lost ground

8/3/2019 Lost Ground

1/49

www.responsiblelending.org

Lost Ground, 2011:

Disparities in Mortgage Lendingand Foreclosures

Debbie Gruenstein Bocian,Wei Li, Carolina Reid

Center or Responsible Lending

Roberto G. QuerciaCenter or Community Capital,

University o North Carolina Chapel Hill

November 2011


2/49

Center for Responsible Lending 1

ExecutiveSummary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7

BackgroundandLiteratureReview . . . . . . . . . . . . . . . . . . . . . . . . .8

Data/Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

AnalysisandFindings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Appendix1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

Appendix2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

Endnotes......................................... 43

Table of ConTenTs


3/49

Acknowledgments

This report was made possible through the generous support o the Sears ConsumerProtection and Education Fund.

About this Research

In June 2010, the Center or Responsible Lending published Foreclosures by Race& Ethnicity: The Demographics of a Crisis, which showed the disparate impacts ooreclosures on Arican American and Latino homeowners and other communitieso color. That report relied on oreclosure rates rom a national dataset o mortgagesthat were matched with loan inormation collected under the Home MortgageDisclosure Act (HMDA).

Like the previous report, the current research examines the geographic anddemographic dimensions o oreclosures by relying on proprietary loan inormationmatched with HMDA data. Here we also incorporate a third data source and newloan-matching methodologies, allowing us to examine outcomes on a larger portiono the mortgage market, and more variables related to those outcomes (e.g., loantype), during the peak years o the subprime boom. Our results are based onapproximately 27 million matched loans, representing 63 percent o mortgagesincluded in HMDA data that were originated between 2004 and 2008.


4/49


eXeCUTIVe sUMMaRY

As the nation struggles through the ith year o theoreclosure crisis, there are no signs that the lood ohome losses in America will recede anytime soon.

Among the indings in this report, Lost Ground,2011,we show that at least 2.7 million households havealready lost their homes to oreclosure, and morestrikingly, that we are not even halway throughthe crisis.

Lost Ground, 2011 builds on the Center orResponsible Lendings longstanding eorts todocument the severity and demographic dimensionso the oreclosure epidemic.In 2006, CRL publishedLosing Ground, which projected subprime oreclosuresand the attendant costs to homeowners prior to

the collapse o the housing market.1 In 2010, wepublished Foreclosures by Race and Ethnicity: theDemographics of Crisis, which estimated completedoreclosures through 2009 and the disparate rates ooreclosure or dierent racial and ethnic groups.2Assessing the scope o the crisis remains daunting,since there is no single, nationwide source o inorma-tion on the number o oreclosures, the demographicso those aected, or the neighborhood distributiono oreclosed properties. In this report, we use anexpanded dataset to update our previous indings

and extend the scope o our analysis.

The report addresses three key questions. First, weconsider who has lost their home to oreclosure, andwho is still at risk. We look at both the race/ethnicityand income o borrowers, and explore how the impacto oreclosures on dierent socioeconomic and demo-graphic groups varies depending on where they live.Second, we look at what kind o mortgages dierentborrowers received to better understand the relation-ship between loan characteristics and deaults. Finally,we examine where the crisis has had the greatest

impact, assessing which areas and types o neighbor-hoods have been most aected.

Top-Line Findings

The nation is not even halfwaythrough the foreclosure crisis.

Among mortgages made between

2004 and 2008, 6.4 percent have

ended in oreclosure, and an

additional 8.3 percent are at

immediate, serious risk.

Foreclosure patterns are strongly

linked with patterns of risky lend-

ing. The oreclosure rates are con-

sistently worse or borrowers who

received high-risk loan products

that were aggressively marketed

beore the housing crash, such as

loans with prepayment penalties,

hybrid adjustable-rate mortgages

(ARMs), and option ARMs.

Foreclosure rates are highest in

neighborhoods where these loans

were concentrated.

The majority of people affected

by foreclosures have been white

families. However, borrowers of

color are more than twice as

likely to lose their home as

white households. These higher

rates relect the act that Arican

Americans and Latinos were

consistently more likely to receive

high-risk loan products, evenater accounting or income and

credit status.


5/49

Lost Ground, 2011: Disparities in Mortgage Lending and Foreclosures4

In brief, these are the key findings:

1. We are not even halfway through the foreclosure crisis. Among homeowners who receivedloans between 2004 and 2008, 2.7 million households, or 6.4 percent, had already lost theirhomes to oreclosure as o February 2011.3 Strikingly, an additional 8.3 percent3.6 millionhouseholdswere still at immediate, serious risk o losing their homes. Aected amilies span

all races, ethnicities, and income levels. It is notable that these serious delinquencies representonly a sub-set o likely oreclosures ahead, since they do not include oreclosures on loansoriginated outside our origination time rame or those that are not yet at imminent risk.

2. Loan characteristics and foreclosures are strongly linked. The study examines outcomes ondierent loan types and inds a pattern o higher oreclosures and delinquencies associated withspeciic mortgage characteristics. Loans originated by brokers, hybrid adjustable-rate mortgages(ARMs, such as 2/28s), option ARMs, loans with prepayment penalties, and loans with highinterest rates (a proxy or subprime mortgages) all have much higher rates o completed oreclo-sures and are more likely to be seriously delinquent.

Loan Status (as of February 2011) by Loan Feature (2004-2008 Originations)

Note:WedefinehybridandoptionARMsasloanswithanyoneofthefollowingcharacteristics:ARMswithinterestrateresetsoflessthanfiveyears,negativeamortization,orinterest-onlypaymentschedules.Higher-rateisdefinedasfirst-lienloansfor

whichtheannualpercentagerate(APR)was300basispointsormoreaboveTreasuryratesofcomparablematurity.

Completed Seriously Foreclosures Delinquent (%) (%)

Broker

BrokerOriginated 4.7 8.0

NotBrokerOriginated 2.7 5.6

HybridorOptionARM HybridorOptionARM 12.8 11.7

FixedRateorStandardARM 3.3 6.9

PrepaymentPenalty PrepaymentPenalty 14.7 14.3

NoPrepaymentPenalty 4.0 6.4

Higher-Rate Higher-Rate 15.6 15.3

NotHigher-Rate 4.6 6.9

3. Although the majority of affected borrowers have been white, African-American and Latinoborrowers are almost twice as likely to have been impacted by the crisis. Approximately onequarter o all Latino and Arican-American borrowers have lost their home to oreclosure or areseriously delinquent, compared to just under 12 percent or white borrowers. Asian borrowershave ared better as a whole than Latino and Arican-American borrowers, but they, too, havebeen disproportionately aected, especially in some metropolitan areas.

LoanFeature


6/49


Asian

PercentofLoans

Rates of Completed Foreclosures and Serious Delinquencies by Borrower Race and Ethnicity(2004-2008 Originations)

6.8

14.213.7

7.3

5.1

9.8

11.9

6.6

Non-Hispanic White

nCompleted Foreclosure n Seriously Delinquent (60+ or in Foreclosure)

African American Latino

30.0

25.0

20.0

15.0

10.0

5.0

0.0

4. Racial and ethnic differences in foreclosure rates persist even after accounting for differencesin borrower incomes. Racial and ethnic disparities in oreclosure rates cannot be explained byincome, since disparities persist even among higher-income groups. For example, approximately10 percent o higher-income Arican-American borrowers and 15 percent o higher-incomeLatino borrowers have lost their home to oreclosure, compared with 4.6 percent o higher-income non-Hispanic white borrowers. Overall, low- and moderate-income Arican Americansand middle- and higher-income Latinos have experienced the highest oreclosure rates.

5. Loan type and race and ethnicity are strongly linked. Arican Americans and Latinos were much more likelyto receive high interest rate (subprime) loans and loanswith eatures that are associated with higher oreclosures,speciically prepayment penalties and hybrid or optionARMs. These disparities were evident even comparingborrowers within the same credit score ranges. In act, thedisparities were especially pronounced or borrowers withhigher credit scores. For example, among borrowers with aFICO score o over 660 (indicating good credit), AricanAmericans and Latinos received a high interest rate loanmore than three times as oten as white borrowers.

6. Foreclosure rates by income groupings vary by housing markets. In areas o the country thathad weak or moderate housing price appreciation during the period leading up to the crisis,oreclosure rates are highest or low-income borrowers and lowest or higher-income borrowers.For example, low- and moderate-income borrowers have been most aected in cities such asDetroit, Cleveland, and St. Louis. However, in areas that had strong housing appreciation beorethe collapse, the opposite is true. In boom-market metropolitan areas located in Caliornia,

am rrr it fICo

cr vr 660 (iicti

crit), aric amric

lti rciv i

itrt rt mr

t tr tim t

it rrr.


7/49


Nevada and Arizona, oreclosure rates are highest among middle-and higher-income borrowers. These patterns are consistent withthe incidence o high-risk mortgages received by dierentgroups o borrowers across the dierent housing markets.

