treasury Department, in cooperation with the
U.S. Department of Housing and Urban Development, with funds allocated as part of Troubled Asset
Relief Program (TARP) appropriations.
Our central study question is: how well does HAMP serve racial minorities, Hispanics/Latinos,
women, and low-income people in seeking to modify loans in order to avoid foreclosures? We examine
first the direct impacts, in terms of those populations’ successful participation in HAMP itself, obtaining
HAMP mortgage loan modifications. Then we address indirect impacts, in which the creation and
implementation of HAMP may affect the same populations’ receipt of all loan modifications and other
mortgage default workout solutions. These may be HAMP-specific modifications, other proprietary15
modifications, or non-modification arrangements between lender/investors and borrowers like payment
plans or forbearances on overdue payments.16
Basics of the Home Affordable Modification Program
HAMP assistance is, during the period of our research data, restricted to owner-occupants of
one to four unit buildings, including fee simple and condominium ownership, for people who obtained
their loans before January 1, 2009.17 Borrowers must be either already delinquent in their payments or
facing imminent default. Only those with housing payments exceeding 31 percent of income can be
considered.
14
The actions of the Federal Reserve to keep interest rates low, especially in their impacts on adjustable rate
mortgages, may have impacted as many people but are not a foreclosure program response per se. Another large
federal program, the Neighborhood Stabilization Program, addresses impacts of foreclosures that have already
been carried to completion, leaving vacant homes, rather than seeking to prevent the foreclosures.
15
“Proprietary modifications” is language often currently used to refer to modifications not involving the federal
government but simply between borrowers and lender/investors.
16
We use a definition of modifications that produces reduced monthly payments through extensions of loan term
or lowering of interest rate and/or principal. See Mayer et al. (2010).
17
Since then, eligibility changes allow for non-resident borrowers, among other adjustments.
2
Lender/investors can choose whether to participate, but the financial incentives have
apparently been sufficient to attract most of those with major market share. Non-participants are not
eligible to receive the incentives on any individual loans; only those who state their participation and
provide all applicants consideration for modification under program guidelines can benefit. Though
nothing precludes borrowers from applying for in-house modifications, borrowers whose lenders are
HAMP participants must be considered for a HAMP modification if they apply for one. Lenders pay for
the costs of reducing borrower payments to 38 percent of income, and the HAMP program pays half the
cost of further reduction to a 31 percent level. The primary financial incentives for borrowers are the
reduction in modified loan payments and payments up to $5,000 in total by Treasury toward principal
reduction, portions of which are triggered by each timely payment on a modified loan. In addition,
servicers are compensated as much as $1,600 at the time of awarding a permanent modification
(depending on how far behind the borrower had fallen), and up to $3,000 per borrower over three years
after the modification agreement is completed (up to $1,000 each year) for modifications that stay in
good standing.
HAMP provides for a standardized reduction of payments to 31 percent of borrower incomes,
for approved modifications. Servicers move through a specified sequence of possible modifications to
loan terms to reach that level (the “waterfall”), beginning with lowering the interest rate, then
extending the term, and finally principal forbearance.18 They then make a calculation of net present
value to the investor of the mortgage in its present form and as modified and, provided that the
borrower meets the program eligibility requirements and has submitted the necessary paperwork, the
servicer must approve modifications that show higher present value for the modified version.19 Net
present value calculations are based on estimated risks that the loan will be repaid under present and
under modified terms, the value of the home at resale if it is foreclosed upon now or later, the interest
rate and payment stream before and after modification, and other financial factors. The Treasury
Department provided a formula and parameters for net present value calculation given the various
factors, but investors can substitute their own formulae and parameters given their own lending
experience, with Treasury’s approval.
Homeowners who apply for and receive modification offers and choose to accept them receive
trial loan modifications initially. They are expected to make three timely and complete payments in the
modified payment amounts as well as to provide updated and complete documentation of income and
other information. Those who continue to meet program eligibility and other standards and make their
payments in satisfactory fashion then receive permanent modifications. These modifications run for five
years at a fixed interest rate of as low as 2 percent, after which interest rates rise by 1 percent annually
until reaching a cap. To retain their modified loans, homeowners must continue to make timely
payments, because falling behind by three months results in a loss of good standing under the program.
The HAMP program has produced large numbers of modifications (Office of the Comptroller of
the Currency and Office of Thrift Supervision 2011), though not nearly as many as initially anticipated. It
18
19
An alternative waterfall applies to the newer Principal Reduction Alternative of the program.
They may approve others as well.
3
has received support in terms of design to fit foreclosure problems, notably other than job loss (Cordell
et al. 2009). It has been criticized for problems with its policies and operations, especially early in its life
(e.g., U.S. Government Accountability Office 2010; White 2011; Stanley 2011), in areas including
consistency, transparency, and promptness of review, heavy paperwork burdens, and cooperation by
and oversight of servicers. Review data has been made available showing improving performance over
time (Making Home Affordable 2011b). But until very recently no information has been available
specifically about the demographic and socio-economic characteristics of its beneficiaries.
Populations and Fair Housing
We assess participation in HAMP of people divided into six racial categories, two ethnic groups,
genders, and five income quintiles. The racial groupings are: African Americans; American Indians and
Alaska Natives (AIANs); Asians; Native Hawaiians and Pacific Islanders (NHPIs); other races, which in the
available HAMP data constitutes people of more than one race; and whites. Our concentration is on the
participation of people in the first five racial groups compared to whites or to all races combined. The
ethnic groups are Hispanic including Latinos and non-Hispanic, focusing on outcomes for the former
compared to the latter; and we focus on female homeowners compared to the large male base. Our
concentration among income groups is on the bottom 20 percent and to a lesser extent the bottom 40
percent, compared to higher income people.
In referring to our set of target populations without having to repeat the lengthy list, we have
chosen a short-cut of calling the first five racial groups, Hispanics, women, and low-income people “fair
housing populations.” This is not literally precise language. The legal definitions of population groups for
attention in fair housing and lending legislation include other groups not in our study (e.g., disabled
people). And low-income people, an important focus in our study, are not among the official fair housing
groups under either the Fair Housing Act or the Equal Credit Opportunity Act. We also refer to our set of
“fair housing” population groups as target populations and focus populations in the report text. The
other definitional designation of note is the use of the terms “counterpart” or “comparison” populations
to refer to comparison by race to whites as the comparison group, ethnicity to non-Hispanics, women to
men, and lowest quintile(s) to the remainder higher quintiles, without having to repeat each comparison
base.
Research Questions
As indicated above, this study addresses two sets of related questions about the extent of
HAMP’s benefit for fair housing populations compared to others. The first concerns the direct benefit to
those homeowners who are able to apply for and obtain HAMP mortgage modifications, with their
lowered payments. The second addresses the benefit that these and other homeowners may obtain
insofar as HAMP’s operation improves their access to HAMP modification, non-HAMP modifications and
other mortgage workouts combined.
Direct Impacts of HAMP for Fair Housing Populations
4
The structure of the HAMP program itself provides a natural shape for components of the
analysis of direct benefits. Homeowners in our target populations can become applicants, meet
eligibility requirements, obtain trial modifications, receive permanent modifications, and sustain those
permanent modifications. The process of pursuing a HAMP modification is presented in Figure 1-1. The
bottom row of boxes represents a borrower moving successfully through various stages we analyze on
the way from deciding to apply to sustaining a permanent modification, and the top row reflects the
steps at which a troubled borrower may opt out or be unsuccessful.
5
Figure 1-1. The HAMP Modification Process.
Source: the authors, based on HAMP documentation.
6
These steps suggest five research questions regarding direct HAMP benefits for the fair housing
populations.
1. To what extent do fair housing populations enter the HAMP applications process in
proportion to their shares of the homeowner population facing mortgage difficulty?
2. What share of our target populations who enter the application system meet HAMP
eligibility requirements, compared to their counterpart majority populations?
3. Among eligible homeowners, how do fair housing populations perform in obtaining trial
modifications in proportion to their numbers?
4. Of homeowners who obtain trials, how do our focus populations fare in receiving
permanent mortgage modifications, compared to respective majority counterparts?
5. How well are fair housing populations able to sustain permanent modifications received, in
proportion to their numbers?
Insofar as fair housing populations are receiving their share of positive benefits at each of these
stages of advancement in program action, HAMP is shown to be serving them appropriately relative to
their shares of need for assistance. If one or more of the focus populations falls short at one or more of
the five steps, then this research identifies the shortfall and measures the extent of the shortcomings. It
also specifies the specific program points at which they occur, in order to suggest where intervention
may be appropriate to move the target group toward benefits more proportionate to their degree of
mortgage difficulty.
Indirect and Net Impacts of HAMP for Fair Housing Populations
HAMP may affect mortgage outcomes for homeowners beyond the direct effect of HAMP
modifications. Various observers have indicated that the structure of the HAMP program, once it was
created, may have been important in shaping lender/investors’ proprietary modification and other
workout decisions. Its standards for affordability such as the housing cost to income ratio of 0.31, its
sequenced waterfall of types of modification, its attention to and methods for comparing present value
of current and modified loans, and other program specifics have apparently had impact on the way
lenders and their servicers consider homeowners in search of assistance with their delinquent or
potentially delinquent mortgages. 20 The standardization HAMP helped to induce may have helped to
lead to increases in the numbers of loan modifications and in the extent to which loan amendments
involve actual reductions in mortgage payments, which many of the first so-called modifications did
20
Mortgage counselors and lending industry participants have both suggested the importance of this kind of effect
(Mayer and Piven 2011).
7
not.21 Indeed, some observers have suggested these effects may be larger than the direct effects of the
HAMP program.
At the same time, HAMP can have effects on the broader picture for mortgage modifications in
an opposite direction. Lender/investors, servicers, and homeowners may enter into HAMP modifications
in cases in which, absent HAMP and its financial incentives, proprietary modifications or other workouts
would have been provided instead. In such cases, HAMP benefits to fair housing and other populations
are not net gains but substitutes22. Thus in directions both positive and negative, measuring only HAMP
modifications themselves may misstate the extent of the program’s benefits for any population.
Comparing HAMP impacts for fair housing populations to majority counterparts in terms only of HAMPspecific modification, as a result, provides only a partial analysis of the possible total comparison of
effects.
Therefore, we conducted a second set of analyses that looks at the net effects of HAMP’s
initiation on a series of mortgage outcomes, combining the direct effects of HAMP modifications and the
indirect effects in treatment of other homeowners. It examines outcomes among a broad sample of
homeowners—including borrowers that sought and in some cases received HAMP modifications and
others that received non-HAMP modifications and other workouts or no assistance. It divides their
experience between the period prior to and following HAMP’s initiation in April, 2009, in order to isolate
HAMP’s effects. It focuses on whether fair housing populations received at least the same level of net
benefits, including direct and indirect effects, from HAMP in terms of the overall outcomes for
homeowners as did majority counterparts. This set of analyses actually took place as a part of another
Urban Institute project, evaluation of the National Foreclosure Mitigation Counseling program (NFMC)
for NeighborWorks America.©
Using that project’s results, we examine how fair housing populations shared in the net impact
of HAMP on overall mortgage outcomes of the following kinds:
1. Size of modifications in terms of reduced monthly payments. Did HAMP contribute as much
to payment reduction for fair housing populations as for other groups?
2.
Foreclosure stops. For loans that had already begun foreclosure process, did HAMP increase
the rate of stopping those foreclosures (leaving them still delinquent or current) as greatly
for fair housing populations as for others?
3. Serious mortgage delinquencies and foreclosures started that were cured to current status,
whether by modification or not. Did HAMP help increase cure rates as greatly for fair
housing populations as for others?
21
The majority of early “modifications” actually raised mortgage payments, often by capitalizing overdue
payments into mortgage principal (Mayer et al. 2010).
22
Although the quality of the modification may be better or worse.
8
4. Cures sustained thereafter, avoiding redefault. Did HAMP increase the share of fair housing
homeowners able to sustain their modifications once completed as much as it did for
counterpart majority populations? 23
5. Taking of homes for fair housing populations as greatly as it did for other borrowers?
In each of questions 2-4, we examine the change in likelihood of obtaining favorable or avoiding
unfavorable outcomes as a result of HAMP’s initiation, for each target population compared to majority
counterparts. The exception is analysis of size of modification, which simply compares dollar monthly
payment savings as a result of modifications as contributed by HAMP’s operation, between fair housing
and counterpart groups.
Neighborhood distribution of HAMP impact: a key question not yet addressable.
The discussion of research questions above focuses entirely on HAMP impacts, direct or indirect,
on mortgage modifications and other mortgage outcomes for individual homeowner households. It does
not address the question of whether HAMP serves neighborhoods with high percentages of households
from among racial minorities, Hispanics, women, and low-income people in proportion to those
neighborhoods’ levels of mortgage difficulties. Certainly there is reason to anticipate concentrated
mortgage difficulties in communities with high levels fair housing populations, given patterns of
subprime lending. It is important to know how HAMP assistance is geographically distributed, both
because inadequate reach to certain neighborhoods may disadvantage their homeowner households
who might use HAMP in gaining modifications, and because other resident households—homeowners
and renters—and property holders may be injured by higher foreclosures in the communities than if
HAMP were more active there.
Unfortunately the Treasury’s publicly released database for HAMP does not include geographic
information disaggregated below the level of MSAs, although the Department does possess addresslevel information. Therefore we have as yet been unable to determine the program’s geographic reach
at the neighborhood or even city or county level. It would be valuable to know, for instance, if
neighborhoods with high concentrations of people of color were receiving HAMP benefits in relation to
their share of problem mortgages, in addition to whether FHV populations as a whole are doing so; but
that study is not feasible with the data that are publicly available.
Limited Previous Analysis
Race and ethnicity of beneficiaries of foreclosure prevention assistance has been at times
highlighted as an important topic for research (e.g., Carr, Anacker, and Mulcahy 2011), along with
income (particularly in relation to people of color). Gender of beneficiaries does not appear to have
received similar mention. Two pieces of recent research have addressed race, ethnicity, and in one case
income of recipients of mortgage modifications in general—not HAMP modifications per se. Been et al.
(2011), focusing on New York City, found that in general race and ethnicity were not among the
significant determinants of mortgage loan modifications for people in mortgage difficulty, with the
23
We are able to address this only for a limited initial period, not including the time after five years in which