7. Impacts vary by neighborhood. Low- and moderate-income

neighborhoods and neighborhoods with high concentrations ominority residents have been hit especially hard by the oreclo-sure crisis. Nearly 25 percent o loans in low-income neighbor-hoods and 20 percent o loans in high-minority neighborhoodshave been oreclosed upon or are seriously delinquent, withsigniicant implications or the long-term economic viabilityo these communities.

The indings presented in this report suggest that we are not even halway through the oreclosurecrisis, as millions o additional amilies are still at risk o losing their home. Meanwhile, Americanso every demographic groupall incomes, races, and ethnicitieshave been aected. Our analysis

shows that non-Hispanic white and middle- and higher-income borrowers represent the vast majorityo people who have lost their homes. However, we also ind that people o color and lower-incomeborrowers and neighborhoods have been disproportionately aected.

Our study provides urther support or the key role played by loan products in driving oreclosures.Speciic populations that received higher-risk productsregardless o income and credit statuswere more likely to lose their homes. While some blame the subprime disaster on policies designed toexpand access to mortgage credit, such as the Community Reinvestment Act (CRA) and the aord-able housing goals o Fannie Mae and Freddie Mac (the government-sponsored enterprises, or GSEs),the acts undercut these claims.4 Rather, dangerous products, aggressive marketing, and poor loanunderwriting were major drivers o oreclosures in the subprime market.

The Dodd-Frank reorms, passed in July 2010, took the irst vital step by strengthening mortgageprotections, restricting the use o risky products and practices, and requiring lenders to consider eachborrowers ability to repay a loan. These new rules will certainly have a positive eect on the successo uture mortgages. Yet responding to todays battered housing market will require policy responseson a variety o other levels as well. In the short term, we need stronger measures to prevent addition-al oreclosures. Over the longer term, policymakers will need to consider how to rebuild the mortgagecredit market, recognizing not only the current challenges but also the broader historical barriers toaccess to credit in this country.

For decades, owning a home has been the most accessible way to build wealth and gain a oothold inthe middle class. Especially or lower-income amilies and middle-class borrowers o color, this crisisthreatens to undo decades o economic, social and educational progress. But in our eorts to stabilizethe housing market and prevent a uture crisis, we must not create an environment where qualiiedborrowers are denied access to reasonably-priced mortgages. Future reormswhether regulatory orlegislative in naturemust prevent unair and abusive lending practices while acilitating a stablesupply o mortgage credit or all qualiied borrowers.

nry 25 rct

i -icm

ir

20 rct

i i-mirityir v

rc u

r r t i ri

ut.


8/49


I. InTRodUCTIon

The United States is in the midst o an unprecedented oreclosure crisis. Since housing prices begantheir precipitous decline in early 2007, millions o homes have gone into oreclosure, and millionsmore remain in distress. The crisis has devastated amilies and communities across the country and isimpairing economic growth or the nation as a whole.

Owning a home has long been the most accessible way to build wealth and gain a oothold in themiddle class. Conversely, when amilies lose their homes, the resulting damage is multiaceted. First,there are the immediate inancial, social and emotional costs associated with oreclosure and housinginstability. Recent studies also indicate that oreclosures take a toll on the physical health o amilieswho lose their home.5 The negative impacts o housing insecurity can be particularly hard on chil-dren, inluencing everything rom behavior and cognitive development to academic perormance.6

Foreclosures also entail long-term consequences or asset building and inancial well-being. Familieswho lose a home cannot tap home equity to start a new business, pay or higher education or securetheir retirement. Loss o a home also removes a inancial cushion against unexpected inancial hard-

ships, such as job loss, divorce or medical expenses, and eliminates the main vehicle or transerringwealth inter-generationally.

Finally, oreclosures have ramiications that extend beyond the amilies who lose their homes.Communities with high concentrations o oreclosures lose tax revenue and incur the inancialand non-inancial costs o abandoned properties and neighborhood blight, while homeowners livingin close proximity to oreclosures suer loss o wealth through depreciated home values. The lowerlevels o wealth associated with oreclosures weaken consumer conidence and spending, therebyacting as a drag on the overall economic growth o the country.

Unortunately, while much has been reported on the oreclosure crisis, inormation on the socioeco-nomic and demographic characteristics o oreclosed borrowers and neighborhoods is diicult to

come by. The degree to which oreclosure data are reported varies tremendously, not just rom stateto state, but even rom county to county, and these data rarely include inormation on the race,ethnicity or income o the aected homeowners. As a result, we have limited knowledge o whohas been impacted by the crisis and which communities will experience the greatest negativeconsequences o oreclosures over the long term.

Better inormation on the impact o the oreclosure crisis can help policymakers crat more eectivepolicies to speed economic recovery, strengthen aected amilies and neighborhoods, and help toprevent additional deaults. More precise oreclosure analysis is also pertinent to current policydebates on down payment requirements and the uture o Fannie Mae and Freddie Mac, theGovernment Sponsored Enterprises (GSEs). The decisions made on those issues will dictate the

availability o credit or generations to come, and should be inormed by accurate inormation onthe impact o the oreclosure crisis.

In this report, we combine multiple data sources and, or the irst time, provide a comprehensiveassessment o the demographic distribution o the oreclosure crisis. We answer several importantquestions. First, we analyze who has lost their home to oreclosure, and who still aces the risk odeault. We look at both the race/ethnicity and income o borrowers, and explore how the impact ooreclosures on dierent socioeconomic and demographic groups varies depending on where theylive. Second, we look at disparities in what types o mortgages dierent borrowers received to better


9/49


understand the relationship between the targeting o unsustainable loan products and deaults.Finally, we examine where the crisis has had the greatest impact, assessing which metropolitanareas and types o neighborhoods have been most aected by oreclosures.

II. baCkgRoUnd and lITeRaTURe ReVIewThe past our years have been among the worst or the U.S. housing market. Millions o U.S.homeowners already have lost their homes to oreclosure and millions more are delinquent or inthe oreclosure process. Though the crisis has been more pronounced in particular states, such asCaliornia, Nevada and Florida, amilies across country have been aected, either through the directloss o their home or by the substantial loss o wealth resulting rom decreases in their home equity.

Although oreclosures are increasingly driven by high and persistent unemployment, the crisis hasits origins in the subprime mortgage market. While subprime loans were initially marketed as nicheproducts, during the latter hal o the 1990s and early 2000s, subprime lending exploded to becomea major driver o the U.S. housing market. From 1996 to 2006, the size o the subprime mortgage

market grew rom $97 billion to $640 billion.7 At the peak o the subprime market in 2006, 27 per-cent o all loan originations were subprime, including 49 percent and 39 percent o loans made toArican Americans and Latinos, respectively.8 As was true with the Alt-A market,9 during this timeperiod the subprime market became increasingly dominated by non-traditional loans, includinginterest-only loans, loans with limited or no documentation o income or assets, and loans with lowteaser rates that adjusted to much higher rates. These loans were oten made on the basis o weakunderwriting and without regard or borrowers ability to repay them.

To a large extent, the deterioration in lending standards was masked during the housing boom;because o the increase in property values, borrowers could reinance or sell their homes when theirrates reset or otherwise became unaordable. However, when house price growth slowed in late2006, the true hazards o these loans were exposed. In its inal report to Congress on the causes othe oreclosure crisis, the U.S. Department o Housing and Urban Development (HUD) conirmedthat, while sotening housing prices were clearly a triggering actor, the oreclosure crisis itsel wasundamentally the result o rapid growth in loans with a high risk o deaultdue both to the termso these loans and to loosening underwriting controls and standards.10 (See The Red Herrings.)

In addition to an overall deterioration in underwriting over this time period, subprime lendersbegan to target historically disadvantaged communities with high-cost and risky loan products. Adual mortgage market emerged, in which low-income borrowers and borrowers o color were servedprimarily by subprime lenders, while higher-income and white borrowers were served primarily byconventional lending institutions. Overwhelmingly, research on the dual mortgage market hasshown that, even ater controlling or dierences in borrower and neighborhood risk characteristics,

borrowers o color were more likely to receive subprime loans and/or loans with other risky producteatures. The studies have also consistently showed that subprime loans were more prevalent inlow-income and minority neighborhoods. (See The Dual Mortgage Market.)


10/49


The Red Herrings

One misperception that continues to circulate in the media is that the oreclosure crisis stemsrom government eorts to ensure air access to credit or low- and moderate-income consumers.Speciically, some observers have charged that the Community Reinvestment Act (CRA)11 and theaordable housing goals o the GSEs, Fannie Mae and Freddie Mac, precipitated the explosion o

risky lending during the subprime boom by requiring banks to make loans to unqualiied borrowers.

The acts simply do not support these accusations. Regarding the claims blaming the CRA, thereare at least three important rebuttals. First, CRA has been on the books or three decades, whilethe rapid growth o subprime and other non-prime loan securitization and the pervasive marketingo risky loan products did not occur until recent years.

Second, the predominant players in the subprime marketmortgage brokers, independent mortgagecompanies, and Wall Street investment bankswere not subject to CRA requirements at all. Inact, only six percent o higher-priced loans, a proxy or subprime, were subject to CRA, meaningthat they were extended by CRA-obligated lenders to lower-income borrowers or neighborhoodswithin their CRA assessment areas.12

Third, studies have shown that loans made to low- and moderate-income homebuyers as part obanks eorts to meet their CRA obligations have actually perormed better than the rest o thesubprime market. In an analysis o CRA-motivated loans sold to a community development inan-cial institution (CRLs ailiate Sel-Help), Ding, Quercia and Ratclie ound that the deault risk othese loans was much lower than subprime loans made to borrowers with similar income and creditrisk proiles.13 A study by the Federal Reserve Bank o San Francisco ound that CRA-eligible loansmade in Caliornia during the subprime boom were hal as likely to go into oreclosure as loansmade by independent mortgage companies, which were not subject to CRA requirements.14

Similarly, research shows no evidence that the GSEs aordable housing targets were a primarycause o the crisis. First o all, the GSEs could not purchase or securitize subprime mortgages

directly because such loans were outside the prescribed GSE guidelines. Subprime mortgage-backedsecurities were created by Wall Street irms, not the GSEs. Second, while the GSEs did purchasesubprime mortgage-backed securities as investments and oten received aordable housing goalcredits or those purchases, their share o such purchases was a raction o that o the private sector,and one that decreased over time, disproving the argument that the GSEs pushed the markettowards unsound, risky lending.15

In addition, the mortgages that accounted or most o the GSEs losses were not aordable housingloans but rather loans that generally went to higher-income amilies. At the end o 2010, amongloans acquired by the GSEs between 2005 and 2008, aordable housing targeted purchases com-prised less than eight percent o their 90-days delinquent portolio, only a small share o overalltroubled assets held by the GSEs.16 Most o the GSEs losses are tied to Alt-A mortgages, and those

loans did not count toward their aordable housing targets, and in act diluted them.17

In short, the assertion that the CRA or GSEs precipitated the oreclosure crisis, while a convenientnarrative or opponents o inancial regulation, is undermined by the acts. Indeed, recent researchby Robert Avery and Kenneth Brevoort at the Federal Reserve Board o Governors has urthershown that neither the CRA nor the GSEs caused excessive or less prudent lending in low- andmoderate-income neighborhoods.18


11/49


Though there is signiicant literature on the implications o the dual mortgage market or subprimelending outcomes among dierent demographic groups, the lack o national, publicly-available dataon oreclosures has made research on the demographic dimensions o the oreclosure crisis muchmore diicult. In 2010, the Center or Responsible Lending published the irst study to use nationaldata to estimate the impact o oreclosures on communities o color. The analysis, which applied theoreclosure rates o groups o loans rom Lender Processing Services to similar groups o loans inHMDA, estimated that 17 percent o Latino and 11 percent o Arican-American borrowers had

already lost or were at imminent risk o losing their home to oreclosure at the end o 2009, com-pared to seven percent or non-Hispanic white borrowers.19 However, an important limitation o thisstudy was its inability to capture oreclosure rate disparities attributable to dierences in the speciicloan products received by dierent populations o borrowers.

Other studies that have tried to answer the question o who has been aected by oreclosures havetended to ocus on smaller geographic areas. Focusing on Caliornia, Bocian, Smith, Green andLeonard ound that over hal o the states oreclosures were experienced by Latino amilies, arhigher than the Latino share o the state's total population or population o homeowners.20 In astudy in Minneapolis, Allen ound that oreign-born minority households were 1.7 times more likelyto go through a oreclosure than native-born white households, and that Latino households wereparticularly vulnerable i they had taken out a loan to purchase rather than reinance their home.21Anacker and Carr ound that high-income Arican Americans in the metropolitan D.C. area were36 percent more likely than whites to go into oreclosure, while Latino borrowers were 79 percentmore likely to be oreclosed on than their white counterparts.22 While these studies cannot begeneralized to the United States as a whole, they nevertheless provide compelling evidence thatthe oreclosure crisis has allen disproportionately on borrowers o color.

There is also growing evidence that oreclosures have been concentrated in particular types o neigh-borhoods. Researchers who have studied the crisis have identiied three key actors in the location ooreclosures. First, oreclosures have been heavily concentrated in suburban neighborhoods that sawconsiderable new construction and rapid housing price appreciation during the subprime lendingboom, particularly in states such as Florida, Caliornia, Nevada and Arizona. 23 These neighborhoodswere attractive to a wide range o moderate- and middle-income amilies hoping to become home-ownersincluding many Latinos, Asians and Arican Americansand who were priced out oneighborhoods elsewhere. Second, older, inner-city neighborhoodsparticularly those with highpercentages o low-income and minority residentshave also seen a disproportionate share ooreclosures.24 Finally, researchers have ound that metropolitan areas with high levels o segregationhave increased levels o oreclosure, even ater controlling or levels o minority populations andsubprime lending.25 These studies all provide strong evidence that low-income and minority neigh-borhoods are being hit hardest, not only by oreclosures, but by the attendant spillover eects ohigher crime, lower property values, and ractured social cohesion.26

This study builds on the existing research by creating a unique loan-level dataset with broad nationalcoverage. The comprehensiveness o this dataset, both in terms o data content and market coverage,

allows us to paint a more complete picture o the borrowers and neighborhoods that have beenaected by oreclosures. Speciically, we are able to look not just at the oreclosure rates or dierentdemographic groups, but also at the speciic loan products received by dierent populations. Byanalyzing these oreclosures and lending patterns in tandem, we are able to get a much richerunderstanding o the demographic dimensions o the oreclosure crisis. In addition, the large numbero loans in the merged dataset allows us to look at patterns o oreclosures and delinquencies at asmaller geographic scale, including disparities in outcomes at the state and metropolitan level.

The ollowing section describes the data and methodology used in this report. This is ollowed by adiscussion o our key indings and our conclusion.


12/49


The Dual Mortgage Market

Researchers trying to understand the disparate impact o oreclosures on lower-income andminority communities have increasingly ocused on the ailings o the dual mortgage market,in which lower-income and minority neighborhoods and borrowers were served primarily by

subprime lenders (even when they could have qualiied or a prime loan), and as a result weremore likely to receive subprime loans.27 Overall, these studies have ound that the race and/orethnicity o a borrower are strong determinants o the probability o receiving a subprime loan.28Even within the subprime market segment, there is evidence that Arican Americans and Latinoswere more likely to be steered into higher-priced loans than white borrowers. Bocian, Ernst andLi ound that Arican-American and Latino borrowers were about 30 percent more likely toreceive higher-cost subprime loans than similarly-situated white borrowers.29 Courchane alsoound that, ater controlling or the likelihood o receiving a subprime loan, Arican-Americanborrowers received higher APRs than white non-Hispanic borrowers.30 Research also has shownthat place matters, and that higher-priced and subprime loans were more requent in low-income and minority neighborhoods than in higher-income or predominantly non-Hispanicwhite neighborhoods.31

The existence o the dual mortgage market was made possible in part by the dominance omortgage brokers in the market. The majority o subprime loans were originated by brokers,who were largely unregulated at the ederal level (and the breadth and depth o state brokerregulations varies considerably). The dominance o broker originations in the subprime marketwas particularly dangerous due to their ability to charge yield-spread premiums (YSPs), whichwere inancial incentives paid by lenders to brokers to charge borrowers higher rates than theyqualiied or. As a result, studies have ound that, especially or borrowers o color, the mortgagebroker channel resulted in unequal loan outcomes, even ater controlling or borrower andneighborhood risk characteristics.32

In addition to brokers perverse incentives, the rise o the dual market was also acilitated byederal regulators who ailed to recognize the risks that subprime loans posed to consumers. TheOice o the Comptroller o the Currency and the Oice o Thrit Supervision both preemptedthe lenders they regulated rom stronger state anti-predatory lending laws,33 and the FederalReserve ailed to use its authority under the Home Ownership and Equity Protection Act(HOEPA) to expand protections or consumers taking out subprime loans until July 2008.34Without adequate consumer protections in place, borrowers in underserved communitiesespecially those with inadequate inormation or knowledge o the mortgage lending marketwere more vulnerable to poor underwriting and other predatory practices.35


13/49


III. daTa/MeThodologY

The analysis in this report relies on a loan-level dataset that matches data submitted by inancialinstitutions under the Home Mortgage Disclosure Act (HMDA) o 1975 with data rom twonational, proprietary datasets on loan perormance: Lender Processing Services (LPS) and BlackBox.HMDA is the largest publicly available database o U.S. home lending activity and includes

demographic inormation on the borrower, including race, ethnicity, and income, the dispositiono the loan (accepted or denied), and some inormation on the loan terms (or example, i it wasused or purchase or reinance, and whether or not the loan was higher-priced). However, HMDAdata is restricted to inormation collected at originationit contains no inormation on whether theloan is delinquent or in oreclosure.