interest rates are allowed to be increased on HAMP modifications.
9
exception of a modest negative effect for non-Hispanic Asians. Collins and Reid (2010), addressing a
wider geography, found no evidence of disparate impact by race or income on the receipt of loan
modifications.
The U.S. Treasury has itself provided one initial analysis of the distribution of trial and
permanent modification recipients, and those turned down for trials, by race and ethnicity (Making
Home Affordable 2011a). Their basic tabulations found no major discrepancies between the percentages
of denials, trials, and permanent modifications received by racial and ethnic minorities—people of color
were, using the simplest analysis without controlling for other factors, receiving temporary and
permanent benefits in proportion to their share of application entries.
The same report provided tabular information on the income ranges of recipients of trials and
permanent modifications. Households in each income range received about the same percentage of
permanent modifications as they did trial modifications, so that no large discrepancies by income
appeared in the movement from trials to permanent. Both the race and ethnicity and income
tabulations were based on the Treasury loan level database which we employ in this study, first released
as of January, 2011.
A recent report by the California Reinvestment Coalition (2011) also took advantage of the
Treasury data, focusing on four MSAs within California. Looking specifically at the database’s
information on reasons why trial or permanent modifications were denied, it found suggestive evidence
that people of color might be facing some extra obstacles. They were more likely to have been denied
because of incomplete applications, which CRC suggested might be related to language barriers and/or
servicers losing documents, based on their interviews with foreclosure prevention counselors. And CRC
found people of color more likely to be designated as declining modifications that were offered, which
counselors thought might be related to borrowers being steered to inferior non-HAMP modifications as
they had been steered to subprime loans. CRC’s survey of counselors produced mixed responses on the
direct question of whether people of color were receiving worse treatment than whites, with a majority
saying no—that their clients were often treated poorly in pursuit of HAMP modifications regardless of
color.
The Treasury and CRC reports begin to address the question of direct benefits of HAMP, in
limited ways. No previous research addresses the indirect and net impacts of HAMP as we have outlined
them earlier in this section.
10
2.
Data
The primary data sources for our study of HAMP’s direct impacts on fair housing populations are
different from those for the indirect and resulting net impacts. The direct impact research focuses on
those homeowners who have at least begun the application process for a HAMP modification and
therefore entered the Treasury’s program database. The indirect and net impact research deliberately
focuses on a sample of homeowners more broadly representing the homeowner population, who may
or not have come into contact with HAMP, in order to encompass HAMP’s impact on the broader
housing market’s foreclosure outcomes. We therefore deal with the data sources for the two sets of
analyses separately.
Data for Analysis of HAMP’s Direct Impacts
Treasury’s publicly available data on HAMP includes detailed data on individual homeowner
households and their mortgages and properties. It is supplied by the participating lender/investors and
servicers from records concerning borrowers seeking HAMP assistance. Updated versions of the data are
supplied monthly, including all borrowers who entered the HAMP system and have either been denied
assistance or approved for trial or permanent mortgage modifications. Homeowners in process of
applying and awaiting a decision regarding approval or denial of a trial are not included until a decision
is reached.
In our research we use the January, 2011 data download,24 which was at the time we began our
research work the most recently issued data. It included the 2,708,234 homeowners who had at least
entered the HAMP application process through the previous month. Available information is provided in
two data files: the loan modification file and the net present value file.
The loan modification file includes at least some information for every participating owner,
including those stopping short of completing applications, eventually ruled ineligible, or denied trials for
other reasons, as well as those reaching trial and permanent modification stages. The file includes data
on the current status of the loan, reason codes for why applicants were denied trial modifications or
were denied a permanent modification after having obtained a trial modification, as well as a variety of
borrower, loan, and property characteristics. Several loan and borrower characteristic variables indicate
whether they are specific to pre- or post-modification conditions. Critically for the purposes of this
research, the loan modification file includes data for participants’ race, ethnicity, and gender. However,
as we shall detail shortly, that information is not complete for every homeowner. Borrower income, the
other population characteristic of focus in this study, is provided not in this file but in the net present
value file (see below). The loan modification file also provides data to establish the status of each
applicant: people who have been denied trial modifications (not approved/not accepted, hereafter
NANAs); approved trials, both currently active and disqualified or cancelled after initial approval; and
approved permanent modifications, both active, cancelled, and loans paid off. And it provides denial
24
US Department of Treasury, “Making Home Affordable Data File,” http://www.treasury.gov/initiatives/financialstability/results/Pages/mha_publicfile.aspx
(accessed December 11, 2011).
11
reason codes for the rejected NANAs in forms that allow us to differentiate those denied for reasons of
eligibility and those denied for other causes.25
The net present value (NPV) file contains information for those homeowners whose application
was advanced to review of the value of the loan to its lender/investor in its potentially HAMP-modified
form, compared to the loan in its existing form (given estimated risks of foreclosure and other factors as
noted earlier). The great majority of loans that were eventually modified, though not all (1.13 million
out of 1.33 million, or 85%), contain entries in the NPV file. But only a minority of the loans that were
not modified, fewer than one in six (about 200,000 out of over 1.25 million), have NPV file information.
Many are not considered for NPV analysis, and thus do not have data recorded for that purpose,
because the applications are denied prior to the NPV step for reasons of eligibility. This more limited
NPV data for borrowers not receiving modifications poses a possible challenge in one specific part of our
analysis. We want to assess which factors, potentially including fair housing population characteristics,
determine whether a homeowner received a trial modification or not. That requires having comparable
data for both those receiving trials and those denied them. We will return shortly to that issue as well.
The NPV file contains data on borrower, loan, and property characteristics, as well as
modification terms and details on the NPV test timing and results. A complete listing and definition of
variables in both files is available online from Treasury.26 Critically for this research, the NPV file includes
data on homeowner household income, though again not for all households in the file. For some
additional homeowners, we were able to infer incomes from the NPV data. For these households, their
housing expenses and the ratio of housing expense to income was reported, allowing us to calculate
income level directly.27
We investigated the possible impacts of two data challenges highlighted above: less than
complete data for the four fair housing variables (race, ethnicity, gender, and income) and limited NPV
file data for homeowners who did not receive loan modifications at even the trial level. Turning first to
the fair housing variables, because of the focus of our research we want to work principally with loans
that include information on all four of the fair housing variables targeted for the study. But it is
important to be satisfied that this choice, given some missing data, does not leave an inadequate set of
usable HAMP applicants nor one that represents poorly the full applicant pool. It is especially important
that there not be evidence of systematic relative under-reporting of fair housing populations with
25
The loan modification file is the source for the variable measuring total monthly housing expenses (before any
loan modification). But that total is provided only for homeowners receiving at least trial modifications, not for
NANAs. For NANAs, we computed total housing expenses by adding together reported housing expense
components (mortgage payment, property taxes, etc.). A difficulty was that some unrealistically high insurance
costs were reported as monthly expenses, when they were apparently annual payments. For certain highinsurance reports, we converted them to monthly.
26
U.S. Department of Treasury, “MHA Data File Data Dictionary v2.0,”
http://www.treasury.gov/initiatives/financialstability/results/Documents/MHA%20Data%20File%20Handbook%20
Data%20Dictionary%2011-4-2011.pdf (accessed December 11, 2011).
27
We were also able to infer housing expense to income ratios for some households beyond those for whom it
was directly provided, using income and housing expense information to calculate the ratios where they were not
directly provided.
12
unsuccessful outcomes, which could bias some of our finding. Table 2-1 indicates the frequencies of
data availability for all borrowers, people denied trial modifications (“NANAs”28), and those receiving
trials or advancing further (“all modified loans”). Leaving aside the income category for a moment, the
other population groups show reporting for absolute numbers of 400,000 cases or more and percent
reporting of from nearly one third to more than three fifths. Separate analysis (not reported here)
shows that reporting improved after December, 2009 when Treasury changed rules to mandate lenders
asking borrowers for the demographic information, although borrowers could still decline to provide it.
Table 2-1. Availability of Data for Fair Housing Characteristics.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
We then tested how well the cases with all four FHVs supplied with data represent the broader
population of those entering the HAMP system. Table 2-2 compares those cases with all borrowers,
those with loan modification file data but not NPV data (the non-NPV file column, including mostly
NANAs), and those with data from both files (NPV file column, including mostly loans with
modifications). For most variables of interest to us in the analysis that follows, the cases with all four
FHV variables supplied match very well their counterparts in which none or only part of the FHV data is
present. That gives us confidence in our four-FHV cases’ representativeness of the overall population of
applicants.
It is also important to note that for two important parts of our analysis, the obtaining of trial and
permanent modifications by differing populations, we are not dependent on this match for in order to
assure that the cases for which we have FHV information are representative.. For those analyses we use
multivariate techniques that allow us to control for many other variables directly, so that if included
cases are different from those without FHV data those differences are explicitly controlled; and we use
cases in which all FHV variables are reported.
28
Not approved and/or not accepted.
13
The significant differences in Table 2-2 do include some over-representation of whites and
under-representation of African Americans in terms of race, and under-representation of Hispanics in
terms of ethnicity, in the four-FHV data. These potentially pose a challenge when Treasury’s HAMP data
about the FHVs themselves are compared to data from outside that database and without extensive
control variables. That is the case in our analyses only for the FHV comparison of all applicants, from the
Treasury database, to the wider set of borrowers having difficulty with their mortgages, from outside
data sources. To limit any impact of missing data for that comparison, we use data on FHV
characteristics of all HAMP applicants who supply data for any given FHV characteristic, even if data is
missing for other FHVs. Thus our HAMP-applicant racial distribution measure includes all cases in which
race information is provided, for example, even if ethnic and gender information is not.
The income variable is reported in a low percentage for cases of borrowers not receiving
modifications (NANAs). That fact ties directly to the more limited data for homeowners without NPV file
data entries. Most NANAs did not have an NPV analysis and reporting, which means that income was
not recorded. Thus only the nearly 196,000 NANAs that did proceed to the stage of having an NPV
analysis, 16 percent of the NANA total, provide income information. All NPV-file NANA loans, and all
modified loans, have income information; but the other NANAs didn’t proceed far enough to record it.
The question remaining is how well the NANAs with income information represent other non-recipients
of any modification. That question applies not only to the use of the income variable as part of the fourFHV data requirement for inclusion in our analysis. The same question is raised in considering the overall
ability of the NANAs with NPV information—on income and many other important variables for our
analysis—to represent all NANAs including those without NPV file data.
We explored that question using the variables that are available for homeowners present only in
the loan modification, not the NPV, data file. We are restricted to a relatively small number of variables,
because we can only use those that are contained in the data obtained for all applicants--in order to
have an all-NANA base for comparison. In Table 2-3, we compare all NANAs, NANAs with NPV data
including income, and NANAs with both NPV data and data for all four of the fair housing variables.29
This third group is the set of NANAs we actually use as the non-modification cases for much of the
analysis.
The three sets of homeowners are quite similar in terms of the variables by which we can
compare them. Hispanics are somewhat over-represented in the 196,000 cases with NPV data but not
necessarily FHV data, but not in the 50,000 cases with both NPV and FHV data. African Americans are
somewhat under-represented in the third group. Condominium owners are over-represented in that
same group, but single family structures dominate in virtually the same percentages throughout the
three sets of cases. To the best of our ability to measure, the necessarily restricted sets of homeowners
who do not receive modifications but for which we have NPV and fair housing data are reasonably
representative of the larger set of all NANAs. Again, for much of our analysis, multivariate controls
29
Note that the third group excludes NANAs which do not meet HAMP eligibility standards. The analysis for which
we need a NANA sample with fair housing and NPV variables, to compare to trial loan modifications, does not
apply to ineligible loans. See the Methodology section of this report.
14
minimize the impact of any remaining differences. And for computing simple race, ethnicity, and gender
distributions, the FHV data are contained in the loan modification file, so that we are not limited to using
only the NANA cases with data in the NPV file.
Some observers have raised the question of whether servicers might deliberately fail to report
race and ethnicity specifically for people they turn down, to avoid showing a disadvantage in HAMP
assistance for those population groups. In addition to the protections against unrepresentativeness due
to missing data provided in our approach to analysis and testing, as just discussed, two additional factors
mitigate against this possibility creating a bias in our research. First, as will be detailed later, we find that
total applicants of color reported by race and ethnicity—combining those approved for trials and those
not—are at least in proportion to their numbers with mortgage problems, relative to white nonHispanics. Deliberately dropping a significant portion of minorities who were disapproved for trials
should have produced lower proportions for total applicants.30 Second, reporting of race and ethnicity
into the Treasury database—if it is done— is made at the time when a borrower is either beginning a
trial or turned down (a NANA case). At that stage, the servicer does not know whether the borrower will
receive a permanent modification nor, if so, whether it will be sustained. Therefore, the servicer cannot
choose to bias those two later outcomes. As we shall see, our results show that FHV households are not
disadvantaged at those stages, nor at the trial stage. 31
30
Unless we assume that people of color were applying for HAMP assistance in much greater numbers relative to
their mortgage troubles than were other people.
31
And even at the trial vs. NANA phase, servicers would have to be believed to be throwing away data they had
collected before they knew the outcome, albeit before it had entered the Treasury’s HAMP database.
15
Table 2-2. Comparison of Data Samples by Key Variables.
(Continued on next page)
16
(Continued from previous page)
(Continued on next page)
17
(Continued from previous page)
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: Shading indicates that there is no information to fill in because the variable comes from the NPV file.
18
Table 2-3. Comparison of All Loans without Modifications to Unmodified Loans with NPV Data and with
Fair Housing Data.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
19
Data for Analysis of HAMP’s Indirect and Net Impacts
The data for the analysis of HAMP’s net impacts, including indirect effects through means other
than HAMP-specific modifications, were obtained principally from three sources. Note that the data
sources cited here are those used for our prior NFMC evaluation work, from which we then drew and
focused on specific results in this study. These sources are:

LPS Applied Analytics, Inc.’s database of loans, encompassing a large percentage of all active
residential loans in the United States;

Home Mortgage Disclosure Act (HMDA) data; and

National Foreclosure Mitigation Counseling Program (NFMC) data, documenting the
experience of homeowners obtaining foreclosure prevention counseling through that large
national program.
The data are described in more extensive detail in Mayer et al. (2011), which reports the original
modeling from which this paper extends to focus on HAMP and fair housing populations; and only key
characteristics are summarized here. The data were assembled in order to evaluate the impacts of
counseling under NFMC on foreclosure outcomes. That included assembling a sample of 155,000
homeowners who received counseling and a comparison sample of similar size of homeowners outside
the counseled group.32 It is these two samples, combined into a single dataset that we use in the netimpact of HAMP research being reported here
LPS Applied Analytics, Inc., (LPS) is a commercial company that compiles home mortgage
performance data from large loan-servicing organizations. As of June 2010, LPS estimates that its
database covered nearly 70 percent of the active residential mortgages in the United States. LPS
compiles loan-level data from mortgage servicers, including nine of the ten largest servicers in the U.S.,
and tracks several aspects of loan performance for active mortgage loans.
The LPS data contain numerous characteristics of each mortgage loan, including the borrower’s
FICO score at loan origination, the original loan amount, the current interest rate of the loan, the loan
type (fixed rate, adjustable rate, option ARM), and the ZIP code of the mortgaged property. The data
also track various loan performance indicators, including when a borrower defaulted on a loan and
whether the loan has gone into foreclosure. The LPS loan performance data are updated monthly, which
permits tracking of delinquency and foreclosure status on a month-to-month basis. It is from this data
that we drew information on the foreclosure outcomes of concern in this study—the dependent
variables of our analysis of net HAMP effects for various populations—for both the NFMC sample and
the comparison sample. We also drew data for many of the mortgage characteristics of the comparison
sample from this source.
32
Some of its members may have also been counseled.
20
The Home Mortgage Disclosure Act (HMDA), enacted in 1975, requires most lending institutions
to report detailed data on mortgage application outcomes and approved loans to the Federal Financial
Institutions Examination Council. HMDA data are routinely used to determine if housing credit needs are
being met in particular neighborhoods and to identify discriminatory lending patterns. HMDA data are
released publicly on an annual basis and the public data include the fields of race, sex, and income of the
borrower; the loan amount and type; and the census tract of the mortgaged property. For this analysis,
we had access to national loan-level HMDA data from 2002 through 2008.
We used the HMDA data to link additional borrower characteristics (notably race, ethnicity, and
income) with the LPS data. The specifics of the loan matching process, based on selected data contained
in both data files, are detailed in Mayer et al. (2011). Furthermore, because census tract is reported on
the HMDA data, by combining LPS and HMDA records we were able to link additional census tract
information for the non-counseled loans. (The counseled loans already had geocoded tract identifiers.)
These census tract characteristics allowed us to control for neighborhood effects in our models.
The NFMC data are collected as part of program operations. NFMC program Grantees are
required to provide client-level data (referred to as production data), along with quarterly reports on
aggregate activity toward overall goals established under the grant award. The production data are
submitted by Grantees on an ongoing basis through an electronic submission system. Production data
consist of a record for each “counseling unit” provided by the Grantee or Subgrantee to an individual
homeowner.
The production data provide the list of homeowners who have received NFMC program
counseling in some form and, therefore, constituted the treatment group for our analysis of NFMC
program impacts in that evaluation study. The data consist of information on the counseled
homeowner, including identifying data (name, address), demographic characteristics (notably including
race and ethnicity), and household income; information on the client’s mortgage loan, including the
current servicer, loan terms, and current default status; and information on the type and amount of
foreclosure mitigation counseling received. We used production data on approximately 960,000 clients
counseled during Rounds 1 and 2 of the NFMC program in 2008 and 2009 and reported to NW America
as of January 22, 2010.
LPS matched NFMC data and LPS data in order to provide mortgage foreclosure status
information—delinquencies, foreclosures, modifications, other cures and foreclosure stops, and
redefaults—for the NFMC-counseled borrowers. They used unique servicer and loan identifiers
maintained by the company (and not available to us) to carry out the matching for us, linking to
comparable identifiers included in the NFMC data. The loans/homeowners for which a match was
founded constitute our NFMC treatment sample.
We then applied a technique called propensity scoring to identify a readily comparable
comparison sample of LPS cases from among those already matched with HMDA data. The result is a set
of about 155,000 cases each of NFMC and non-NFMC loans, which we combined and use for analysis.
Each includes outcome variables for each of our measures of mortgage results (foreclosure, cure, etc.)
21
from LPS. Each includes race, ethnicity, and income variables—the NFMC sample from the counseling
program’s records and the comparison sample from the HMDA match with LPS data—to use in isolating
HAMP’s impacts for fair housing populations and majority counterparts. And each includes information
about loans, counseling provided, and neighborhood and regional economic conditions to be used as
control variables in the research reported here.
Because LPS records are monthly, we have monthly observations about mortgage status which
can be designated as either pre- or post- initiation of HAMP in April of 2009. These are key to identifying
HAMP impacts. That same information enables us to track additional time-varying explanatory variables,
including months since key events, such as starts of foreclosure or delinquency and timing of cures
which may then redefault. These enable us to build models tracking foreclosure outcomes over time.
22
3.
Models and Methods
In describing our models and methods for analysis, we again treat separately the direct effects
and the indirect and net benefits of HAMP to fair housing populations.
Models and Methods for Analyzing HAMP’s Direct Impacts
We address direct impacts for our sequence of individual research questions about fair housing
populations: their participation in applications in relation to their share of mortgage difficulties, in
receipt of trial modifications in relation to applications, in permanent modifications compared to trials,
and in sustaining of permanent modifications in comparison to their receipt.
Applicants Compared to Homeowners in Trouble
Our first piece of analysis compares the distribution of troubled mortgages in terms of fair
housing characteristics to the distribution of people who set out to pursue HAMP loan modifications.
We conduct tabulations of the proportions of applicants in the race and ethnicity categories in the total
HAMP applicant pool, drawing from the Treasury database. We compare them to estimates for the race
and ethnicity distributions of all homeowners who are seriously delinquent (60+ days) or have entered
or completed foreclosure on their mortgages, using the results of analysis by Bocian et al. (2011). In
their research, the researchers at Center for Responsible Lending (CRL) drew on LPS and Black Box data
about individual loan status matched to HMDA information including race and ethnicity to estimate the
distribution of homeowners in mortgage trouble, in terms of those two fair housing population
characteristics. We perform Chi-Square tests to determine the statistical significance of differences
between the HAMP distribution by race and ethnicity and the corresponding all-troubled-loans
distributions.
These comparisons have two limitations (in addition to the inability to use multivariate control
methods). One is that the LPS and Black Box data inevitably themselves include HAMP modification
applicants whom we cannot distinguish. It would be a cleaner comparison if it were possible to compare
HAMP modification populations to those only of loans not applying for the program. The second is that
the CRL researchers limited themselves to loans originated in 2004-2008, whereas the Treasury data
does not allow us to match that constraint. It does not provide date of origination for loans that were
not modified under HAMP. But data from Bocian, Li, and Ernst (2010) suggest that 2004-2008
originations account for 85 to 90 percent of foreclosures completed between January 2007 and
December 2009. The CRL information thus represents a major part of the total of troubled loans
(assuming it accounts for much of the seriously delinquent loan population as well).33
33
Ideally, we would have liked to construct a multivariate model of the likelihood of a HAMP-eligible homeowner,
with a defaulted or imminently defaulting mortgage, entering the HAMP application process, as a function of its
FHV characteristics and a variety of control variables, testing for the role of the FHV characteristics if any. We have
detailed data about at least many of the homeowners who begin the HAMP application process and proceed to
various levels, in the Treasury database. The existence of records in the loan modification file of Treasury’s
23
Unfortunately, we do not have a means to conduct parallel analysis for gender, because neither
the CRL researchers nor others we are aware of provide gender estimates of this type. And for income,
we can carry out only a much rougher comparison between troubled homeowners and applicants,
which we explain in detail in the section on findings. While we are unable to compare applications to
overall need in terms of gender and can make that comparison only crudely in terms of income, we will,
as shown in succeeding sections, include gender and income in assessing the other four elements of
HAMP activity: eligibility, trial modifications, permanent modifications, and homeowners’ ability to
sustain them.
Eligibility of Applicants
The HAMP program specifies which mortgages and borrowers are eligible for the program, as
we have described in summarizing the program elements. Some of the applicants may be simply
ineligible. Therefore a next step is to test whether the fair housing population groups have differing
frequencies of eligibility for the program. Eligibility criteria could, without intent, have disproportionate
impact on fair housing populations if they focus on characteristics that apply in unusually high
proportions to people of color, women, or low-income people.
At least in theory, eligibility should be a “yes” or “no” issue in terms of impact on receiving a
trial. If an applicant is ineligible under one or more criteria, there should be complete certainty that the
applicant does not obtain a trial. While it might otherwise be desirable to include eligibility criteria in a
multivariate regression analysis of the impact of those and other factors in the likelihood of receiving a
trial, we cannot do so for two reasons. First, for measures of some eligibility criteria, data are not
provided at all in the Treasury data for those who do receive trials, because logically only those denied
trials get assigned “denial reason codes”—from which important elements of the eligibility information
derive. Second, we cannot properly methodologically use regression methods under these “yes or no”
conditions, which violate assumptions about statistical properties of regression models. Therefore we
conducted a separate analysis of eligibility issues alone, assessing whether our population groups of
interest were more frequently rejected because of eligibility rules than were others.
The eligibility analysis is a two-step one. The Treasury data provides reasons for denial of trials, a
subset of which we determined are elements of eligibility. The reasons that relate to eligibility are
shown on the left side of Table 3-1. We compare the share of borrowers in our target populations
rejected on eligibility grounds with the share of total borrowers --receiving trials and not—who are
database signals the entry of borrowers into the HAMP process, and the loan modification and in some cases the
NPV files provide the specifics. As indicated above, these include data for the FHV characteristics of homeowners.
But we do not have a source of comparable specifics for other households potentially in mortgage trouble and
eligible to participate in HAMP. It might have been possible to obtain access to proprietary data including some of
the key variables determining potential for HAMP assistance, from sources including LPS and Black Box, and even
to match them to HMDA to determine some FHV characteristics, although these tasks were beyond the scope and
resources of this study. But crucially, as yet the Treasury has not released loan identifiers that would have made it
possible to accurately match individual HAMP loans to such external records of loan specifics and status. Therefore
there is no reliable way to separate HAMP applicants from other homeowners in the broader data, in order to
perform an analysis of the ideal type.
24
ineligible, using all ineligibility criteria taken together. Then as a second step, insofar as there are any fair
housing populations with greater shares of ineligibility-driven rejections than their counterparts, we
examine which of the eligibility criteria affect them more than do others.
Table 3-1. Trial Denial Reason Codes and Their Relationship to Eligibility.
Source: Treasury (list of denial reason codes) and Mortgage Bankers Association (what qualifies as
eligibility criteria).
Recipients of Trial Modifications and Permanent Modifications
The next two central pieces of our analysis are assessing the role of fair housing population
characteristics in obtaining HAMP trial mortgage modifications and permanent mortgage modifications.
The models and methods for analyzing receipt of trial modifications and permanent modifications by fair
housing populations are very similar to each other. We employ two separate but parallel models—one
for each type of modification. In each case, we model the percentage of homeowners receiving and
accepting the respective modification level (trial, permanent modifications) as the dependent variable,
as a function of:







homeowner characteristics, notably including the fair housing characteristics of race,
ethnicity, gender, and income;
homeowner financial condition including housing expense, credit score, and debt-to-income
ratios;
region in which the property is located34;
basic property characteristics (e.g. single family vs 2-4 units);
terms and characteristics of the mortgage in existence before any possible modification;
characteristics of the loan at origination;
terms of the modification; and
34
Treasury data also includes location by state, but we did not use it because the field’s frequent (about half)
missing data would have sharply limited our data pool.
25

results and timing of the net present value analysis for a possible modification.
Trial modifications are modeled for all homeowners who enter the HAMP application process,
who then either obtain a trial or fail to do so. The trials modeling includes those borrowers who may go
further into the HAMP process, obtaining permanent modifications, having trials or permanent
modifications cancelled or disqualified, or paying off loans. Permanent modifications are modeled for all
homeowners who obtained trials, distinguishing those who made those trials permanent from those
who did not.
The complete list of variables for the two models is provided in Table 3-2. The variables other
than homeowner characteristics provide information that lenders/investors employ to assess the value
of the loan in modified and non-modified form, including the ability of the borrower to repay each and
the risk of a foreclosure under each alternative, and information that homeowners employ to assess the
value of accepting a modification if offered, including their financial position under each alternative.
Controlling for those variables allows us to isolate the impacts, if any, of race, ethnicity, gender, and
income. Note that for many of the control variables, we do not have a clear expectation, a priori,
regarding what direction their effects may be on the likelihood of modifications being approved and
accepted. For many variables, for example, higher (lower) values may encourage the lender/investor to
take the risks that a modified mortgage will be repaid and thus encourage them to agree to make the
modification, while the same higher (lower) value lowers the worth of the modification to the borrower
and discourages acceptance.
The vast majority of the control variables are the same in trial and permanent modification
models. The logic is that these same variables are assessed by lenders and borrowers considering the
value of modifications at the trial stage and then again at the permanent modification stage.
A few variables are unique to the permanent modifications analysis (see again Table 3-2). One is
a measure of whether the trial modification is at least three months old, important because three
months is the minimum trial period for successful payments under HAMP rules. The others are data
unavailable for NANAs and thus unusable in the trials analysis, but available for both trials and
permanent modification recipients and thus usable in the analysis of permanent modifications. Those
two are condition of residence and reasons why homeowners are in hardship positions (loss of income,
medical, etc.). A few variables are included in the trials analysis and dropped in the permanent
modifications analysis. These include a categorical variable representing missing data in one variable in
which there is no missing data among modification recipients (only among NANAs), a categorical
variable for ratios of housing expense to income below the 0.31 program minimum,35 and two variables
regarding original mortgage terms that become superfluous in analysis of permanent modifications
when immediately pre-modification mortgage terms are available for all modifications.
In both the trials and permanent modifications analyses, we first eliminate the borrowers who
are or become ineligible, for the reasons explained above in the eligibility analysis description. Then we
35
An artifact of our need to estimate this variable by adding housing cost components together for NANAs in the
trials analysis.
26
can use multivariate regression techniques applied to data for the sets of eligible applicants, in the trials
analysis and of eligible trial recipients, in the permanent modifications analysis.
Table 3-2. Variables for Models of Trial and Permanent Modifications.
(Continued on next page)
27
(Continued from previous page)
28
(Continued on next page)
(Continued from previous page)
29
Source: Authors’ selections of variables from the U.S. Treasury HAMP Public Database.
Note: As noted later in the text, some of the variables listed differ between the two analyses in how they
are calculated, even though they are called the same thing.
For each of trials and permanent modifications, we use a binomial LOGIT regression model for
estimating the impacts of the fair housing and control variables on the percentage of (respectively)
applicants receiving trial modifications and trials recipients obtaining permanent modifications. This is a
standard technique for models with dichotomous dependent variables-- here the approval or denial of a
HAMP modification-- as a function of multiple determining factors.
The primary difference between the estimation procedures employed in the two models is the
use of weighting in the trial modifications model and not in the permanent modifications model. As we
30
have detailed earlier, people who were denied trials—NANAs—are under-represented in the usable
data for regression analysis of trial modifications, because many of them were eliminated from
consideration without receiving an NPV analysis and thus without collecting much of the data needed
for that analysis. People receiving trials, on the other hand, are fully represented in the NPV file. Had we
not given extra weight to the NANA cases, the LOGIT analysis for trials would be weighted too heavily
toward explaining the cases that did result in a trial relative to those that did not. We instead attach a
weight of approximately 6.7 to each NANA included in the analysis, in order to balance the complete
trials data and the limited NANA data. Trials recipients and NANAs are thus each given weight in
proportion to their actual numbers in the overall applicant pool.
We do nonetheless estimate our trial modifications model both with and without weighting, to
be certain that our findings are not highly sensitive to the weighting assumption. As reported in our
findings below, eliminating the weights has no major impacts on our results regarding outcomes for fair
housing populations relative to other HAMP applicants.
In the analysis of recipients of permanent modifications, all trials recipients being considered –
both those ultimately rejected for permanent modifications and those approved—have roughly equally
complete data. Both those receiving permanent modifications and those being stopped with only trials
will have been assessed for net present value of loans and have NPV file data. Weighting is not needed,
because NANAs, with their limited NPV data, are not involved in the analysis. The cases involved in the
permanent modifications analysis are only those that received a trial modification previously.
Sustaining Permanent Modifications
Fair housing populations may have differential success in sustaining permanent modifications
they receive, compared to majority homeowner groups. Our study did not include sufficient resources
for a full scale analysis of this possibility. We include simple tabulations and Chi-Square testing of the
comparative experience in sustaining permanent modifications, rather than having them disqualified
because modified payments were not met.36 These provide a suggestive first take on the likelihood that
fair housing populations face additional challenges in holding onto their homes after permanent
modifications. Multivariate analysis parallel to that used for trial and permanent modifications would be
useful to further pin down that possibility.
Methods and Models for Analyzing HAMP’s Indirect and Net Impacts
To analyze the net impacts of HAMP, including indirect effects along with HAMP modifications,
for differing population groups, we assess overall mortgage foreclosure outcomes before and after the
initiation of HAMP in April, 2009. We focus on the interaction between HAMP’s creation and operation
and the fair housing characteristics of homeowners, in determining various mortgage outcomes. We
examine the impact of a combination of race, ethnicity, or income with HAMP’s initiation and continuing
36
There are also conditions of modification which, if violated, would cause a disqualification—for example, taking
on another loan which makes the modified loan subordinate. But it is our understanding that the primary reason
for being disqualified is missing payments.
31
work, using multivariate models of foreclosure outcomes as a function of these interactions of HAMP
and fair housing characteristics and a series of other control variables for borrower, loan, and locational
characteristics.
We do this by first creating categorical variables indicating on which side of the April 1, 2009
date events relevant to foreclosure outcomes occurred. We create variables that are the product of
those categorical variables and each of variables reflecting race, ethnicity, and gender (categorical) and
income (continuous). We test the impact of those products on each of our series of foreclosure
outcomes—size of loan modification payment reduction, foreclosure stops, foreclosure cures through
modification and otherwise, and redefaults of modifications and of other cures. Each outcome is
represented by a separate multivariate regression model.
The exact structure of the interaction variables differs across the models involving differing
foreclosure outcomes. Specifically, the HAMP indicator variable going into the product interactions
takes the following forms for models analyzing the respective outcomes:




For size of modification payment reduction (a cross-sectional analysis), the HAMP indicator
signals whether the modification that occurs took place before or after the start of HAMP. If
HAMP is to influence modification size, it has to be in place before the modification is
approved; and we need to specify that in the model.
For foreclosure stops, the HAMP indicator signals whether a given period being observed in
terms of a foreclosure stop taking place (or not), in a time-series analysis, is before April,
2009.
For foreclosure cures, the HAMP indicator again signals when a given period observed is in
relation to April, 2009.
For redefaults, our principal concern is whether HAMP was underway by the time the
original curing of the original default occurred. The HAMP categorical variable represents
whether the cure (modification or non- modification cure in the two redefault models
respectively) took place before or after the start of HAMP.
In the analysis of each outcome, the product of the corresponding HAMP variable and each of
the fair housing variables is tested for its impact on foreclosure prevention results. In the case of the size
of modification, the multivariate model is an ordinary least squares regression, and the test for the
statistical significance of the interaction is a standard one.
For the other outcomes, which are dichotomous time series measures for events occurring or
not in each of multiple time periods, the appropriate multivariate analysis is a logistic regression
(LOGIT). For the interactions estimated for these logistic regressions, we employ a significance-testing
method recommended by Norton, Wang, and Ai (2004) that is required due to the non-linear
relationship in LOGIT models between explanatory variables and the dichotomous outcome used as a
dependent variable.
The results of these interaction models indicate whether HAMP has differential effects for
particular borrowers. For example, if we find a positive and significant interaction effect of income and
32
HAMP on the probability of curing a seriously delinquent loan with a modification, it means that HAMP
has a greater impact for higher income homeowners, when compared to lower income homeowners. A
negative and significant interaction effect would mean that lower-income homeowners, on average,
benefit more from HAMP. If there is no significant interaction income effect, it means that HAMP has
the same impact regardless of an owner’s income.
An important caveat regarding this analytic approach is the assumption that we can well
approximate the impact of HAMP by representing its start date and that date’s relationship to
foreclosure prevention activity. Obviously HAMP’s impact does not neatly turn from zero to full throttle
on the day it opens, so that there is inherently some approximation in signaling its operation in the way
we have. Perhaps more importantly, there may be other trends underway both before and after HAMP’s
initiation, which lead to improved foreclosure prevention outcomes. It is possible that mortgage
servicers and counselors may have improved their performance with time even without HAMP. Certainly
we know that, since the beginning of the foreclosure crisis, the market participants have instituted bestpractices over time as they learned strategies and methods to increase their effectiveness. We know as
well that counseling organizations and mortgage servicers have increased their capacity to deal with the
rising volume of troubled mortgages. Some of these changes might have been a result of the
introduction of HAMP, while others might be coincidental in timing. Insofar as they were separate, and
trend toward increased foreclosure prevention success, HAMP’s impact could be overstated in the
analysis.
We have in the various foreclosure outcome models reduced the extent to which such effects
bias our tests of HAMP’s benefits to fair housing and counterpart populations. Each model except the
modification amount contains a “counter” variable specifying the month in which each observation
takes place and increasing each month.37 A significant piece of any trends over time not dependent on
HAMP’s initiation should be captured by the counter variable, separate from any distinct change
observed specifically in April, 2009. In addition, it is important to note that the HAMP variables we are
focusing on are those in interaction with fair housing variables of race, ethnicity, gender, and income—
not HAMP alone. There is no a priori reason to anticipate that the impacts of possible trends discussed
in the previous paragraph are differentially distributed with respect to fair housing populations and their
counterparts. If for example servicers were generally improving their approach to foreclosure
prevention over time, but they were not making greater improvements for Hispanics than for nonHispanics, then any servicing trend not captured by the counter would nonetheless not bias our
estimate of the impact of being Hispanic on the benefit obtained from HAMP. The same would be true
for other fair housing populations.
37
In the foreclosure stops model, the counter counts the period since foreclosures started. In the cures model, it
counts the period since foreclosure or serious delinquency began. In the redefaults models, it counts the period
since cure occurred. Each of the counters captures a combination of external trends and timing for individual
loans.
33
4.
HAMP Applicants Compared to Homeowners in Trouble
We are able to make tabular comparisons between the racial and ethnic composition of all
HAMP applicants on the one hand and that of all homeowners who are seriously delinquent (60+ days),
are in the foreclosure process, or have already completed a foreclosure. As described in section 2, we
compare applicants’ race and ethnicity from Treasury database data for all applicants to the
race/ethnicity estimates for all troubled borrowers from the research by Bocian et al. (2011). The results
are in Table 4-1.
For those racial and ethnic groups for whom we have data, people of color fare at least as well
in becoming HAMP applicants as do non-Hispanic whites. As the table indicates, each group of racial and
ethnic minorities forms at least as high a percentage of entry applicants to HAMP as it does a
percentage of loans in trouble—whether seriously delinquent loans and those with foreclosures begun,
those with completed foreclosures, or a combination of the two.38 Hispanics have higher percentages of
applicants than homeowners in each type of difficulty and in total. African Americans have higher
percentages of applicants by margins of two percentage points and more, depending on the measure.
Asian applicant figures very closely match the percentages of Asian homeowners among those in
default. The total of the smaller racial groups among applicants exceeds the “others” category’s share of
all troubled owners. At least on race and ethnicity dimensions, HAMP is doing well with initially
attracting or making accessible the program to fair housing populations.39
We do not have a source of data for the gender of all homeowners in difficulty, to compare to
the Treasury data regarding gender of HAMP applicants. HMDA offers a source for the future, if work
similar to that of Bocian et al. (2011) were to be conducted including gender measurement.
Unfortunately we do not have a way to make a high-precision comparison between income
levels of HAMP applicants and income levels of people with seriously delinquent mortgages or
38
The sole exception is that, for completed foreclosures only, Asians have a slightly higher percentage than they do
applicants.
39
The Treasury Department performed its own analysis of race and ethnicity (Making Home Affordable 2011b)
using data only from December, 2009 onward after requesting these items of borrowers became mandatory
(though they could still decline), whereas we use all borrowers reporting from the start of the program. Their
results are very similar to ours for the percentages of all applicants (recipients of trials and not) who are people of
color. As a result, a comparison of Bocian’s numbers on troubled borrowers to Treasury tabulations produces
virtually identical results to those arising from our computation. When we use Treasury computations, African
Americans and the small racial groups have a higher share of applicants than they do of all troubled mortgages or
the categories of troubled mortgages, and Asians essentially have the same share, as we found. Hispanics have a
higher share as well, compared to all troubled loans, as we found. The one difference is that Hispanics’ share of all
applicants is slightly below that of completed foreclosures (without including seriously delinquencies) alone, in the
Treasury analysis, but not for all trouble loans combined, whereas our computation showed Hispanic share of
applicants higher than either completed foreclosures or the total of seriously delinquent and foreclosures begun
and completed. In sum, using only the numbers from December, 2009 onward as Treasury does would not
substantively change the results we found using data for all homeowners reporting race and ethnicity form the
start of the program as we did.
34
mortgages having entered or completed foreclosure. CRL’s research in Bocian et al. (2011), drawing on
HMDA for income information, necessarily is based on incomes at the time mortgages were originated,
as HMDA reports. The income data we have from Treasury’s database are for homeowners at the time
they apply for HAMP assistance. The latter will frequently reflect significant income declines since the
time of mortgage origination, in the context of the broad recession and more specifically because
people seeking HAMP’s reduced mortgage payments are more likely to be borrowers with reduced
incomes.
Table 4-1. Comparison of Homeowners with Troubled Mortgages (Foreclosed or Seriously Delinquent) to
Applicants for HAMP Assistance, by Race and Ethnicity.
Source: For the right column, authors’ computations, based on U.S. Treasury HAMP Public Database; for
the other columns, Bocian et al. (2011).40
We can make very rough comparisons for a more limited question: where in the spectrum of
troubled homeowners’ incomes at loan origination do HAMP applicants’ incomes at HAMP application
lie? We might find that HAMP applicants are at least far down the income scale of what people, now
delinquent, in foreclosure process, or foreclosed, were earning when they first acquired their premodification mortgages (or that they are not).
Even that approach requires some significant approximations. The CRL work computes numbers
of homeowners with loan trouble by income categories defined in each MSA by a household’s income as
a percentage of MSA median income, with four income categories. Our income computations divide
40
Figures from Bocian et al. are based on originations from between 2004 and 2008 only, whereas the right
column is based on all origination years. Bocian et al. consider someone who indicated non-Hispanic, AfricanAmerican, and another race (e.g., white or Asian) to be non-Hispanic African American, though our analysis of
HAMP data puts those few borrowers into the bottom row. The “other” categories include “American Indian or
Alaska Native” and “Native Hawaiian or Pacific Islander.” In this analysis, if a borrower does not indicate any
ethnicity (i.e., neither “Hispanic” nor “non-Hispanic”) they are treated as non-Hispanic. Elsewhere in the report, we
include as Whites, African Americans, and Asians HAMP participants who indicate “Hispanic” as their ethnicity,
though here, as specified in the table, we refer to non-Hispanic Whites, African Americans, and Asians, in order to
match Bocian’s categories. Because of the differences in treatment of the smaller groups, we conducted a separate
analysis in which we excluded the group in the bottom row above, recomputed the distributions for the other
groups, and again found that HAMP applicants have a similar composition to homeowners in trouble.
35
applicant households into quintiles, based on their absolute income level and combine all households
nationally. CRL computes median incomes for each of its categories and we compute mean income
within each category. We have to make approximations and structure broad comparisons to bridge
these differences. Still we think the findings are illustrative of HAMP’s reach.
Table 4-2. Incomes of HAMP Applicants and of Homeowners with Seriously Delinquent and Foreclosed
Mortgages.
Sources: For the top section, authors' computations from U.S. Treasury HAMP database; for the bottom
section, Bocian et al. (2011).41
Table 4-2 presents the income distribution for HAMP applicants and then for borrowers with
loan difficulties. The latter are separated into serious delinquencies or foreclosures in process,
completed foreclosures, and totals of the two. Because the percentages differ little among the three
measures of loan difficulty, we will focus on the total for all troubled loans. Consider the lowest quintile
(20 percent) of applicants. Mean annual income within the quintile is $22,200. Compare that to the lowincome category for people with troubled loans. It contains only 6.6 percent of all troubled borrowers
and has a median annual income within the category of $26,000. The incomes at entry of applicants to
HAMP are thus sharply more concentrated among incomes averaging $22,200 (and $26,000) than are
the incomes of troubled borrowers at the time of loan origination.42 Very roughly speaking, HAMP
41
In the bottom section of this table, “low” income means less than 50% of area median income; “moderate”
greater than or equal to 50% and less than 80%; “middle” greater than or equal to 80% and less than 120%; and
“higher” greater than or equal to 120%.
42
Two of the approximations of import here are the following. First, the $26,000 figure is a median, not a mean,
within the low-income group. Within the inherently constrained low-income category, that should not make a
36
appears at a first approximation to be drawing substantially from people in the lower reaches of the
now-troubled-homeowner income spectrum, including those who started with lower incomes and no
doubt others whose incomes have dropped since origin (in ways we cannot directly observe).
Consider also the bottom two quintiles of applicants combined, comprising 40 percent of HAMP
entrants. Their average income is $28,650. Compare that to the low-income and moderate-income
categories of all troubled borrowers, which contain 6.6 percent and 20.8 percent of troubled borrowers
respectively. Their (weighted) average income43 is, at $37,400, substantially higher than the average for
the lowest 40 percent of HAMP applicants, yet they total only 27.4 percent of borrowers. Again,
applicants are more concentrated among the lower range of incomes of homeowners in difficulty, when
the latter incomes are those at origination.
Thus our crude first approximation suggests—and we underline how rough this suggestion is—
that the HAMP program is reaching well down the income spectrum, in terms of the current income of
applicants relative to the spectrum of loan-origination incomes of now-troubled borrowers. Obviously
more research would be valuable in this area, insofar as better information on current incomes of all
troubled borrowers could be obtained.
large difference. Second, some people in CRL’s low-income category may have incomes higher than the lowincome quintile because they are ranked as percentages of income in higher-income MSAs. Conversely, some in
CRL’s higher income categories as a result of residing in low-income MSAs may have low incomes in absolute
terms.
43
The weighted average of two medians.
37
5.
Fair Housing Characteristics and HAMP Trial Loan
Modifications
The HAMP program provides for consideration of applications initially for what are expected to
be three-month trial loan modifications. Those trials are to be provided on the terms on which
permanent modifications would also subsequently be offered, subject to possible changes in factors
such as borrower income during the period of the trial modifications and, especially in the early stages
of the program, subject to verification of borrower financial information that did not necessarily take
place before the trial was considered.44 At least in the ideal case, borrowers would after three months of
making timely and full payments in their modified amount, receive permanent modifications. If they fail
to make the payments, or their circumstances (e.g., income) change in negative ways in terms of
investor assessments, their trials may be disqualified (permanently not allowed), cancelled (with the
opportunity to return if circumstances again change), or extended for further consideration.45
In order to receive trial modifications, borrowers need first to meet program eligibility
requirements. Then their modifications must be approved by servicers as being in the net economic
interest of lender/investors, and be accepted by the borrowers. The significant eligibility criteria are
comprised of the following items:

The loan is a first lien mortgage which originated on or before January 1, 2009;

The loan is owned by Fannie Mae or Freddie Mac, or it is administered by a servicer
participating in the Making Home Affordable program and the servicer is granted permission
by the owner of the loan to modify it;

The property which secures the mortgage is owner-occupied and contains four units or less;

The mortgage is in foreclosure, the borrower is delinquent, or default is reasonably
foreseeable;

The amount of unpaid principal is less than a cap for a given number of units (e.g., $729,750
for a one-unit home);

The loan has not been previously disqualified during a previous trial or permanent HAMP
modification; and