To obtain inormation on loan perormance, we rely on data rom LPS and BlackBox, which arecomplementary, proprietary loan-level databases.36 LPS is collected rom loan servicers, whileBlackBox, which is exclusively comprised o loans that are in private-label securities, is collectedrom investor pools. While these databases do not contain any demographic inormation, they docontain data on loan perormance. The advantage o using both LPS and BlackBox in this analysis is

that it allows us to analyze a broader segment o the mortgage market than using either one by itsel.Importantly, the addition o the BlackBox data allows us to assess the outcomes or a greater share osubprime and Alt-A mortgagesthe risky loans that precipitated the crisissince these were morelikely to be securitized on the private secondary market.

To combine these three databases, we employ a probabilistic matching technique that links loanperormance data rom LPS and BlackBox to the loan origination data rom HMDA, which servesas our proxy or the universe o mortgage originations. That is, we allow loans in LPS and BlackBoxto be matched to loans in HMDA along loan characteristics that are common to all three datasets.HMDA loans that match to more than one LPS or BlackBox loan are weighted so that each HMDAloan is given a inal weight o one. For example, i one HMDA loan was matched to two loans inLPS and one loan in BlackBox, each match would be given a weight o one-third. The matching

algorithm is explained in greater detail in Appendix 1.

The advantage o using this probabilistic matching technique is that we do not lose the observationsthat would need to be dropped i we insisted on unique matches between the databases or i weexcluded loans that were in census tracts with overlapping zip codes. Relying on unique matches canintroduce bias: or example, loans with ewer matches tend to be prime loans, government loans,non-broker originated loans, and loans in non-boom housing markets. Thereore, ocusing only onloans with unique matches would bias the results towards loans with better perormance. While weundoubtedly have alse positive matches in our sample, the weights that are assigned to these loansdiminishes their eect and should not have any biased direction.

The inal sample used in our analysis contains approximately 27 million matched loans, representing63 percent o the HMDA universe o 42.9 million irst-lien owner-occupied mortgages originatedbetween 2004 and 2008. Loan perormance is observed through February 2011. We include bothpurchase and reinance loans in our analysis; we also consider the perormance o FHA/VA loans inaddition to conventional loans. Table 1.2 in Appendix 1 provides descriptive statistics or thematched HMDA/LPS/BlackBox dataset.


14/49


We assess two measures o loan perormance: (1) completed oreclosure, meaning that the propertyhas gone through the oreclosure process and has been lost to the homeowner; and (2) seriouslydelinquent, which reers to loans that were more than 60 days delinquent or in the oreclosureprocess as o February 2011. Both completed oreclosures and serious delinquency rates are calculatedas a percentage o all 2004-2008 originations.37

The comprehensive nature o our merged dataset as well as its national coverage allow us to provideimportant insights into the distribution o oreclosures to date. However, we are duly cautious aboutthe limitations o matching records across disparate sources o data, especially when the data areproprietary and are not reviewed and cleaned as part o a public reporting system. Furthermore, thepresence o alse positive matches in our sample, resulting rom the limited number o commonields between the datasets, highlights the need or a publicly-available dataset that includes inor-mation rom origination to termination on a representative sample o U.S. mortgage loans. Thistype o data would allow researchers to answer a much broader set o questions, leading to moreeective, evidence-based policymaking. In the absence o such data, the HMDA/LPS/BlackBoxmatch provides us with the opportunity to build on the existing literature and or the irst timedescribe the relationship between lending patterns and oreclosure rates or dierent demographic

and geographic groups.

It is also worth noting that our analysis does not cover all loans and/or borrowers aected by theoreclosure crisis. The analysis is based on individual loans included in the datasets, not individualborrowers. As a result, some borrowers are almost certainly in our dataset more than once, orexample, because they reinanced over the 2004-2008 time period. Unortunately, there is noway to link multiple loans associated with a single borrower.38 Second, the analysis is limited to2004-2008 originations by HMDA-reporting institutions. Thereore, loans originated outside thattime rame or by non-HMDA covered institutions will not be relected in our analysis. Our estimateso the number o serious delinquencies and oreclosures are thereore conservativethat is, an evengreater number o borrowers have been aected by the crisis.


15/49


IV. analYsIs and fIndIngs

In this section, we present our analysis and seven separate indings. We begin by calculating thetotal numbers o completed oreclosures and seriously delinquent loans and analyzing the peror-mance o dierent types o loan products. To assess which borrowers have been aected by theoreclosure crisis, we examine oreclosure and delinquency rates by race, ethnicity, and income.

We also look at loan perormance in dierent types o neighborhoods, characterizing the neighbor-hood by its demographic and socioeconomic composition. Throughout the analysis, we also look atthe incidence o dierent types o loans by demographic group and neighborhood type. While it isbeyond the scope o this report to develop a ull regression model to isolate the impact o variousactors on oreclosures, our analysis o loan products provides useul inormation or understandingoreclosure patterns and explaining disparities in oreclosure rates among dierent demographicgroups, neighborhoods and types o housing markets.

Finding #1: The foreclosure crisis is less than halfway over. Among borrowers who received

loans between 2004 and 2008, 6.4 percent have already lost their homes to foreclosure.

Strikingly, an additional 8.3 percent were still at serious risk of losing their homes. Affectedborrowers span all races, ethnicities, and income levels.

The impact o the oreclosure crisis has been ar-reaching. Based on the HMDA/LPS/BlackBoxdata, we estimate that more than 2.7 million homeowners who received loans between 2004 and2008 had already lost their homes to oreclosure as o February 2011, representing 6.4 percent othose borrowers. Another 8.3 percent o these loans, or 3.6 million households, were 60 or more daysdelinquent on their mortgage or in some stage o the oreclosure process, and at serious risk o losingtheir homes.39 (See Figure 1.) This suggests that, ive years into this crisis, we are not even halwaythrough it.

Figure 1: Rates of Completed Foreclosures and Serious Delinquencies (2004 2008 Originations)

Completed Foreclosures Seriously Delinquent

(60+ or in Foreclosure)

9.0

8.0

7.0

6.0

5.0

4.0

3.0

2.0

1.0

0.0

6.4

8.3

Per

centofLoans


16/49


As shown in Figures 2 and 3, oreclosures have reached into every U.S. community, aectingborrowers across racial, ethnic, and income lines. The vast majority o amilies who have losttheir homes have been white non-Hispanic40 and middle- or higher-income.41 As o February 2011,over 1.5 million white borrowers who received their loans between 2004 and 2008 have lost theirhomes to oreclosure, and over 2 million were 60 or more days delinquent or in the oreclosureprocess. Among middle- and higher-income households, 2 million have lost their homes, andanother 2.6 million homes were seriously delinquent. The large number o borrowers still indistress shows that the weak economy is taking its toll, as long durations o unemployment areaecting the ability o households to make their mortgage payments.

Note. Volumes are based on applying rates calculated from the merged sample to all 42.9 million 2004-2008

HMDA originations.

Non-Hispanic White

African American

Latino

Others

Asian

Figure 2: Number of Completed Foreclosures and Seriously Delinquent Loans by Race/Ethnicity(2004-2008 Originations)

500,0000 1,000,000 1,500,000 2,000,000 2,500,000

nCompleted Foreclosure nSeriously Delinquent (60+ or in Foreclosure)


17/49


Figure 3: Number of Completed Foreclosures and Seriously Delinquent Loans by Borrower Income(2004-2008 Originations)

Note: Volumes are based on applying rates calculated from the merged sample to all 42.9 million 2004-208 HMDA

originations. Low-income refers to borrowers at 50% below area median income (AMI), moderate-income refers to

borrowers at 50-80% of AMI, middle-income refers to borrowers at 80-120% of AMI, and higher-income refers to

borrowers at 120% or above AMI.

Low-Income

Moderate-Income

Middle- Income

Higher- Income

400,0000 800,000 1,200,000 1,600,000 2,000,000

n Completed Foreclosure nSeriously Delinquent (60+ or in Foreclosure)


18/49


Finding #2. Loans with certain product featureshigher interest rates, prepayment penalties,and hybrid or option adjustable interest rates, as well as loans originated by brokershave much

higher rates of completed foreclosures and are more likely to be seriously delinquent. However,as the crisis has continued, these types of loans make up a smaller share of all troubled loans.

Consistent with other research that has documented the importance o loan products andunderwriting in precipitating the oreclosure crisis, we ind that oreclosure rates vary by loanproduct type. Loans originated by mortgage brokers,42 adjustable rate mortgages (ARMs) withnon-traditional eatures (speciically, interest-only payments, negative amortization options or hybridrate structures with low initial teaser interest rates), loans with prepayment penalties,43 and loanswith higher interest rates44 are all associated with higher rates o completed oreclosures and seriousdelinquencies. (See Table 1.) The higher rates o deault or higher-rate/subprime loans are notsurprising, given that these products were supposed to serve higher-risk borrowers. However, thereis evidence that higher-rate loans were oten inappropriately targeted: as many as 61 percent oborrowers who received subprime loans had credit scores that would have enabled them to qualiy ora prime loan.45

Although risky loan eatures are strongly associated both with higher completed oreclosure rates

and higher rates o delinquencies, their signiicance in driving new delinquencies appears to bedecreasing. For example, loans with adjustable interest rates that had payment options or that beganwith a ixed lower interest rate that changed to an adjustable rate ater a year or twocomprised36.8 percent o seriously delinquent loans.46 While this is still much higher than their share o allloans (25.4 percent), it represents a sizable decrease rom their 56.7 percent share o completedoreclosures. Similar patterns are seen or loans with prepayment penalties or higher-interest rates.

There are two likely explanations or the declining inluence o loan characteristics on oreclosures.First, many o the riskiest loan products have already completed oreclosure and, thereore, theirtotal share o active loans has decreased. Second, as the crisis has evolved, unemployment and othereconomic actors, as opposed to product eatures, have increasingly been driving new delinquencies.

Table 1: Loan Status (as of February 2011) and Features (2004-2008 Originations)

LoanFeature

LoanStatus ProportionofLoanswithHigh-RiskFeatures byStatus

Completed Seriously Shareof Shareof Shareof Foreclosures Delinquent Originated Completed Serious Loans Foreclosures Delinquencies

Broker

BrokerOriginated 4.7 8.0 53.5 66.4 61.9

NotBroker 2.7 5.6 Originated

Hybridor HybridorOptionARM 12.8 11.7 25.4 56.7 36.8

OptionARM FixedRateor 3.3 6.9 StandardARM

Prepayment PrepaymentPenalty 14.7 14.3 16.2 41.5 30.2

Penalty NoPrepayment 4.0 6.4 Penalty

Higher-Rate Higher-Rate 15.6 15.3 16.8 40.6 30.7

NotHigher-Rate 4.6 6.9


19/49


While the data show that the oreclosure crisis has been widespread, our analysis also provides newand compelling evidence o the disproportionate impact o the oreclosure crisis on borrowers ocolor. Among Arican-American and Latino borrowers, 9.8 and 11.9 percent have lost their home tooreclosure, respectively.47 (See Figure 4.) Among Asian borrowers, 6.6 percent have lost their hometo oreclosure. For non-Hispanic whites, the oreclosure rate is 5 percentalthough this, too, isextremely high when compared to historical levels, it is approximately hal the rate o AricanAmericans and Latinos.

Equally troubling is the large share o borrowers o color who are still in distress. Among Latinoand Arican-American households, an additional 14 percent o loans were seriously delinquent,compared with 6.8 percent or non-Hispanic whites. It is unlikely that all o these delinquencies

will result in oreclosure, but given that the Mortgage Bankers Association data showed thatdelinquencies ticked upwards in the irst quarter o 2011, it is quite possible that nearly 25 percento loans to Arican-American and Latino borrowers will end in oreclosure during this crisis.

Figure 4: Rates of Completed Foreclosures and Serious Delinquencies by Race/Ethnicity (2004-2008 Originations)

PercentofLoans

6.8

14.2

13.7

7.3

9.8

11.9

6.6

Non-Hispanic White

n Completed Foreclosure nSeriously Delinquent (60+ or in Foreclosure)

African American Latino Asian

30.0

25.0

20.0

15.0

10.0

5.0

0.0

5.1

Finding #3: Although the majority of foreclosures have affected white borrowers,African-American and Latino borrowers are almost twice as likely to have lost theirhome to foreclosure as non-Hispanic whites.


20/49


When we examine oreclosure patterns at the metropolitan level, we ind signiicant variations inthe locations where borrowers o color have been the most aected. (See Table 2.1 in Appendix 2.)For example, more than one third o Arican-American borrowers in Detroit have already losttheir home to oreclosurea staggering statistic or a city that once had one o the highest Arican-American homeownership rates in the country.48 Arican Americans also have experienced veryhigh rates o oreclosure in areas that saw large price declines, such as Riverside-San Bernardino

and Phoenix.

The high oreclosure rates or Latinos result rom two distinct issues. First, like Arican Americans,Latinos were disproportionately targeted by subprime lenders. Thereore, within given geographicareas, Latinos oten have relatively higher oreclosure rates. In Washington, D.C., or example, theLatino oreclosure rate is three times the metro-wide average. Second, Latino borrowers tend to beconcentrated in cities in the sand states, such as Caliornia, Nevada and Florida, which have amongthe highest overall oreclosure rates in the country.

The metropolitan data also show that while Asian homeowners generally have ared better thanArican Americans and Latinos, in some markets the oreclosure rate or Asians is signiicantlyhigher than the metropolitan average, including in Minneapolis-St. Paul and Sacramento.

Finding #4: Racial and ethnic differences in foreclosure rates persist even after accountingfor differences in borrower incomes. Low- and moderate-income African Americans andmiddle- and high-income Latinos have experienced the highest foreclosure rates.

Importantly, we ind that the dierences in oreclosure rates among Arican Americans, Latinosand non-Hispanic whites are not simply the result o income disparities. As shown in Figure 5, theracial and ethnic disparities in oreclosure rates persist even within income groups. For example,the oreclosure rate or low- and moderate-income Arican Americans is approximately 1.8 times

higher than it is or low- and moderate-income non-Hispanic whites. The gap is smaller or Latinos,especially among low-income households, but even among low-income Latinos the oreclosure rate is1.2 times that o low-income whites. Low- and moderate-income Asian borrowers have a loweroreclosure rate than lower-income non-Hispanic whites, but the pattern is reversed amongmiddle- and higher-income Asians.

Particularly striking are the high rates o oreclosure among middle- and higher-income Arican-American and Latino borrowers. Approximately 10 percent o higher-income Arican Americansand 15 percent o higher-income Latinos have lost their home to oreclosure, compared with4.6 percent o higher-income non-Hispanic whites.


21/49


Interestingly, the data also show that the relationship between income and oreclosure rate variesacross dierent racial and ethnic groups. Among non-Hispanic whites, the completed oreclosurerate decreases as income goes up. For Arican Americans, the oreclosure rate is also highest orlow-income borrowers, but there is not much variation across income groups. However, the patternis reversed among Asian and Latino homeowners, with middle-and higher-income borrowers experi-encing the highest rates o oreclosure or these groups. This may be due to the act that Latino andAsian borrowers were concentrated in boom markets with higher-than-average house prices, wherethey were particularly vulnerable to being targeted or hybrid and option ARMs as well as other riskyloan products.49

Figure 5: Rates of Completed Foreclosures by Borrower Race/Ethnicity and Income (2004-2008 Originations)

Note:Low-incomereferstoborrowersat50%belowareamedianincome(AMI),moderate-incomereferstoborrowersat50-80%ofAMI,middle-incomereferstoborrowersat80-120%ofAMI,andhigher-incomereferstoborrowersat120%oraboveAMI.

PercentofLoans

16.0

14.0

12.0

10.0

8.0

6.0

4.0

2.0

0.0

nNon-Hispanic White

nAfrican American

nLatino

nAsian

Low-Income Moderate-Income Middle- Income Higher-Income

Borrower Income


22/49


While income cannot account or the racial and ethnic disparities in oreclosures, it appears thatloan product can explain at least part o it. We ind that borrowers in minority groups were muchmore likely to receive loans with product eatures associated with higher rates o oreclosures, specii-cally higher interest rates, hybrid and option ARMs, and prepayment penalties. Table 2 below showsthe disparities in the incidence o receiving a loan with a risky product eature (measured as a ratioo the incidence compared to non-Hispanic whites). Arican Americans and Latinos were 2.8 and2.2 times as likely, respectively, to receive a higher-rate loan as whites, which is consistent withprevious research on the distribution o higher-rate lending. We also ind that the incidence ohybrid and option ARMs and loans with prepayment penalties is much higher or borrowers o

color, including Asian borrowers.

Table 2: Incidence and Increased Incidence (Disparity Ratio) of High-Risk Loan Features by Race/Ethnicity(20042008 Originations)

IncidenceofHigh-RiskLoanFeatures

OneorMoreHigh-Risk

Feature

Higher-Rate

HybridorOptionARM

PrepaymentPenalty

OneorMoreHigh-Risk

Feature

Higher-Rate HybridorOptionARM

PrepaymentPenalty

DisparityRatio(versusNon-HispanicWhites)

Non-HispanicWhite 38.2 12.5 21.5 12.3 NA NA NA NA

AfricanAmerican 62.3 35.3 32.0 24.8 1.6 2.8 1.5 2.0

Latino 61.9 27.9 37.1 28.5 1.6 2.2 1.7 2.3

Asian 48.3 9.8 33.5 15.6 1.3 0.8 1.6 1.3

Note:WedefinehybridandoptionARMsasloanswithanyoneofthefollowingcharacteristics:ARMswithinterestrateresetsoflessthan5years,negativeamortization,orinterest-onlypaymentschedules.Higher-rateisdefinedasfirst-lienloansforwhichtheAPRspreadwas300basispointsormoreaboveTreasuriesofcomparablematurity.Thefollowingloanfeaturesaredefinedashighrisk:hybridandoptionARMs,prepaymentpenalties,orhigherinterestrates.