The borrower’s monthly mortgage payment is greater than 31% of his or her monthly
income.
The economic interest of mortgage lenders/investors is assessed in terms of a substantial set of
financial characteristics, enabling investors to take into account risks and likelihoods of repayment with
44
The program was changed to require income documentation before granting trial modifications.
In reality, servicers often took more than three months to consider permanent modifications for borrowers in
trials, and some borrowers required additional time to submit full documentation.
45
38
and without the modification, the losses associated with foreclosure, and other elements of a
comparison of net present value of returns before and after a potential modification.
Experience of Fair Housing Population Groups in Trial Modifications: Tabulations
The Treasury HAMP database provides information on the status of all applications made to the
program. In particular in terms of obtaining trial modifications, it indicates whether applications have
been rejected short of obtaining a trial—whether denied by lenders or rejected by borrowers. These
cases are here referred to as NANAs, reflecting their being either “not approved” or “not accepted.” The
database also indicates whether a trial loan modification has been obtained by a borrower. It may be
active, received but thereafter disqualified or cancelled, or advanced to a permanent modification (with
various outcomes following that). Note that applications that are currently under review for trial
modifications at the time we download the Treasury database are simply not shown in the data and will
appear at a later date when a decision is reached.
This piece of our analysis focuses on the comparison between NANAs—applications for trials
that have not been successful--and all applications that have reached trial or beyond, including current
trials, trials obtained but then cancelled or disqualified, and those moved forward to permanent. Table
5-1 presents the outcomes for our key population groups, by race, ethnicity, gender, and income
quintile, in terms of simple tabulations. It does not yet control for eligibility, financial characteristics of
the borrowers, mortgage characteristics before and after potential modification, housing market prices,
and the relations among these items. Primary concentration is on columns 1 and 2, which report total
rejections of trials and total recipients of them.
The percentage figures indicate the share of each population group among the borrowers
achieving a given outcome. Population groups that have higher percentage representation among those
who received trials than among those who did not (NANAs) are doing better than the overall applicant
pool in obtaining trial modifications, and those with a lower percentage are doing worse.
The table shows that, on the basis of simple tabulations, only people in the bottom quintile of
incomes46 and in racial categories Asian, Native Hawaiian and Pacific Islander (NHPI), and other (two or
more races) are obtaining trials less frequently than the rest of the population. Only the differences for
low income people and NHPIs are more than a small fraction of a percentage point. However the
differences are all statistically significant at the 1 percent level (Chi-square tests are presented in
Appendix Table 1). The data suggest that, in terms of obtaining trials once they have entered the
application process, differences among the key population groups may be small. Separate analysis of the
role of program eligibility, as well as multivariate analysis to control for differences in other factors than
population group, is required to test these rough initial estimates.
46
The top income quintile also has a lower percentage of trials but is not a population group of interest for this
study. Not shown is that the bottom two quintiles together have a lower percentage of trials than the top 3
together, extending the range of the possible disadvantage to people of lower incomes.
39
Table 5-1. Fair Housing Populations in NANAs and Modified Loans.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: Column 2 includes all loans which advanced at least as far as receiving a trial modification, some of which made it to the permanent
modification stage.
40
Trial Modifications and Eligibility
The tabulations in the previous section consider all applicants for trial modifications for which
the Treasury provides data on the population categories of interest and the outcomes of applications,
while some of the applicants are in fact ineligible. Eligibility criteria could disproportionately affect fair
housing populations, depending on the criteria and characteristics of those homeowners and their
mortgages.
As indicated in discussing methods, we use a two-step eligibility analysis. We compare the share
of borrowers in our target populations rejected on eligibility grounds with the share of total borrowers -receiving trials and not—who are ineligible, using all eligibility criteria taken together (see criteria in
Table 3-1). For fair housing populations with greater shares of ineligibility-based denials than their
counterparts, we see which of the eligibility criteria most affect them.
Table 5-2. Analysis of Eligibility for Trial Modifications by Fair Housing Population.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
* The difference in eligibility rates between people with income data and people with gender, ethnicity,
and race data is due to the fact that relatively few NANAs have income data (a larger percentage of
NANAs have gender, ethnicity, and race data).
41
Note: The over-representation of Native Hawaiians and Pacific Islanders among ineligibles was
significant in a chi-squared test (p-value = 0.000). The other tests are also significant but indicate underrepresentation of ineligibles among fair housing populations.
The results of the first step tabulations are presented in Table 5-2. For most of the fair housing
population groups, rejection due to ineligibility is less rather than more common than for the general
population. Only Native Hawaiian and Pacific Islanders have a higher likelihood of ineligibility than do
whites, among racial population groups. And Hispanics, women, and low-income borrowers (in either
the bottom or bottom two quintiles), have lower ineligibility than their counterparts.47
Therefore for NHPIs alone, we tested in which eligibility denial reason codes they are
concentrated. They are over-represented in 3 reason codes compared to whites (see Table 5-3). Those
three reason codes are (1) date of origination, maximum principal balance, and whether the mortgage
to be modified is a first lien (referred to in Table 5-3 as “ineligible mortgage”), (3) whether the property
is owner-occupied, and (5) whether the investor is a HAMP participant. The first two of these are basic
elements of program definition. The last is more amenable to policy revision, in that investor
participation could be made mandatory. But the difference in percentage ineligibles is only 1.1 percent
between NHPIs and whites, involving a maximum of 112 borrowers becoming eligible, only some of
whom would ultimately receive trials. Ineligibility rules are thus having very limited impact on chances
for any fair housing population to pursue trial modifications.
Table 5-3. Comparison between Whites and Native Hawaiians and Pacific Islanders (NHPIs) in Effects of
Trial Modification Eligibility Standards.
47
Note that the percentage of ineligibles is sharply lower in every income category than it is for the other
population categories. The reason is that we are measuring ineligibility as a percentage of all applicants--NANAs
and trial recipients-- together, but ineligibility denial codes to indicate ineligibility are provided only for NANAs. But
income itself is far less often reported for NANAs than is gender, race, and ethnicity. What is important is that
42
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Multivariate Analysis of Trial Modifications
For borrowers who are eligible for the HAMP program, we need to separate out the impact, if
any, of membership in the various fair housing population groups on likelihood of actually receiving a
trial modification, controlling for other determinants of modification approval. Therefore the next key
piece of analysis is a LOGIT regression of probability that a trial modification is granted to a borrower,
given the borrower’s entry into the application process, his/her fair housing characteristics, and control
variables for financial condition and mortgage risk, housing value, mortgage type, and nature of
modification offered if any, as detailed in section 2 and section 3.
Ineligible borrowers are first eliminated from the analysis, given their treatment in the previous
section. They are filtered out using the appropriate denial reason codes for the NANAs, for which the
codes are available. Borrowers who actually received trial modifications are assumed to have been
eligible.48 The Treasury system logically does not provide reasons for denial for trial modifications,
because they were by definition not denied.
A weighted regression is our principal analytic focus, as explained in our earlier models
discussion (section 3). We weight, by 7.8 times, the observations of people not receiving trials. The
weights offset the lower rate of inclusion of NANAs than of trial recipients in the Treasury data we
employ, which, as discussed earlier, is due to lack of data availability for many borrowers among NANAs
who received no reported net present value analysis and no entries in that data file.
The regression we estimate performs extremely effectively in modeling actual approvals of trial
modifications from among eligible applicants. Nearly all of the included explanatory variables have
statistically significant coefficients at the 1 percent level or better. Within the cases used to estimate the
regression (and including weighting), the actual outcomes are that 54 percent are granted modifications
and 46 percent denied. As Table 5-4 records, the model as estimated is able to predict correctly 91.4
percent of the denials of trial modifications as denials, and 94.7 percent of the approvals as approvals,
using a threshold level of 50 percent probability of approval to make the predictions. The regression
explains high levels of variation in outcomes, by both of the reported pseudo R-squared standards. The
full regressions, both weighted and unweighted versions, are attached in Appendix Table 2.
Key results for the fair housing populations are in Table 5-5. We focus on the weighted
regression, especially the odds ratio column. In a LOGIT regression, odds ratios of greater than 1.0 mean
that a one unit increase in the variable (here, except for the case of income, a dummy variable
representing membership in a population group) increases the odds of achieving the desired outcome—
here a trial modification—compared to that for cases without that unit increase (here the majority
population). A majority of our population groups of concern have at least as high odds of obtaining trial
48
However, we have found several exceptions where several variables indicate, in about 67,000 cases, that
borrowers were given modifications even though some variables appear to indicate that the borrower should have
been ineligible. The largest portion of these, by far, is borrowers who appear to be current on their loans and
whom servicers did not determine were at risk of imminent default.
43
modifications given they entered the application process as do their counterpart traditional majority
populations, after controlling for a host of other variables. Among racial groups, American Indian and
Alaska Natives (AIANs) have a slightly lower odds ratio for obtaining trials at 92 percent of the odds for
whites, and NHPIs have a larger disadvantage at 78 percent of the odds for whites. But African
Americans and Asians have significantly higher odds of success in gaining trials than do whites and
people of multiple races have statistically the same odds as whites.
Table 5-4. Predictive Accuracy and Summary Statistics for Trial Modification Multivariate Regression
Model.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Hispanics have modestly lower odds of obtaining trial modifications than do non-Hispanics (93
percent as high). Women have higher odds than do men, but the difference is not statistically significant.
And lower income people have substantially higher odds of gaining trials than do higher income people.
Each $100 in additional monthly income lowers the relative odds of a trial modification by 12 percent.49
Table 5-5. Trial Modification Regression Results for Fair Housing Populations.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
49
That is, the income variable is measured as one unit equals $100/month. The odds ratio of 0.88 indicates an
impact of one unit of income on odds of 1.00 - 0.88 = 0.12 or 12 percent.
44
These regression results, controlling for a large number of non-fair-housing characteristics, differ
in some significant ways from the simple tabulations of trial outcomes for the various population
groups. Other things being equalized with the regression controls, Asians obtain better outcomes than
whites rather than worse as suggested by the tabulations. But AIANs move to modestly worse
outcomes, while NHPIs continue to show a consistent and substantial disadvantage. Hispanics show
disadvantage in trials in the weighted (but not unweighted) regression, whereas they did not in simple
tabulations; but the level of disadvantage is modest. People with lower incomes, whose success
obtaining trials showed as worse than higher income borrowers in our tabulations, turn out to succeed
more frequently once other characteristics are controlled for.
Note that the results differ only slightly between the weighted and unweighted versions of our
trial modification model. The weighted model is, we believe, to be preferred in properly correcting for
the exclusion of many NANAs from the analysis because of the lack of net present value file data for
them. But the choice between weighted and unweighted models makes only limited difference in the
nature of outcomes for target populations. The only two differences in the unweighted outcomes are
both of one type. The modest disadvantages of AIANs and Hispanics showing in the weighted analysis
keep their negative coefficients but become not statistically significantly different from zero in the
unweighted analysis. Use of the weighted figures thus also insures that we do not understate any
disadvantages to these groups of people of color.
Table 5-6. Percentages of Eligible Applicants Receiving Trial Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
45
Odds ratios are somewhat difficult to interpret in commonplace terms. We can restate the
impacts of population group on trial modifications in terms of ordinary probabilities of obtaining such a
modification given an eligible applicant. We assign mean and mode values to other regression variables
and compute probabilities for target populations and their reference counterparts. Table 5-6 reports
these probabilities. We concentrate first on the three population groups at a disadvantage according to
the odds ratios results in the weighted regression: AIANs, NHPIs, and Hispanics. The difference between
AIANs and whites is two percentage points, as is the difference between Hispanics and non-Hispanics.
The largest difference is for Native Hawaiians and Pacific Islanders, with a disadvantage of six
percentage points in obtaining trials for eligible applicants compared to whites.
The one truly large difference in the table is the higher probability of receiving a trial
modification for people in the bottom income quintile of eligible applicants, compared to those in other
quintiles combined. The advantage is between 98 percent probability of a trial modification being
obtained for those in the lowest fifth of incomes and 46 percent for the rest of the income spectrum.
These figures are themselves difficult to interpret alone, although they clearly indicate that lower
income people are not being disadvantaged in the program at the trials stage. The regression analysis,
by its nature, estimates the difference income makes holding all other variables in the analysis fixed—
and we assume they are at mean values as if people with low incomes nonetheless had houses of
average value, DTIs of average size, etc. In fact, there are likely to be differences in many of those other
variables in correspondence with income differences, with impacts on typical success in obtaining
modifications for different income groups as we show in the analysis of non-FHV variables’ impacts in
the next section.
Other Characteristics of Fair Housing Populations Affecting Trial Outcomes
The regressions discussed above test the direct impact of belonging to a population group, by
race, ethnicity, gender, or income, on the likelihood of obtaining trial modifications. However there is a
less direct way in which population group can affect HAMP trials as well, insofar as the fair housing
populations systematically differ in other characteristics from majority populations. Suppose, as one
hypothetical example, that Hispanics commonly live in metro areas in which house prices have suffered
especially deep drops and as a result they have very high loan-to-market-value ratios as a group,
compared to non-Hispanics. Suppose investors are especially reluctant to provide HAMP trial
modifications to people with high current LTVs. Then Hispanics could be disadvantaged in obtaining
trials if investor decisions are significantly shaped by LTVs. No bias against Hispanics as a group would be
required to produce such an effect—just decision-making formulae that strongly weighted LTV. If one
found LTV to be having such an impact, it would be worthwhile to test its effect on borrowers being able
to sustain HAMP modifications, to see if its use as a decision filter were justified by its value as a
predictor or redefault, despite its negative effect on Hispanic outcomes or whether it ought to be deemphasized in the trials approval process.
It makes sense to examine such possibilities for the population groups that, in simple
tabulations, trail in proportions of approved trials. (For fair housing populations that actually did better
than their reference counterparts, we would simply be explaining why they didn’t perform even more
46
highly in obtaining trials). The goal is to identify which variables contribute the most to such indirectly
caused differences in likelihood of obtaining trials, for each population group that obtained relatively
few.
To do this analysis for any given variable such as LTV, we evaluate our model of trials at the
average LTV value for an FHV group such as Hispanics and at the average LTV value for its reference
group, non-Hispanics in the example, and compute the difference between the two values. More
generally, we perform this analysis by computing the mean of each regression variable for each
population group: Asians, NHPIs, and their reference counterpart whites; and bottom quintile income
borrowers and their reference group of all four other income quintile borrowers combined. Taking the
coefficients from our (weighted) regression, we multiply each variable’s coefficient by the difference in
mean between a fair housing population and its counterpart (Asians versus whites, NHPIs versus whites,
and bottom income quintile versus others). In a logistic regression form, the largest negative products
make the most difference in reducing the target population’s probability of obtaining trials compared to
the reference population.
Table 5-7 lays out, for Asians compared to whites, the seven variables with the largest negative
indirect impacts on relative Asian rates of trials. The most powerful variables—income, property value,
borrower total monthly obligations, and pre-modification net present value of the mortgage—are ones
in which Asians have higher mean values, with (according to the regression) negative effects on trials. It
seems difficult to argue that one would want to instead avoid taking these variables into account in
approving modifications, or take them into account in the reverse direction, so that wealthier Asians
would obtain more trial mods relative to the majority population. But it is possible that one or more of
the seven represents a greater need for assistance or unnecessarily filters out borrowers who could
significantly benefit from and sustain a modification.
Table 5-7. Seven Variables with Largest Negative Impact for Asians in Trial Modification Regression
Analysis.
47
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.50
Table 5-8 lays out the variables with largest impacts on NHPI versus white probabilities of
obtaining HAMP trials, based on the same analysis. The set of variables is, except for NHPI status itself,
much the same as for Asians. The list again suggests little that on its face would argue for program
adjustment to improve probabilities of trials for NHPIs, although they as a population have the most
sharply lower likelihood of obtaining those modifications. Perhaps those borrowers with greater total
monthly obligations are in greater need of assistance, although this is only the third most influential
variable. A multivariate analysis of the ability of borrowers with high total obligations to sustain
modifications, beyond the scope of this study, would be helpful in assessing the value of adjusting use of
that variable in assessing trials.
Table 5-8. Seven Variables with Largest Negative Impact for Native Hawaiians and Pacific Islanders in
Trial Modification Regression Analysis.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
50
The term front-end DTI refers to the ratio of mortgage expenses to income, whereas back-end DTI refers to the
ratio of all debt payments (including mortgage payments as well as, for example, car loan payments) to income.
48
Table 5-9. Seven Variables with Largest Negative Impact for Low-Income Group in Trial Modification
Regression Analysis.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
The parallel analysis for people of bottom-quintile incomes is contained in Table 5-9. The mean
values used in the computation are for the bottom quintile and for the other four quintiles combined.
Here we have a variety of variables, within the top seven impacts, that lower low-income people’s trials
apparently because of their smaller resources. Examples include their smaller pre-mod housing expense,
the longer post-mod amortization term, and the higher pre-mod total debt to income ratio. It may be
important to use these variables to lower risk of failed modifications, even at some cost to lower income
applicants’ ability to obtain trials. Whether these variables actually are usefully in signaling risk could
again be tested in multivariate analysis of the sustainability and redefault probabilities of trials granted.
Combining Eligibility Analysis with Regression Analysis of Trial Modifications for Eligible Applicants
As detailed above, fair housing populations may potentially be at a disadvantage in obtaining
HAMP trial modifications either because they have lower rates of eligibility for the program or because
eligible borrowers are receiving and accepting trial approvals at a lower rate, or a combination of the
two. Table 5-10 summarizes these outcomes, allowing us to assess combined impacts.
For many population groups, the results are consistent in direction and thus clear in outcome.
African Americans, Asians, women, other (two or more) races, and low-income people—indicated in
blue in the table-- all have higher rates of eligibility among applicants than do their reference groups and
either higher rates of trial approvals among those eligible borrowers than their counterparts or equally
high rates. On both counts, they do at least as well as whites, men, and higher income people
respectively. Therefore they do at least as well in obtaining trials. For the most part, the differences are
small, because the differences in both eligibility rates and success rates in gaining trials once eligible,
even when present are small. The exception is in substantially higher trials rates for low-income people,
49
everything else equal. But as we have previously noted, it is not clear that incomes can differ widely
without influencing other explanatory variables as well.
Table 5-10. Fair Housing Population Results for Eligibility and Success in Obtaining HAMP Trial
Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: In the eligibility analysis, positive indicates fair housing population applicants are eligible in higher
percentages than counterpart whites (for race), non-Hispanics (for Hispanics), males (for females), and
higher income people (for bottom income quintile). In the regression analysis, positive indicates fair
housing population eligible applicants have higher probability of obtaining trial modifications than
respective counterparts. Neutral indicates no significant difference from counterparts.
The situation is equally clear for Native Hawaiians and Pacific Islanders, but to the disadvantage
of NHPIs compared to whites. NHPIs applicants—red in the table-- are less likely to meet eligibility
criteria and less likely to obtain a trial modification approval if eligible. They clearly have a lower
likelihood of obtaining trials than do their white counterparts, other things equal.
The picture is mixed only for American Indian and Alaska Natives and Hispanics—yellow in the
table. Both of those populations have an advantage in eligibility and a disadvantage in approval among
eligibles, although the AIANs’ partial disadvantage is of more marginal significance in the weighted
regression and not significant in the unweighted version.
We can combine the eligibility and approval rate findings to (1) estimate the total disadvantage
of NHPIs from the two negative effects and (2) determine the net advantage/disadvantage for AIANs
and Hispanics from the impacts operating in conflicting directions.
We have seen in Table 5-2 that 41.4 percent of NHPI entrants to the HAMP process are
ineligible, compared to 21.6 percent for whites. That makes NHPIs 75 percent as likely to be eligible.51
51
(1- .414)/(1-.216) = .75 = 75 percent.
50
And NHPIs’ and white’s percentages of eligible applicants obtaining trial modifications are respectively
54 and 48 percent (see Table 5-6). The combined effect is that NHPI applicants have only two-thirds (28
percent vs. 42 percent) the likelihood of receiving a trial modification that whites do (see Table 5-11)
Most of the difference derives from the eligibility factor based in very basic program criteria.
For AIANs, the eligibility differential is positive but small at 4 percent, and the trials percentage
disadvantage among eligible borrowers is 2 percentage points. The combined impact just balances out,
so that AIAN applicants have effectively the same likelihood of ending up with a trial modification as do
whites (42 percent).
Table 5-11. Net Differential Rates in Obtaining Trial Modifications, Combining Eligibility and Approval
among Eligible Applicants, for Population Groups with Lower Rates of Eligibility, Approval of Eligibles, or
Both.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: The third column is a product of the first two.
For Hispanics, the eligibility advantage (8 percentage points, 86 percent compared to 78
percent) is larger than the disadvantage in trial approvals among eligibles (2 percentage points, 52
percent vs. 54 percent). The result is a net advantage to Hispanic applicants over non-Hispanics in
obtaining HAMP trial modifications overall (45 percent to 42 percent), despite the regression result that
among eligible applicants Hispanics fare slightly less well than their counterparts.
In sum, among fair housing populations, only Native Hawaiians and Pacific Islanders combine
eligibility and approval factors in a way that result in disadvantage compared to majority counterpart
populations in obtaining HAMP trial modifications. The absolute numbers of these populations,
compared to the total application population, is small.
51
6.
Fair Housing Characteristics and HAMP Permanent Loan
Modifications
The HAMP program is designed to convert trial modifications into permanent modifications
following a three-month trial period, provided the homeowner makes full timely payment of the
modified mortgage payment amounts, maintains eligibility status, and provides any necessary additional
documentation consistent with the modification application on which the trial was based (or amended
information should conditions change). The permanent modifications are, like the trial modifications,
designed to reduce housing expenses to 31 percent of homeowner incomes, at interest rates as low as 2
percent, but for extended terms beyond the trial period. For some loans, the interest rate is fixed for the
remaining duration, while with others, it stays at a base level for five years, and then rises at 1% per year
until reaching an interest rate cap.52 Except in situations in which conditions change between the trial
modification approval and the consideration of the permanent modification, the decision to approve
and the basic structure of the modification should not change. The determinants of the approval should
be essentially the same ones that drove the decision to grant the trial modification, with the important
exception of the requirement for timely payment on the trial. Additional documentation of
circumstances is also required for permanent modifications, especially as compared to early trial
modification applications, so that further changes may occur because of differences between initially
stated and documented conditions.
As with the analysis of trial modifications, we analyze fair housing populations’ experience in
HAMP permanent modifications using simple tabulations, a separate analysis of program eligibility,
multivariate regression, and a combination of the eligibility and regression analyses. The remainder of
this section details the findings.
Experience of Fair Housing Population Groups in Permanent Modifications: Tabulations
The Treasury HAMP database indicates whether trial modifications have been advanced to
permanent status. They may be currently active permanent modifications or, after being made
permanent, may have been paid off or disqualified and therefore become no longer active.
Alternatively, a trial modification may remain active as a trial or be cancelled or disqualified. Our focus is
on whether the advancement to permanent modification took place, including all three of the options
(active, paid off, disqualified) for current status, because all three involve an affirmative decision by
lender/investor and homeowner to proceed with the permanent modification and could be sustained by
the homeowner thereafter. Having already analyzed above the likelihood of applicants advancing to trial
modifications, we now consider the share of those who made it to trials who now succeed in obtaining
permanent modifications.
52
If the modified interest rate is below the Freddie Mac Primary Mortgage Market Survey rate when the
permanent modification begins, then the rate will increase over time before hitting the cap. If the modified rate is
at or above the Freddie Mac rate, the rate will stay fixed. For more information, see U.S. Department of Treasury,
“Making Homeowner Frequently Asked Questions,” http://www.makinghomeaffordable.gov/faqs/homeownerfaqs/Pages/default.aspx (accessed December 1, 2011).
52
Table 6-1 summarizes the distribution of permanent modifications and trial modifications
among our target fair housing populations and their corresponding reference groups. The first column
includes all loans modified at any stage: those receiving only trials (at least thus far) and those that have
gone on to permanent modifications. Since every modified loan initially was a trial modification, this
column is also the total number of trial modifications initiated. The second column specifies the
numbers that have received permanent modifications (after trials). It includes the permanent
modification subcategories in columns 6, 7, and 8: active permanent modifications and those that have
been cancelled (generally for failure of timely payment) or their loans paid off. Loans that were given
trial modifications but have not gone on to permanent modifications (the difference between columns 1
and 2) are broken down in columns 3, 4, and 5, including active trials which may or may not become
permanent later and trials that have been cancelled or disqualified. Finally column 9 includes every loan
whose borrower entered the HAMP application process, both all with any modification and the NANAs
who failed to attain trials.
Forty percent of all homeowners in trials made the transition between trials and permanent
modifications: 579,661 permanent modifications in column 2 out of 1,450, 630 trial modifications from
column 1.
The percentages in the table’s first two columns indicate the share of a given population among the
total trial modifications recipients and permanent modifications recipients respectively. If the
percentage is higher among permanent modification recipients than among trial recipients for a
population group, then the population group performed better than the overall average in obtaining
permanent modifications. These tabulation results of course do not yet control for other factors than
population group, including eligibility, financial characteristics of borrowers, mortgage characteristics
before and after modification, housing market prices, and the relations among these items.
Homeowners in the fair housing population groups in general did as well in obtaining permanent
modifications as they did in getting trials. The lowest income quintile (and the highest) fell slightly short
of gaining their share of modifications (19.1 percent), given their receipt of trials (19.5 percent); but the
difference was less than half a percentage point. And the mean income of those who obtained
permanent modifications was actually lower than those who got trials and significantly lower than the
mean for those whose trials were disqualified or cancelled. On the other hand, those in the bottom
quintile remaining in active trials were a lower percentage than the all-trials group, with more being
disqualified and cancelled. It is possible, therefore, that when all current trials are finally settled as
either permanent modifications or not, the fall-off in permanent modifications for those of lowest
income may be somewhat larger than the 0.4 percentage point differential.
Women who had obtained trials actually did better than men in obtaining permanent modifications.
They received a higher percentage of the permanent modifications (39.1 percent) than of trials (37.9
percent). Women also received a higher percentage of active trials than cancelled and disqualified trials,
so that their positive margin in permanent modifications may grow as remaining active trials are settled.
53
Hispanics exactly matched their share of all trials in obtaining permanent modifications, at 31.2
percent. But their percentage of still-active trials slightly trailed that of cancelled or disqualified trials,
which might eventually lead to Hispanics falling slightly behind non-Hispanics in permanent
modifications.
Among racial groups, African Americans obtained a higher percentage of permanent modifications
than trials (21.0 versus 20.5 percent). Further, their percentage of active trials exceeded that of
cancelled or disqualified trials by a full percentage point, so that the margin may widen a bit as more
trials receive final decisions. Asians on the other hand received a very slightly lower percentage of
permanent modifications than trials (one-tenth of a percentage point). Further, a slightly lower
percentage of active trials remain for them, versus cancelled and disqualified trials. They are the one
racial group trailing overall average in conversion to permanent modifications and only by the narrowest
margin.
54
Table 6-1. Fair Housing Populations in All Modified Loans and Permanent Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
* Column 1 includes all loans which advanced at least as far as receiving a trial modification, some of which made it to the permanent
modification stage. Column 2 includes all loans which advanced at least as far as receiving a permanent modification.
55
American Indians and Alaska Natives obtained a very slightly higher percentage of
permanent modifications than trials and the same percentage of active and
cancelled/disqualified trials. Native Hawaiian and Pacific Islanders received a higher percentage
of permanent modifications than of trials and have a higher percentage of active than cancelled
and disqualified trials, suggesting a slightly improving margin as trials are completed. People of
more than one race just matched average receipt of permanent modifications and remaining
active trials.
Thus among the fair housing populations, only modestly for lowest quintile income people,
plus to a very small extent Asians, is there an overall disadvantage, compared to their respective
counterparts, in likelihood of obtaining a permanent modification given having obtained a trial.
The same is true in terms of trials maintained as active—thus still with a reasonable chance of
obtaining a permanent modification eventually—versus trials disqualified or cancelled, except
that Hispanics also trail slightly in that comparison.
These are of course simply tabulations. Our regression analysis sorts out whether these or
any other categories of fair housing groups suffer once many other factors are controlled for.
Permanent Modifications and Eligibility
People not meeting HAMP eligibility standards should in general not have been granted
trial modifications and should thus not be eligible for permanent ones. However there is the
possibility of changing circumstances, during trials, with eligibility consequences, such as moving
out of home and becoming no longer an owner-occupant as required. Borrowers experiencing
such circumstances would “fall out” of trial modifications, with trials either cancelled or
disqualified. Borrowers falling out of trials receive trial fall-out codes in the Treasury’s database,
some though not all of which are for reasons involving eligibility. The four fall-out codes (out of
a total of eleven) that involve eligibility are:

The loan is considered an “ineligible mortgage” for one of two reasons:
o
The loan is not a first lien mortgage which originated on or before January 1,
2009
o
The amount of unpaid principal is greater than a cap for a given number of units
(e.g., $729,750 for a one-unit home)

The borrower’s monthly mortgage payment is less than 31% of his or her monthly
income

The property which secures the mortgage is renter-occupied

The property which secures the mortgage is vacant or contains more than four units
56
We test whether fair housing population groups are more likely to lose access to
permanent modifications than others, by dropping out of trials for these eligibility reasons. For
the same reasons involving certainty in eliminating ineligible borrowers and for lack of data
(here for recipients of permanent modifications) that were discussed previously for trial
modification applicants, we apply cross-tabulation and Chi-Square testing to this question rather
than regression analysis.
The comparison is between all applicants who received at least a trial modification and
those in fair housing populations who did so. We examined whether any of the fair housing
populations were more likely to receive one of the four ineligibility codes than their reference
comparison population. Borrowers not receiving such a code received either another, noneligibility-related fall-out code or, if they were still in active trials or had advanced to a
permanent modification, a missing data designation.
Not a single one of the target population groups has a higher percentage of ineligibility
determinations during trials than did its reference population. The results are in Table 6-2. Many
of the population groups—for example African Americans and Asians--had eligibility fall-out
percentages very similar to those for their majority counterparts (in that case whites), while
some others had significantly lower percentages (Native Hawaiians, lowest income quintile).
None had more acute eligibility fall-out than their comparison group.
Table 6-2. Eligibility Analysis for Permanent Modifications, by Fair Housing Population.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
57
Because none of the tabulations showed a disadvantage for our target populations,
there was no reason to undertake testing of the statistical significance of the differences.53 In
testing ineligibility among all applicants, we found some fair housing populations with higher
ineligibility than their reference groups and sought to break down the specific standards that
expanded their ineligibility. Because there are no such populations in the trials households
seeking permanent modifications, there is no corresponding analysis to conduct here.
Multivariate Analysis of Permanent Modifications
For borrowers who are still eligible for the HAMP program after they have obtained
trials and while they are being considered for permanent modifications, we need to separate
out the impact of membership in the various fair housing population groups on likelihood of
actually receiving a permanent modification. We undertake a LOGIT regression of probability
that a permanent modification is granted to a borrower that has received a trial modification,
given his/her fair housing characteristics, and control variables for financial condition and
mortgage risk, housing value, mortgage type, and nature of modification offered, as detailed in
section 3. Ineligible borrowers are again eliminated from the analysis, given their treatment in
the previous section.54
As explained earlier, no weighting of cases is required for analysis of permanent
modifications. Besides eliminating weighting, the other primary differences between the
regression for trial modifications and that for permanent modifications are a few changes in
control variables, discussed under methods and models earlier.
The regression we estimate performs very well in modeling actual approvals of
permanent modifications from among eligible applicants who have received trial modifications
previously. Nearly all of the included explanatory variables have statistically significant
coefficients at the 1 percent level or better. Among all trial modification recipients, actual
outcomes are that 54 percent are granted modifications and 46 percent are denied.55 The model
as estimated is able to predict correctly 79.9 percent of the denials of permanent modifications,
and 88.1 percent of the approvals, using a threshold level of 50 percent probability of approval
to make the predictions (see Table 6-3). The regression explains substantial levels of variation in
outcomes, by both of the reported pseudo R-squared standards. The full regression results are
laid out in Appendix Table 3.
Key results for the fair housing populations are in Table 6-4. Once again, we focus on the
odds ratios column, where ratios of greater than 1.0 mean that a one unit increase in the in the
variable (here, except for the case of income, a dummy variable representing membership in a
population group) increases the odds of achieving the desired outcome of a permanent
53
A test of a populations’ lower percentage ineligibility could not possible reveal a statistically significantly
higher percentage.
54
Ineligible homeowners whose ineligibility was indicated before the decision on a trial modification have
previously been deleted in that they did not make it to the trials modification status.
55
This is only coincidentally the same percentage of all eligible applicants who receive trial modifications.
58
modification compared to that for cases without that unit increase (here the counterpart
majority population).
Table 6-3. Predictive Accuracy and Summary Statistics for Multivariate Regression Model for
Permanent Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: The “modified loans which have not reached permanent status” category includes active
trials, canceled trials, and disqualified trials. The “permanent modifications” category includes
active permanent modifications, loans which have been paid off, and disqualified permanent
modifications.
Table 6-4. Permanent Modifications Regression Results for Fair Housing Populations.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Most of our population groups of fair housing interest have at least as high odds of
obtaining permanent modifications, given their previous receipt of trial modifications, as do
59
their counterpart populations, after controlling for the many other variables included in the
regression analysis. Among racial groups, African Americans have slightly higher odds56 than
whites of obtaining permanent modifications; American Indians and Alaska Natives, Asians, and
Native Hawaiian and Pacific Islanders have odds of receiving permanent medications not
statistically different from the majority whites; and only the very small other (more than one)
race category has statistically significantly lower odds than whites.
Hispanics are the one population of major size that does show lower odds (odds ratio of
0.95) of receiving permanent modifications than its counterpart, non-Hispanics, with a high level
of statistical significance. Women, on the other hand, have higher odds than men for permanent
modifications. People of lower income are estimated to obtain permanent modifications with
the same odds as those with higher incomes (estimated as an odds ratio slightly larger than but
not statistically significantly different from 1.0).
Translating the odds ratios into simple percentages for the two categories, Hispanics
and more than one race, which have reduced odds of permanent modifications compared to
their reference groups shows that the differences in approval rates are modest. With other
explanatory variables equal at overall mean levels, 51 percent of Hispanics in trials obtain
permanent modifications, compared to 54 percent for non-Hispanics. The difference is sharper
for people of more than one race, at 7 percentage points (47 vs. 54 percent), but the percentage
of homeowners in that category is very low.
One specific concern we had about our regression specification was whether using a
dummy variable to represent trial modifications that had proceeded for fewer than three
months, during which according to program rules permanent modifications could not be made
because of the requirement for three months of timely and full payments, was a satisfactory
approach. Alternatively, one could treat those trials as temporarily ineligible for conversion to
permanent. We tested an alternative specification, without that dummy variable and excluding
from the regression analysis the cases in which trials were too new to allow for permanent
modification. None of the results for fair housing variables were materially changed from those
we report above—not surprising given that the dummy’s coefficient when the variable is
included is, while large, not statistically significant.
Other Characteristics of Fair Housing Populations Affecting Outcomes
The previous section lays out the apparent direct impact of fair housing population
group membership on likelihood of obtaining permanent modifications, once receiving a trial. As
with homeowners obtaining trial modifications, we want to examine any effects on receipt of
permanent modifications that may arise due to differences between fair housing groups and
their counterpart populations other than their group membership itself.
56
Statistical significance is somewhat marginal at the .06 level.
60
For permanent modifications, this analysis of other differing characteristics focuses on
the two population categories in which simple tabulations showed lower rates of success than
their counterparts: people in the bottom income quintile and, to a small extent, Asians. It
identifies which other characteristics most contribute to the shortfall in permanent modification
odds for those two groups. (As explained earlier, the analysis makes less conceptual sense in a
fair housing context for groups outperforming the norm, because we would be explaining why
they might not be outperforming the norm by a wider margin rather than why they were at a
disadvantage). As with our trials analysis, we compute the mean of each regression variable for
each population group and their counterparts: here bottom quintile versus combined higher
income, and Asians versus whites. Then we multiply the respective differences in means by their
regression coefficients and look for the largest negative products, which in a logistic regression
most reduce relative odds of the target group’s success in gaining permanent modifications.
For Asians, the two largest factors are net present value of their loans before
modification and pre-modification housing expense, for both of which they have higher mean
values than do whites (see Table 6-5).57 The former relates to a basic HAMP program tenet, that
loans with higher value to investors in their current form, relative to modified form, should less
likely be modified. That has a small differential impact on Asians. The latter may, insofar as it is
being used as grounds to deny permanent modification, be a candidate for further analysis of
the appropriateness of its own role and its impact on NPV. There is at least a question of
whether modifications should be less likely for those starting out with higher housing costs,
other things equal. The fact that the differential in modification likelihoods is so small for Asians
compared to whites, however, mitigates possible concern.
Table 6-5. Seven Variables with Largest Negative Impact for Asians in Permanent Modification
Regression Analysis.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
57
The third largest factor is timing of the effective date of the modification squared. This is misleading.
Coupled with the effective date itself, the higher value for Asians actually benefits them compared to
whites, at least at the mean values.
61
For people of lowest income, the prime contributors to lower permanent modifications
seem to reflect their lower level of resources: post-modification net present value, post-mod
housing expense, as-is value of home, and borrower total monthly obligations (see Table 6-6),
for all of which they have lower values. For the first, there is again the question of protecting a
basic program element versus its effect on people in the lowest income quintile. How the last
three would add to, rather than subtract from, risk of a modified loan is not obvious. Such
factors could be candidates for further examination in terms of their value for financial analysis
versus their impact on low-income people. Again the small size of the disadvantage of low
income people versus others in obtaining permanent modifications mitigates the seriousness of
possible issues.
Table 6-6. Seven Variables with Largest Negative Impact for Low-Income Group in Permanent
Modification Regression Analysis.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Combining Eligibility Analysis with Regression Analysis of Permanent Modifications for Eligible
Applicants
Fair housing populations may potentially be at a disadvantage in obtaining HAMP
permanent mortgage modifications, given their receipt of trials, because they have lower rates
of eligibility for the program, because eligible trial recipients are receiving and accepting
permanent modifications at a lower rate, or a combination of the two. Table 6-7 summarizes
these outcomes, allowing assessment of combined impacts.
The great majority of the fair housing populations do not suffer a disadvantage
compared to their reference groups in obtaining permanent modifications after receiving trials.
For most of the focus populations, the combined results of any differentials in eligibility and in
approvals of those eligible are readily apparent. Because all of the eligibility differentials are
positive or statistically neutral, only those populations with significant negative results in the
regression analysis, relative to their counterpart populations, can even possibly be at a
disadvantage in obtaining permanent modifications. Thus we can see directly that African
Americans, AIANs, Asians, NHPIs, women, and lowest income people (shaded blue) all do at
least as well respectively as whites, men, and higher income homeowners in obtaining
62
permanent modifications once they have received trials. They neither lose more in eligibility
filtering of all trials, nor in overall approvals among those who have trials and are eligible once
other factors are controlled for.
Table 6-7. Fair Housing Population Results for Eligibility and Success in Obtaining HAMP
Permanent Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
The two remaining populations of concern are homeowners of more than one race and
Hispanics (shaded yellow). In each of these cases, trial recipients have a first step in their favor:
lower likelihood of being ruled ineligible (higher percentages eligible) than whites and nonHispanics respectively. And they have a second step in their disfavor: lower likelihood of gaining
a permanent modification for eligible borrowers with trials than their counterparts.
Table 6-8. Net Differential Rates in Obtaining Permanent Modifications, Combining Eligibility
and Approval among Eligible Trials, for Population Groups with Lower Rates of Eligibility,
Approval of Eligibles, or Both.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: The third column from left is a product of the first two.
63
Drawing on the results in tables 6-2 and 6-4 for the size of these impacts, we find that
people of more than one race do in fact have modest combined disadvantage. But Hispanics do
not. The combined results are in Table 6-8. Homeowners of more than one race already in trials
have a higher eligibility rate than whites (90 vs. 87 percent). But that is more than offset by their
disadvantage in approvals among eligible trials (47 to 54 percent). As a result, their approval
rate for permanent modifications is smaller at 43 percent than the comparable rate for whites at
47 percent. For Hispanics, the advantage in eligibility just offsets their disadvantage in approval
rate among eligibles compared to non-Hispanics, producing a 46 percent rate of permanent
modifications among people with trials for both Hispanic and non-Hispanic populations.
64
7.
Fair Housing Characteristics and Sustaining HAMP
Permanent Modifications
Obviously HAMP is much more valuable to homeowners in general and to fair housing
populations in particular insofar as owners are able to meet the payments of their permanent
modifications and thus sustain those modifications over time. Homeowners having received
permanent modifications and sustained them with timely payment are, in Treasury’s HAMP
terms, in “active” permanent modifications.58 If the borrower continues to make his or her
payments, the permanent modification will be sustained.59 More than 90 percent of all
permanent modifications in our data are still active.
Due to resource constraints, this study did not include a full-scale analysis of sustaining
permanent modifications by fair housing and counterpart populations. Ideally, one would
construct a multivariate logistic regression model of the likelihood of sustaining or defaulting on
the modification, structured to estimate the changing risk of failure over multiple periods of
time since the execution of the permanent modification. We include here only a simple
tabulation of the share of fair housing populations’ permanent modifications that remain active
and those disqualified. Parallel to our tabulations for trial modifications versus NANAs and
permanent modifications versus trials, we compare the focus populations’ share of active
permanent modifications versus their share of all permanent modifications received. Those
populations with a lower share of active permanent modifications than their share of all
permanent modifications are sustaining those modifications less well than others.
As shown in Table 7-1, fair housing population groups except African American and
multiple race homeowners are sustaining their modifications at least as well as are their
respective counterparts, at the simple tabular level of analysis. Among races, Asians and AIANs
have higher sustainability than the rest of the population, and NHPIs’ level is equal. Hispanics
have a higher share of active permanent modifications than they do of all permanent
modifications, by just over a half percentage point; and very low-income people have a bit
smaller but still positive margin in sustainability. Women have the same share of active
permanent modifications as they do of all permanent modifications.
African Americans have, by half a percentage point, a smaller share of still-active
permanent modifications than their percentage of all permanent modifications received. At
least prior to controlling for other factors, including time since modification issuance, they have
a modest disadvantage in sustainability. The small population of homeowners of more than one
race has a small disadvantage as well.
58
A tiny number (0.1% of permanent modifications) have received permanent modifications and then
paid off their loans and are separately delineated.
59
Permanent modifications are subject to a series of conditions, like keeping the modified loan as a first
lien loan. For a full list, see U.S. Department of Treasury, “Making Home Affordable Handbook v3.3,”
https://www.hmpadmin.com/portal/programs/docs/hamp_servicer/mhahandbook_33.pdf (accessed
December 2, 2011).
65
Table 7-1. Fair Housing Populations Compared to Counterparts in Sustaining of Permanent
Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
These rough tabular comparisons provide a first estimate of the relative experience of
fair housing populations in sustaining their modifications. They suggest that disadvantages, if
they exist once other factors are controlled, may be of modest size. But additional modeling is
required to know for certain whether disadvantages persist once other factors are controlled,
for African Americans and homeowners of more than one race and for our other target
66
populations; how large they are if any; and what aspects of the nature of HAMP modifications,
characteristics of different borrowers, and/or housing markets may pose additional challenges
for fair housing populations.
67
8.
Summary of Findings for HAMP Direct Effects for
Fair Housing Populations across Stages of HAMP
Activity
This report’s series of analyses of direct impacts of HAMP for fair housing populations
compared to their respective majority counterparts allows us to look across all the stages of
HAMP action. We have population group results for HAMP applicants among all borrowers with
loan troubles, for trial mortgage modification recipients among all applicants (in terms of
eligibility and among eligibles), for permanent mortgage modification recipients among all trial
recipients (again in terms of eligibility and among eligibles), and those able to sustain permanent
modifications among all recipients of permanent modifications. The directions of all these
results are summarized in Table 8-1.
Overall, there is broad evidence that the HAMP program serves fair housing populations
at least in proportion to their shares of mortgage difficulty throughout the mortgage
modification process. Among racial groups, African Americans’ share of HAMP activity exceeds
its share of candidate borrowers at every stage but the last. For them, the positive differentials
they obtain at application and trial and permanent modification stages outweigh the single small
negative differential in terms of sustaining modifications.60 Asians’ positive or neutral shares
relative to whites at every stage produce an overall positive position. Among the smaller racial
groups61, American Indians and Alaska Natives have positive or neutral experience compared to
whites at each stage except review of eligibility for trials, and we have seen that that single
disadvantage is fully offset by just their positive position in trial approvals among eligibles.
People of more than one race come very close to balancing between positives and negatives.
Native Hawaiians and Pacific Islanders are the one racial group which benefits from HAMP less
than in proportion to their entrance into it. We lack separate data on their share of troubled
loans. But among those who start applications, we know they lag in gaining trials, both in
eligibility and in approvals among eligibles; and their advantage in eligibility at the permanent
modification stage does not fully offset those lags.
60
Recall that even that negative is based only on simple tabulations thus far and needs further
multivariate analysis.
61
We acknowledge, without repeating for each population group, that AIANs, NHPIs, and Other (more
than one) race, as well as women, do not have separate comparison data for borrowers with mortgage
trouble and all applicants. The conclusions necessarily apply to the modification stages, for which we do
have data. The three small racial populations together total slightly less as a share of all applicants than
does the “other” category in the analysis of troubled borrowers by Bocian et al. (2011); but that analysis
does not provide further breakdown.
68
Table 8-1. Summary of Findings for HAMP Direct Effects for Fair Housing Populations across Stages of HAMP Activity.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: Positive means that the fair housing population fares better than its counterpart. In the left column only, the racial categories include only
non-Hispanics, whereas in the other analyses, those groups include Hispanics.
69
Hispanics present a slightly more mixed picture. Among HAMP-eligible homeowners, their
percentages of trial and permanent modifications approved are modestly smaller than for nonHispanics. But they are more likely to be eligible, and they enter the application system more than in
proportion to their share of troubled loans. Overall Hispanics have a very slender positive margin in
HAMP participation through its multiple steps, compared to non-Hispanics.
Women obtain and sustain HAMP modifications at least as well as men, with a positive or
neutral share of modifications out of candidates for each stage for which we have data. An important
gap is that, as with smaller racial groups, we do not have data for women’s share of troubled mortgage
situations, to which to compare their modification success.
Finally, people of lower income are faring positively in the HAMP program. Among applicants,
they are more than proportionally eligible at the trial and permanent stage; they receive more than a
proportionate share of trial modifications among eligibles; and they more frequently sustain permanent
modifications. An important limit to this conclusion is that we lack a fully comparable measure of
troubled loans among low-income borrowers to compare to their share of entering applicants to HAMP.