Particularly troublesome is the act that these high disparities are evident even within credit ranges.In act, as Table 3 below shows, as we move up the credit spectrum, the disparities in the incidenceo loans with risky eatures actually increases. For borrowers with a FICO score o over 660, or

example, 21.4 percent o Arican-American borrowers and 19.3 percent o Latino borrowersreceived a higher-rate loan, 3.5 and 3.1 times the incidence o white borrowers. This increaseddisparity is much higher than among borrowers with FICO scores o less than 580, where theincidence o higher-rate lending among Arican Americans and Latinos was 1.2 and 1.1 timesthat o whites, respectively. We also ind that, while Asian borrowers were no more likely thannon-Hispanic whites to receive a higher-rate loan, they did receive a greater share o loans withprepayment penalties and loans with non-traditional adjustable interest rate eatures.

Finding #5: In addition to receiving a higher proportion of higher-rate loans, African

Americans and Latinos also were much more likely to receive loans with other risky features,such as hybrid and option ARMs and prepayment penalties. Disparities in the incidence ofthese features are evident across all segments of the credit spectrum, and are particularlypronounced for minority borrowers who received a loan from a mortgage broker.


23/49


Non-HispanicWhites 50.2 52.5 52.0 NA NA NA

FICO


24/49


Table 4: Incidence and Disparities of High-Risk Loan Products, by Borrower FICO and Mortgage Broker Channel(2004-2008 Originations)

IncidenceofLoanswithOneorMoreHigh-RiskLoanFeature

DisparityRatio(Ratio:BrokervsNon-Broker)

Note:Thefollowingloanfeaturesaredefinedashigh-risk:hybridandoptionARMs,prepaymentpenalties,orhigherinterestrates.

Broker Non-Broker

Non-HispanicWhites 50.0 48.4 1.03

FICO


25/49


Figure 6. Rates of Completed Foreclosures and Serious Delinquencies by Borrower Income(2004 2008 Originations)

8.68.7 8.5 8.2

7.36.6 6.2 6.4

18.0

16.0

14.0

12.0

10.0

8.0

6.0

4.0

2.0

0.0

Low-Income Moderate-Income

Borrower Income

Middle- Income Higher-Income

PercentofLoans

Note:Low-incomereferstoborrowersat50%belowareamedianincome(AMI),moderate-incomereferstoborrowersat50-80%ofAMI,middle-incomereferstoborrowersat80-120%ofAMI,andhigher-incomereferstoborrowersat120%oraboveAMI.Foralistingofstateswithineachhousingmarkettype,pleaseseeTable1.3inAppendix1.

Figure 7: Rates of Completed Foreclosures by Borrower Income and Housing Market Type, 2004-2008 Originations

12.0

10.0

8.0

6.0

4.0

2.0

0.0

nLow-Income

nModerate-Income

nMiddle-Income

nHigher- Income

Housing Market Type

PercentofLoans

Weak Stable Moderate Growth Boom

Borrower Income

nCompleted Foreclosure n Seriously Delinquent (60+ or in Foreclosure)

Note:Low-incomereferstoborrowersat50%belowareamedianincome(AMI),moderate-incomereferstoborrowersat50-80%ofAMI,middle-incomereferstoborrowersat80-120%ofAMI,andhigher-incomereferstoborrowersat120%oraboveAMI.


26/49


While overall the dierences in delinquencies and deaults between income groups seem modest,separating the analysis by dierent housing market types yields more striking results. In Figure 7,we look at the distribution o oreclosures by borrower income and housing market type. For thisanalysis, we categorize regional housing markets as either weak, stable, moderate, or booming basedon state-level house price appreciation trends between 2000 and 2005.51

We ind that in areas with weak house price appreciation, the impact o oreclosures has allendisproportionately on lower-income borrowers. In these communities, the oreclosure rate amonglow-income borrowers is 2.8 times that o higher-income borrowers. Stable and moderate growthmarkets also have oreclosure rates that decline as incomes rise. This contrasts with the experienceo borrowers in boom market areas, where middle- and higher-income borrowers have experiencedthe highest oreclosure rates. While these borrowers may have had higher incomes (with a mediano $61,000 or middle-income borrowers and $108,000 or higher-income borrowers), the extremelyhigh cost o housing in these boom markets, even or modest homes52 suggests that the majority othese borrowers were not the very wealthy buying mansions, but rather working amilies aspiring tohomeownership and the middle class.

Table 2.2 in Appendix 2 presents these same statistics or selected metropolitan areas. As thenational statistics suggest, low- and moderate-income borrowers have been most aected incities such as Detroit, Cleveland, and St. Louis. By contrast, in boom-market metropolitan areaslocated in Caliornia, Nevada and Arizona, oreclosure rates are highest among middle- andhigher-income borrowers.

To better understand these patterns in loan outcomes, we explored dierences in the incidenceo high-risk products received by dierent income groups in the dierent types o housing markets.Figure 8 presents data, by borrower income and housing market type, on the incidence o loanswith one o any o the ollowing risky eatures: hybrid or option ARMs, prepayment penalties,or higher interest rates. The pattern mirrors that o Figure 7, providing urther evidence o a linkbetween loan type and oreclosure rates. While in weak market areas, low-income and moderate-

income amilies have the highest incidence o mortgages with at least one high-risk eature, thepattern is reversed in boom markets. In act, a slightly smaller percentage o low-income borrowersin boom areas (45.3 percent) received a loan with any risky eature than low-income borrowers inweak markets (47.3 percent).


27/49


Figure 8: Percent of Loans with One or More High-Risk Loan Feature, By Borrower Income and Housing Market Type(2004-2008 Originations)

70.0

60.0

50.0

40.0

30.0

20.0

10.0

0.0

nLow-Income

nModerate-Income

nMiddle-Income

nHigher- Income

Housing Market Type

PercentofLoans

Weak Stable Moderate Growth Boom

Borrower Income

Note:Low-incomereferstoborrowersat50%belowareamedianincome(AMI),moderate-incomereferstoborrowersat50-80%ofAMI,middle-incomereferstoborrowersat80-120%ofAMI,andhigher-incomereferstoborrowersat120%oraboveAMI.Thefollow-ingloanfeaturesaredefinedashigh-risk:hybridandoptionARMs,prepaymentpenalties,orhigherinterestrates.

Drilling down to the speciic types o high-risk lending by market yields additional inormation.(See Table 5.) Low- and moderate-income borrowers were more likely to get higher-rate loansthe only marker or subprime loans in the HMDA dataacross housing market types, but this isnot the case or loans with other risky eatures. The pattern or hybrid and option ARMs is especially

striking: while only 20 percent o lower-income borrowers in boom markets received a hybrid oroption ARM, nearly 42 percent o higher-income borrowers did. Also interesting is the act thatin boom housing markets, a smaller share o low-income borrowers received higher-rate loans:19.5 percent compared with 28 percent in weak housing market areas.

While still exploratory, this analysis suggests that there were important interactions between housingmarkets and how dierent loan product types were marketed that inluenced who received riskyloans and, consequently, the oreclosure rates or dierent demographic groups in dierent partso the country.


28/49


Note:Low-incomereferstoborrowersat50%belowareamedianincome(AMI),moderate-incomereferstoborrowersat50-80%ofAMI,middle-incomereferstoborrowersat80-120%ofAMI,andhigher-incomereferstoborrowersat120%oraboveAMI.Thefollowingloanfeaturesaredefinedashighrisk:hybridandoptionARMs,prepaymentpenalties,orhigherinterestrates.

Table 5. Incidence of High-Risk Loan Features by Borrower Income and Market Type (2004-2008 Originations)

IncidenceofHigh-RiskLoanFeatures

HousingMarketType

BorrowerIncome OneorMoreHigh-RiskFeature

Higher-Rate HybridandOptionARM

PrepaymentPenalty

Low-Income 47.3 28.2 17.6 17.4

Weak

Moderate-Income 43.7 24.4 17.7 16.0

Middle-Income 38.6 19.9 16.4 13.5

Higher-Income 31.1 12.1 15.2 9.4

Low-Income 47.9 26.0 20.4 13.9

Stable


Middle-Income 42.1 18.9 21.1 10.6

Higher-Income 36.2 11.5 20.9 7.7

Low-Income 40.8 20.2 16.8 15.6

ModerateGrowth Moderate-Income 39.5 17.8 19.4 15.7

Middle-Income 37.3 15.5 19.9 14.3

Higher-Income 36.2 10.9 22.1 11.4

Low-Income 45.3 19.5 20.1 18.0

Boom


Middle-Income 52.6 19.2 31.8 22.3

Higher-Income 58.3 15.3 41.7 23.4


29/49


Higher-Income

Neighborhood

In addition to looking at the impact o the oreclosure crisis by borrower demographics, wealso examined which types o neighborhoods have seen the most delinquencies and deaults.Neighborhoods with high concentrations o oreclosures lose tax revenue and incur the inancial

and non-inancial costs o abandoned properties and neighborhood blight, and homeownersliving in close proximity to oreclosures typically lose signiicant wealth as a result o depreciatedhome values.53

To what extent have oreclosures allen disproportionately on low-income and minority neighbor-hoods? To answer this question, we group neighborhoods by census tract into our income categories(low, moderate, middle and higher)54 and our minority status quartiles, based on the percentage oresidents that are not non-Hispanic white.55

The indings rom this analysis are presented in Figures 9 and 10, and show the impact o theoreclosure crisis on lower-income neighborhoods and neighborhoods o color. In low-income neigh-borhoods, the oreclosure rate is 2.6 times that o higher-income neighborhoods, with approximately

12 percent o loans in oreclosure and another 12 percent seriously delinquent. Similarly, the oreclo-sure rate in neighborhoods with the highest concentration o minority residents is 1.7 times higherthan in neighborhoods with the lowest concentration o minorities. In high-minority neighborhoods,8.7 percent o loans taken out between 2004 and 2008 have resulted in oreclosure, and another10.8 percent are at risk o deault.