But our rough approximation indicates a significant advantage to people at the lower end of the income
scale, so that given the positive results for other stages of HAMP consideration, we can reasonably
expect low-income people to do well in the program.
70
9.
Indirect and Net Impacts of HAMP by Fair Housing
Populations
As we have noted earlier, observers suggest that HAMP can have impacts not only directly in
providing HAMP loan modifications but also potentially indirectly. Direct impacts may understate
HAMP’s impacts, because HAMP may expand its impact by shaping the response of lenders/investors to
requests for proprietary modifications not within the HAMP program. In the other direction, HAMP’s
direct impacts may overstate its net effects because HAMP modifications may substitute for non-HAMP
loan workouts that would otherwise have occurred in its absence. Either or both of these impacts may
be different for fair housing populations than for majority homeowners.
Here we draw on results of a related research project by ourselves and colleagues, developed in
an evaluation of the NFMC foreclosure prevention counseling program, to assess net impacts of HAMP
including both indirect and direct effects. We do that by looking at a series of foreclosure prevention
outcomes for homeowners—without regard to their participation in HAMP-- during the period before
HAMP was implemented beginning in April, 2009, compared to the outcomes in the period from
January, 2008 through March, 2009 (see Mayer et al. 2011). We compare the changes in outcomes
between the two periods for majority populations with the corresponding changes between periods for
fair housing populations. These differences in HAMP impact are visible in the coefficients of regression
variables representing the interaction between a HAMP indicator designating the pre- and post-April
periods and indicators of membership in the fair housing population groups. These interactions indicate
whether being in a fair housing population group (each one separately) changes the impact HAMP has
on mortgage outcomes once it comes into operation. In the study we draw upon, we have analysis for
HAMP’s initiation interacting with race in categories of African Americans, Asians, and all other races
combined, compared to whites; with Hispanic and non-Hispanic ethnicities; and with income levels
(measured as a continuous variable).
We consider the impacts on six measures of help to homeowners in dealing with potential
foreclosures: the size of loan modifications that take place in terms of reduced monthly mortgage
payments, the percentage of all cases with foreclosures begun in which the foreclosure is stopped
(changed to some non-foreclosure status) during the study period, the percentage of seriously
delinquent (here 90+ days) loans and foreclosures underway in which mortgages are cured to current
status through a modification (loan modification cures), the percentage curing of the same loans
without modifications (non-modification cures), the percentage of modification cures that later
redefault, and the percentage of non-modification cures that redefault.
Table 9-1 summarizes the direction and significance of the effects of HAMP interacting with fair
housing population on these six outcomes, compared to HAMP’s impacts for majority populations.
These are estimates of net effects, including all modifications, foreclosure stops and cures, and
redefaults, including both HAMP and proprietary workouts and homeowners regardless of whether they
initiated HAMP or other modification applications. Note that a positive effect is the more desirable
outcome for a population group in the table’s first four columns, signaling larger modification size and
71
higher likelihood of foreclosure stops and cures. A negative effect is desirable for the last two columns,
signaling reductions in redefaults. “None” means no statistically significant effect. Complete estimated
equations supporting Table 9-1 are available from the authors.
72
Table 9-1. Summary of Net (Direct and Indirect) Effects of Fair Housing Variables on Impacts of HAMP for Various Outcomes.
Source: Author's computations from NFMC and LPS data.
Note: For the first four outcomes, “positive” results are beneficial to the homeowner and “negative” are disadvantageous. In the last two
outcomes on redefaults, “negative” results (fewer redefaults) are beneficial.
73
African Americans present the most problematic picture among people of color, in terms of net
benefit from HAMP creation and operation. The positive indicator in the first column means this
population group gets more benefit in increased modification size from HAMP’s initiation than do
whites. But the indicator is negative for likelihood of stopping foreclosures, meaning African Americans
get less help from HAMP than do whites in halting foreclosure processes once begun. The picture is
favorable for African Americans with respect to curing delinquencies and foreclosures. But African
Americans get less benefit from HAMP in lowering redefault rates than do their white counterparts,
whether redefaults of modification or non-modification cures.
The situation for Asians is more consistently favorable or neutral in terms of HAMP’s impacts
compared to its effect for whites. Asians receive higher HAMP benefits in loan modification size and
foreclosure stops, and equal benefits for three of the remaining outcomes. Only in loan modification
cures is their benefit from HAMP smaller than that of whites. Other races combined fare equally with
whites in terms of HAMP benefits on all outcomes but modification size, in which they have an
advantage.
Hispanics quite consistently reap more impact from HAMP than do non-Hispanics. The HAMP
impacts are larger for Hispanics in terms of modification size and foreclosure stops and equal in both
types of cures. HAMP contributes more greatly to decline in Hispanic redefaults from both modification
and non-modification cures. Thus for neither Hispanics nor races other than African American and
Asians are we given reason for concern regarding their net HAMP impacts.
Lower income homeowners consistently receive less benefit from HAMP’s creation than do
people of higher income. Higher earners get a larger boost in loan modification size, foreclosure stop
frequency, and likelihood of modification cures as a result of HAMP. They don’t differ in obtaining
modification cures but they do get more non-modification cures. HAMP gives them greater help in
avoiding redefaults of both types of cures. Most of the advantages in foreclosure outcomes go to those
with relatively high incomes.
It is instructive to look at the quantitative scale of the difference in impacts of HAMP on differing
population groups and to place those measures in the context of HAMP’s total impact including the
population group differentials. Table 9-2 provides the specifics. Each row of the table represents a
separate regression analysis which includes a single interaction between the variable in question and
the initiation of HAMP. For each foreclosure outcome, we have a figure for the impact of each
population group on HAMP’s net effect (the same measure whose direction we report in Table 9-1). And
we have a measure of the overall net impact of HAMP on that same population group, from a
combination of HAMP’s effect independent of FHV group and its interaction with the population group.
Consider for illustration the fourth and fifth columns of the Hispanic row of the table. The fourth column
indicates that Hispanics have a 1.3 percentage point higher benefit each month from HAMP in likelihood
of stopping a foreclosure than the benefit non-Hispanics receive. The fifth column in the same row
indicates that HAMP’s total impact on foreclosure stops for Hispanics is a positive 1.6 percentage points
per month, adding together the benefit obtained by the general population and the 1.3 points for
Hispanics specifically. In this case, Hispanic status accounts for most of the HAMP benefit.
74
75
Table 9-2. Detailed Net (Direct and Indirect) Effects of Fair Housing Variables on Impacts of HAMP for Various Outcomes.
(Continued on next page)
76
(Continued from previous page)
Source: Author's computations from NFMC and LPS data.
Note: Grey shading means that there is no significant impact. The impact of HAMP on the size of loan modification is not available because it
depends on the homeowner's income level. The analysis says that HAMP initiation has a negative $49 impact on modification size by itself (for
whites). The impact for people of given income X is thus ($X * $1.70) - $49. Therefore people with incomes above $28,800 will have a positive
impact for HAMP on modification size, and people with lower incomes will have a negative impact.
77
We concentrate on the population groups facing disadvantage with respect to at least one
mortgage prevention outcome, as outlined in our discussion of Table 9-1: African Americans, Asians to a
limited extent, and people of low income. For African Americans, we have already seen that foreclosure
stops is the first outcome in which they benefit less than whites from HAMP in net terms. This is
indicated by the -0.8 monthly percentage of serious delinquencies and foreclosures in process that are
stopped, in column 4. The next column indicates there is no total net monthly impact (not statistically
significant) of HAMP for African Americans. Whites would have a positive 0.8 benefit from HAMP, but
the African American-specific interaction balances that out to zero for blacks. In loan modification cures,
the African-American HAMP-specific effect is a material but not predominant component of the total
HAMP effect for African Americans (0.2 of 0.7 percent monthly). In redefaults, African Americans get
lowered benefit in modifications and other cures. The African American-specific effect considerably (1.5
percent) offsets what would have been a larger reduction in redefaulting modification cures (-5.0
percent for whites becomes -3.5 percent HAMP impact for blacks). There is a somewhat smaller effect
for non-modification cures (0.5 percent African American interaction reduces HAMP impact to -5.5
percent). The total effect of HAMP in redefaults for all homeowners is about 6 to 7 percentage points
per month reduction, so that the impact of African Americans on the percentage rate is significant in
relation to average HAMP impact as well. Overall, the disadvantage to African Americans in the size of
net benefits from HAMP makes a substantial difference in the value of HAMP to them, especially in
stopping foreclosures and in avoiding modification redefaults.
The impact of race on HAMP’s net impact is much more limited for Asians. Only loan
modification cures are reduced relative to those for whites. The reduction is only two-tenths of a
percentage point. And even with that reduction, the overall effect of HAMP, while lower than that for
white by the 0.2 points, is still positive at 0.3 percentage points per month in those cures. HAMP still
helps Asians, in loan modification cures as well as other outcomes—just a bit less than it helps whites for
that single outcome.
The impact of income on HAMP’s net benefit is broad but in most cases limited. As we have
seen, lower income people get less benefit from HAMP than those with higher income in all six
foreclosure outcomes. But outside of the size of payments on modified loans, income’s impact on HAMP
benefits ranges from modest to very small. Even a $50,000 lower income would register very modestly
on foreclosure stops, non-mod cures, and mod-cure redefaults. The effect of income on size of
modifications’ loan payment reductions from HAMP is more substantial. The effect of HAMP alone on
the typical dollar saving from modification is negative (-$49) for people with no income.62 At $50,000
income, the saving would be positive at $36; and each added $50,000 produces an addition to the
HAMP benefit of $85 per month. The impact of income on non-modification cure redefaults is in the
middle range. A $50,000 increase in income would increase the impact of HAMP on redefault rates by a
62
That is, at zero and other low incomes, people who get non-HAMP modifications get larger ones—more deeply
reduced payments-- than do people with HAMP modifications.
78
substantial 2 percent per month63, but the HAMP benefit for people with very low incomes is already a
reduction of 6 percent and more.
63
(-0.04 x 50)
79
10. Summary and Conclusions
This study focuses not on the question of how successful HAMP is overall as a program but on
the extent to which it serves important population groups evenly. Our consistent conclusion from these
multiple analyses is that race, ethnicity, gender, and income have very little impact on homeowners’
direct successful participation in HAMP and on the net benefits of HAMP for homeowners’ overall
mortgage outcomes.
In terms of direct benefits, HAMP serves most racial minorities, Hispanics, women and lowincome people at least in the same proportion as they do their respective reference populations of
whites, non-Hispanics, men, and higher income people. This is true at every step in the program: for
program applicants compared to all households with mortgage problems, for eligible homeowners
compared to all applicants, for trial modification recipients compared to all eligible households, for
permanent modification recipients compared to trial recipients, and for sustained modifications
compared to all permanent modifications. Most fair housing populations enter the program and
advance through the program steps in at least the proportion of their share of candidates for
advancement.
Among racial groups, African Americans’ share of direct HAMP activity exceeds its share of
candidate borrowers at every stage but the last. For them, the positive differentials they obtain at
application and trial and permanent modification stages outweigh the single small negative differential
in terms of sustaining modifications. Asians’ positive or neutral shares relative to whites at every stage
produce an overall positive position. Among the smaller racial groups, American Indians and Alaska
Natives have positive or neutral experience compared to whites at each stage except review of eligibility
for trials, and that single disadvantage is fully offset by just their positive position in trial approvals
among eligibles. People of more than one race come very close to balancing between positives and
negatives.
Native Hawaiians and Pacific Islanders are the one racial group which benefits from HAMP less
than in proportion to their entrance into it. We lack separate data on their share of troubled loans. But
among those who start applications, we know they lag in gaining trials, both in eligibility and in
approvals among eligibles; and their advantage in eligibility at the permanent modification stage does
not fully offset those lags.
Hispanics present an only slightly more mixed picture in terms of direct HAMP benefits than do
populations by race. Among HAMP-eligible homeowners, their percentages of trial and permanent
modifications approved are modestly smaller than for non-Hispanics. But they are more likely to be
eligible, and they enter the application system more than in proportion to their share of troubled loans.
Overall Hispanics have a very slender positive margin in HAMP participation through its multiple steps,
compared to non-Hispanics.
Women obtain and sustain HAMP modifications at least as successfully as men, with a positive
or neutral share of modifications out of candidates for each stage for which we have data. An important
80
gap is that, as with smaller racial groups, we do not have data for women’s share of troubled mortgage
situations, to which to compare their modification success.
Finally, people of lower income are faring positively in direct HAMP program participation and
success. Among applicants, they are more than proportionally eligible at the trial and permanent stage;
they receive more than a proportionate share of trial modifications among eligibles; and they more
frequently sustain permanent modifications. An important limit to this conclusion is that we lack a fully
comparable measure of troubled loans among low-income borrowers to compare to their share of
entering applicants to HAMP. But our rough approximation indicates a significant advantage to people
at the lower end of the income scale. Given the positive results for other stages of HAMP consideration,
we can reasonably expect that low-income people do well in the program.
In terms of net benefits of HAMP—direct and indirect-- on total mortgage outcomes, fair
housing populations also do well.64 But the findings are not quite as consistent as for direct participation
in HAMP. While most of the population groups fare at least as well as their counterparts, in most steps
in the HAMP process, African Americans, Asians (to a lesser extent), and low-income people do less well
in one or more program steps.
African Americans get more help from HAMP than do whites in modification size (reduced
payments) and in curing delinquencies and foreclosure, but less help in halting foreclosure processes
once begun and in sustaining cures of defaults. African Americans do benefit from HAMP in all but one
of those cases, but the gains are smaller than for whites. Asians receive higher HAMP benefits than
whites in loan modification size and foreclosure stops, and equal benefits for three of the remaining
outcomes. Only in loan modification cures is their benefit from HAMP smaller than that of whites. Other
races combined fare equally with whites in terms of HAMP benefits on all outcomes except modification
size, in which they have an advantage.
Hispanics quite consistently reap more net impact from HAMP than do non-Hispanics. The
HAMP impacts are larger for Hispanics in terms of modification size and foreclosure stops and equal in
cures. HAMP contributes more greatly to decline in Hispanic redefaults from both modification and nonmodification cures.
Lower income homeowners consistently receive less net benefit in overall mortgage outcomes
from HAMP’s creation than do people of higher income. Higher earners get a larger boost in loan
modification size. They get a modestly higher benefit in terms of foreclosure stop frequency, and
likelihood of modification cures as a result of HAMP. They don’t differ in obtaining modification cures
but they do get modestly more non-modification cures. HAMP also gives higher income people greater
help in avoiding redefaults of both types of cures. However, only the difference by income in
modification payment reduction is substantial in size. Even that necessarily assumes all other factors but
income are held constant. A low income homeowner with the housing payments, loan amount, and
other factors at the same level as an average borrower will get significantly less of a payment reduction
64
Because the net impacts analysis was not undertaken by gender, we do not have outcomes for women.
81
benefit from HAMP than a household with the same other characteristics but average income. However,
typically the low-income owner will also have other financial characteristics that differ from their higher
income counterparts. These may well moderate the observed differences in this modification size
impact that HAMP appears to be having for differing income groups.
In sum, while the HAMP program has experienced both successes and failures, it systematically
benefits people—both directly and indirectly—in very much the same way regardless of homeowners’
race, ethnicity, gender, and income. There is additional research to be done to tighten these results, as
appropriate data and resources are available. Especially important is a look at differential impacts of
HAMP across neighborhoods of varying racial and ethnic composition. But our results about impacts on
households are robust and consistent.
Measures that strengthen HAMP’s ability to produce good mortgage outcomes for all
homeowners would be highly desirable. Other program analyses have shown important reasons for
modifications in program structure and operations, and our findings do not refute (nor explicitly
support) them. The central lesson of our research is that at least in terms of the distribution of
household (as distinct from neighborhood) benefits and outcomes, attention would best be focused on
the changes and additional supports that benefit all homeowners with mortgage difficulties. Emphasis
can be placed on extending the program’s reach and effects to all homeowners consistent with its
equitable distribution to date, rather than on changing the mix of beneficiaries in terms of fair housing
populations.
82
References
Been, Vicki, Mary Weselcouch, Ioan Voicu, and Scott Murff. 2011. “Determinants of the Incidence of
Loan Modifications.” Paper presented to the Federal Reserve Community Affairs Research
Conference in Arlington, Virginia. April 28–29.
Bocian, Debbie Gruenstein, Wei Li, and Keith S. Ernst. 2010. “Foreclosures by Race and Ethnicity: The
Demographics of a Crisis.” Durham, NC: Center for Responsible Lending. June.
Bocian, Debbie Gruenstein, Wei Li, Carolina Reid, and Roberto G. Quercia. 2011. “Lost Ground, 2011:
Disparities in Mortgage Lending and Foreclosures.” Durham, NC: Center for Responsible
Lending. November.
California Reinvestment Coalition. 2011. “Race to the Bottom: An Analysis of HAMP Loan Modification
Outcomes by Race and Ethnicity for California.” San Francisco, CA: California Reinvestment
Coalition. July.
Carr, James H., Katrin B. Anacker, and Michelle Mulcahy. 2011. “The Foreclosure Crisis and Its Impact on
Communities of Color: Research and Solutions.” Washington, DC: National Community
Reinvestment Coalition. September.
Collins, Michael J., and Carolina Reid. 2010. “Who Receives a Mortgage Modification? Race and Income
Differentials in Loan Workouts.” San Francisco, CA: Federal Reserve Bank of San Francisco.
Working Paper 2010-07. December.
Cordell, Larry, Karen Dynan, Andreas Lehnert, Nellie Liang, and Eileen Mauskopf. 2009. “Designing Loan
Modifications to Address the Mortgage Crisis and the Making Home Affordable Program.”
Washington, DC: Division of Research & Statistics and Monetary Affairs, Federal Reserve Board.
Finance and Economics Discussion Series Paper 2009-43. October.
Making Home Affordable. 2011a. “Making Home Affordable Data File Summary.” Washington, DC: U.S.
Department of the Treasury and U.S. Department of Housing and Urban Development.
November.
———. 2011b. “Program Performance Report through August 2011.” Washington, DC: U.S. Department
of the Treasury and U.S. Department of Housing and Urban Development. October.
Mayer, Neil. 2011. “Foreclosures and the Federal Response.” Progressive Planning Magazine 189, 31-33.
Mayer, Neil, and Matt Piven. 2011. “National Foreclosure Mitigation Counseling Program Evaluation:
Issues in NFMC Counselors Use of the HAMP Program in Preventing Foreclosures.” Washington,
DC: Urban Institute. April.
83
Mayer, Neil, Peter A. Tatian, Kenneth Temkin, and Charles A. Calhoun. 2010. “National Foreclosure
Mitigation Counseling Program Evaluation: Preliminary Analysis of Program Effects.”
Washington, DC: Urban Institute. December.
———. 2011. “National Foreclosure Mitigation Counseling Program Evaluation: Final Report.”
Washington, DC: Urban Institute. December.
Norton, Edward C., Hua Wang, and Chunrong Ai. 2004. “Computing Interaction Effects and Standard
Errors in Logit and Probit Models.” Stata Journal 4(2), 154–67.
Office of the Comptroller of the Currency and Office of Thrift Supervision. 2011. OCC and OTS Mortgage
Metrics Report, Fourth Quarter 2010. Washington, DC: U.S. Department of the Treasury. March.
Stanley, Marcus. 2011. “Designed to Fail.” American Prospect 22(5), A10.
Taylor, Paul, Rakesh Kochhar, Richard Fry, Gabriel Velasco, and Seth Motel. 2011. “Wealth Gaps Rise to
Record Highs between Whites, Blacks and Hispanics.” Washington, DC: Pew Social and
Demographic Trends. July.
U.S. Government Accountability Office. 2010. “Troubled Asset Relief Program: Further Actions Needed
to Fully and Equitably Implement Foreclosure Mitigation Programs.” Washington, DC: U.S.
Government Accountability Office. June.
84
Appendix
Appendix Table 1. Significance Testing of Tabulations of Whether Various Groups Are Disadvantaged in
Eligibility for Trial Modifications.
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
Note: Three chi-squared tests for race were conducted, each of the three minority groups versus the
white group.
Appendix Table 2. Variable List and Results for Weighted and Unweighted Regressions for Receipt of
Trial Modifications.
(Continued on next page)
85
(Continued from previous page)
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
86
Appendix Table 3. Variable List and Results for Regression for Receipt of Permanent Modifications.
(Continued on next page)
87
(Continued from previous page)
Source: Authors’ computations, based on U.S. Treasury HAMP Public Database.
88
Download