Figure 9: Rates of Completed Foreclosures and Serious Delinquencies by Neighborhood Income(2004-2008 Originations)

12.3

11.4

8.8

6.512.2

6.6 6.6

4.6

n Completed Foreclosure n Seriously Delinquent (60+ or in Foreclosure)

Low-Income

Neighborhood

Moderate-Income

Neighborhood

Middle- Income

Neighborhood

PercentofLoans

30.0

25.0

20.0

15.0

10.0

5.0

0.0

Note:Low-incomeneighborhoodsarethoseforwhichthemedianincomeforthecensustractislessthan50percentofthemetropolitanstatisticalarea(MSA)medianincome;moderate-incomeatleast50percentandlessthan80percentoftheMSAmedianincome;middle-incomeatleast80percentandlessthan120percentoftheMSAmedianincome;andhigher-incomeatleast120percentofMSAmedianincome.

Finding #7. Across the country, low- and moderate-income neighborhoods and neighborhoodswith high concentrations o minorities have been hit especially hard by the oreclosure crisis.


30/49


Percent Minority-

Lowest Quartile

Figure 10: Completed Foreclosures and Seriously Delinquent Loans by Neighborhood Minority Status(2004-2008 Originations)

10.8

8.3

7.57.1

8.7

6.5 5.65.0

n Completed Foreclosure n Seriously Delinquent (60+ or in Foreclosure)

Percent Minority-

Highest Quartile

Percent Minority-

3rd Quartile

Percent Minority-

2nd Quartile

PercentofLoans

25.0

20.0

15.0

10.0

5.0

0.0

Once again, we can see the strong link between mortgages with risky product eatures andoreclosures. As shown in Table 6 below, low-income neighborhoods had the highest incidence omortgages that were hardest to sustain (68.1 percent), whereas the incidence o higher-risk mortgag-es declines to 41.1 percent or higher-income neighborhoods. Similarly, neighborhoods with thehighest concentrations o minorities experienced the highest proportion o higher-risk loans.Fity-our percent o the loans in the neighborhoods with the highest proportion o minorityresidents had at least one high-risk eature, compared to 40.4 percent o those in neighborhoodswith the lowest proportion o minorities. These indings build on existing research showing thatthese areas were targeted by subprime lenders during the boom56 and documents the consequenceso weak consumer protection during this time period and its disproportionate impact on low-incomeneighborhoods and neighborhoods o color.

Low-Income 68.1 HighestQuartile 53.9

Moderate-Income 56.9 3rdQuartile 44.9

Middle-Income 44.1 2ndQuartile 41.8

Higher-Income 41.1 LowestQuartile 40.4

Table 6: Incidence of High-Risk Loans by Neighborhood Type (2004-2008 Originations)

NeighborhoodMinorityStatusCategory

IncidenceofLoanswithOneorMoreHigh-RiskFeature

IncidenceofLoanswithOneorMoreHigh-RiskFeature

NeighborhoodIncomeCategory

Note:Thefollowingloanfeaturesaredefinedashighrisk:hybridandoptionARMs,prepaymentpenalties,orhigherinterestrates.


31/49


The implications o these neighborhood oreclosure patterns will be elt by metropolitan areasthroughout the country. (See Tables 2.3 and 2.4 in Appendix 2.) In Clevelands low-incomeneighborhoods, a quarter o borrowers have already lost their home to oreclosure, threateningthe viability o these neighborhoods. While much has been written about Cleveland, low-incomeneighborhoods in other cities such as Atlanta, Las Vegas and Minneapolis-St. Paul also havecompleted oreclosure rates o over 20 percent. Such high levels o concentrated oreclosures will

place a signiicant burden on these neighborhoods and also the wider communities, which, withoutsubstantial interventions, will almost certainly suer reduced revenues or vital city services, higherrates o crime, and myriad other adverse eects. Furthermore, in once ast-growing metropolitanareas such as Phoenix and Riverside/San Bernardino, the more limited governmental and non-proitinrastructure and experience related to vacant properties may make it especially diicult to respondto the inventory o bank-owned homes.57


32/49


V. ConClUsIon

The United States is in the midst o an unprecedented oreclosure crisis, one that shows little signo abating. In act, our indings suggest that we are not even halway through the crisis, as millionso amilies remain at risk o oreclosure. Our report shows that Americans o every demographicgroupall incomes, races, and ethnicitieshave been aected, urther undermining the notionthat policies used to promote homeownership in disadvantaged communities are to blame or thecrisis. Such narratives, while providing convenient scapegoats or opponents o inancial regulation,are simply belied by the data.

Nevertheless, our analysis also shows that low-income and minority borrowers and neighborhoodshave been disproportionately impacted by oreclosures and that this relects the higher incidence ohigher-risk products received by these groups. For communities o color in particular, the oreclosurecrisis will have long-term consequences or wealth and opportunity. A recent study by the PewResearch Center ound that the wealth gap between whites and Arican Americans is at a historicalhigh and attributes this increase in inequality, at least in part, to the housing crisis. The averagewhite household now has twenty times more wealth ($113,149) than the average Arican-American

($5,667) and Latino ($6,325) household.58

The ongoing oreclosure crisis will only exacerbate thisgrowing inequality, as more Arican-American and Latino households will see their assets erode,either directly through oreclosure or through the loss o home equity resulting rom concentratedoreclosures in their neighborhoods.

Adequately addressing the crisis will require policy responses on a variety o levels. In the short term,preventing additional oreclosures must remain at the top o the policy agenda. The policy choiceswe make now regarding loan servicing and loan modiications have the potential to inluence boththe duration and severity o the crisis.

Over the longer term, policymakers will need to consider how to rebuild the mortgage credit market,recognizing not only the current challenges but also the broader historical context o access to credit

in this country. For decades, the predominant air lending issue was basic access to mortgage creditor groups excluded rom the economic mainstream. Whether through overt redlining or other ormso discriminatory underwriting, lower-income and minority borrowers aced barriers to obtaininghome loans. In response to discriminatory practices, in 1975 Congress enacted the Home MortgageDisclosure Act (HMDA). Shortly thereater, Congress passed the Community Reinvestment Act(CRA) to encourage lending in previously ignored communities, and amended the Equal CreditOpportunity Act (ECOA) to prohibit lending discrimination based on race and national origin,among other criteria. In the early 1990s, President George H. W. Bush signed the Housing andCommunity Development Act o 1992, which established aordable housing goals or the GSEs.While these policies did not by any means eliminate discrimination in home mortgage lending orerase homeownership disparities among demographic groups, they established a regulatory ramework

that recognized the importance o air access to mortgage credit.

Over the past two decades, the revolution in mortgage inance, including the widespread adoption ocredit scoring, automatic underwriting, and innovations in the secondary market, drastically alteredthe landscape o the mortgage market. Access to credit was no longer a major threat to lower-incomeand minority borrowers, but the terms o credit became a new, insidious issue. The explosion o exot-ic loan products and the act that they were so requently underwritten without regard or borrowersability to repay them, undermined homeownership by moving away rom proven wealth-building


33/49


loan eatures, such as ull amortization and ixed interest rates.59 In this new environment, thepolicies designed to expand access to mortgage credit, including the CRA and the GSEs aordablehousing goals, became less relevant, with both losing substantial market share at the peak o thebubble.60 Instead, a dual mortgage market arose where lower-income and minority borrowers andcommunities were served primarily by mortgage brokers and independent mortgage companies oper-ating largely outside existing consumer protections. These lenders had inancial incentives to steer

large numbers o borrowers into riskier mortgage products without regard or their ability to repay.

The negative consequences o this subprime boom have become glaringly apparent, and our researchshows that the oreclosure crisis has allen disproportionately on low-income and minority borrowersand neighborhoods. However, in our eort to prevent a similar crisis rom occurring again, it iscritical that we not return to an environment where lower-income and minority amilies are deniedaccess to reasonably-priced mortgages. Stability in the mortgage market should be a top priority inederal housing inance policy, but so, too, should airness and equal access to credit. Thereore,reormswhether regulatory or legislative in naturemust prevent unair and abusive lendingpractices like those that caused the crisis. At the same time, reorms must recognize the importanceo homeownership as a wealth-building mechanism and oothold into the middle class, and acilitate

a stable supply o mortgage credit or all qualiied borrowers.


34/49


appendIX 1. daTa and MeThodologY

The three databases in the analysis are Home Mortgage Disclosure Act (HMDA), BlackBox (BBx)and Lender Processing Services (LPS). HMDA is considered to be the universe o loans, as ederallaw requires that the majority o mortgage originations be reported in HMDA.61 To obtain inorma-tion on loan perormance, CRL purchased data rom LPS and BlackBox, which are complementary,proprietary loan-level databases. BlackBox data covers over 90 percent o non-agency pools, includ-ing jumbo, subprime and Alt-A loans, and includes approximately 7,600 deals, over 5,700 o whichare active.62 LPS, which uses loan-level data collected rom servicers, is not limited to private-labelsecurities and is requently used by researchers studying mortgage perormance. We estimate thatLPS has coverage equal to 66 percent o the irst-lien mortgages reported to ederal regulators inHMDA data rom 2005 through 2008.

Because we consider HMDA to be our universe o loans, our irst step in merging these databases wasto assign each HMDA loan a unique identiier. Next, we match loans rom HMDA to loans in LPSby geography and loan amount. Because HMDA reports data by census tract and LPS by zip code,we created a cross-walk ile using spatial location within ARCGIS, a mapping sotware program.

To create the crosswalk, or each census tract, we estimate the share o housing units that are in theoverlapping zip code. These become our geographic weights, which we use to represent the probabili-ty that a loan is, in act, in that zip code. For loans that are in tracts that overlap multiple zip codes,we create additional HMDA records or each potential zip code that the loan could be in. The resulto this step in matching HMDA to LPS is, in essence, a Cartesian product, where every HMDArecord with a given loan amount-zip code combination is allowed to match to every LPS recordwith the same combination. We then ilter out those matches where other common ields betweenthe two databases (e.g., loan purpose and loan type63) are inconsistent with each other.

Ater matching HMDA to LPS in this manner, we separately match HMDA to BlackBox using thesame methodology. The resulting two combined datasets (i.e. HMDA-LPS and HMDA-BlackBox)are then appended, and a matching weight is assigned to each loan. The matching weight is given

based on the number o times a unique HMDA loan was matched to loans in LPS and BlackBox.64A inal weight is calculated by multiplying the geographic weight by the matching weight. Formally,the weights are developed as ollows. or a HMDA loan A in census tract X, suppose there are totalo n LPS/BBx loans matched to A. For the ith loan matched to HMDA loan A, suppose it is romzip code Y. X and Y overlap at Z, as illustrated in Figure A1.


35/49


The advantage o using this probabilistic matching technique is that we do not lose the observationsthat would need to be dropped i we insisted on unique matches between the databases or onexcluding loans that were in census tracts with overlapping zip codes. Dropping observations wouldincrease the likelihood o introducing bias. While we undoubtedly have alse positive matches in oursample, the weights that are assigned to these loans diminish their eect and should not have anybiased direction.

LetX,YandZalsodenotetheirareasize.TheprobabilitythattheHMDAloanAisinZisgivenby

assumingthatAhasanequalchanceofbeinglocatedanywhereinX.Similarly,theprobabilitythatthe ithloanisinZisgivenby

ThejointprobabilitythatboththeHMDAloanAandthe ithloanofLPS/BBxareinZisgivenby

TheprobabilitythatHMDAloanAandthe ithloanarethetruematchisgivenby

ForHMDAloanA,anysupplementalinformationobtainedfromthe ithloanofLPS/BBxisweightedbyQiA.Thisfinalweightisusedintheanalysis.


36/49


Table 1.2 presents the descriptive statistics or the data in our sample. Overall, the data show that

the merged sample is largely representative o the mortgage market between 2004 and 2008. Thesample shows a distribution o loan purpose (51 percent are reinance loans and 49 percent are pur-chase loans), that is consistent with the large number o reinances that occurred over this time peri-od. Our racial distribution is also consistent with HMDA data: non-Hispanic whites comprise thelargest share o borrowers, at 61.3 percent, ollowed by Latinos at 10.7 percent, Arican Americansat 8.1 percent, and Asians at 4.5 percent. Approximately 22.4 percent o the loans in the sample arehigher-priced as indicated in the HMDA data. O all the loans in the sample, 42.8 percent werepaid o by February 2011, 6.4 percent had completed oreclosure, and 8.3 percent were 60+ daysdelinquent or in the oreclosure process.

Table 1.1. Distribution of Number of LPS or BlackBox loans matched per HMDA loan

Note:Theabovedistributiononlyaccountsforthemergedsample.36.7%ofHMDAloansdidnotmatchtoanyBlackBoxorLPSloan.TherearemultiplereasonswhyaHMDAloanmightnotmatchanyloaninBlackBoxorLPS,includingHMDAsmorecompletemarketcoverage,missinginformationonmatchingvariables,andreportingerrorsineitherHMDAorLPS/BBx.

1 79.83 74.50

2 14.39 17.22

3 3.72 4.83

4+ 2.07 3.45

Table 1.1 shows our match rate or the three datasets. O HMDA loans that matched, approximately75 percent were uniquely matched, meaning that there was only 1 loan in either LPS or BlackBoxthat successully matched to a HMDA loan with the same characteristics. Around 15 percent hadtwo matches to each HMDA Loan, and between 3.7 to 4.8 percent o the loans had 3 matchingloans. Less than 3.5 percent o HMDA loans were matched to more than 3 loans in BlackBox or

LPS.65

We do ind dierences between loans that were uniquely matched to one HMDA loan andthose that have multiple matches, conirming that limiting the sample to uniquely matched loanswould introduce some bias into the results. Loans with more than one match are slightly morelikely to be conventional loans (as opposed to FHA/VA) and used or reinance. They are also morelikely to be loans originated to borrowers o color (32.4 percent or multiple matches compared with26.3 percent or unique matches), and more likely to be in oreclosure or at imminent risk o deault(15.3 percent compared to 14.4 percent). As a result, we believe that our probabilistic weightingapproach leads to less biased results than ocusing solely on unique matches.

#ofLPSorBlackBoxLoans

matchedperHMDAloan

%ofallHMDAloans

matchedwithBlackBox

%ofallHMDAloans

matchedwithLPS


37/49


Table 1.2. Descriptive Statistics of Merged HMDA/LPS/BlackBox Dataset

OriginationYear

FullSample 2004 2005 2006 2007 2008

TotalNumberof LoansinSample 27,117,189* 7,175,942 7,221,122 4,904,838 4,612,297 3,202,990

LoanPurpose(%)

Purchase 48.4 44.5 47.9 52.1 50.3 50.2

Refinance 51.0 54.9 51.4 47.2 49.1 49.3

Race/EthnicityofBorrower(%)

Non-HispanicWhite 61.33 57.47 61.20 60.29 63.37 68.93

AfricanAmerican 8.09 7.59 8.35 9.38 8.23 6.48

Latino 10.71 9.92 11.66 12.76 10.48 7.54

Asian 4.54 4.79 4.52 4.31 4.37 4.62

BorrowerIncome(%)

Low-Income

(lessthan50%AMI) 5.79 6.14 5.79 5.62 5.24 6.08Moderate-Income (50-80%AMI) 18.87 18.99 19.10 18.16 18.36 19.86

Middle-Income 26.10 25.15 26.94 25.84 25.78 27.17(80-120%AMI)

Higher-Income (morethan120%AMI) 42.61 36.75 43.10 45.76 46.85 43.68

LoanCharacteristics

ConventionalPrime 74.83 73.18 73.26 75.26 83.93 68.35

Alt-A 3.55 4.39 4.99 4.45 1.53 0.00

FHA/VA 12.97 11.00 7.39 7.34 10.23 45.55

Higher-PricedLoan(HMDA) 22.43 11.37 22.66 24.71 14.76 7.34AverageFICO 701 698 692 689 704 723

AverageLTV 77 76 77 78 78 76

AverageLoanAmount $241,915 $213,350 $237,712 $249,575 $259,150 $249,788

LoanPerformance(%)

PaidOff 42.8 56.6 43.3 36.0 32.2 36.4

CompletedForeclosures 6.4 4.0 8.1 10.9 6.2 1.6

SeriouslyDelinquent(60+DelinquentorinForeclosure) 8.3 4.0 8.0 12.9 12.4 6.0

*Note:Thetotalnumberoffirst-lienowner-occupiedHMDAloansforyears2004-2008was42,871,874(36.7%ofHMDAloans

didnotmatchtoanyLPSorBlackBoxloan.ThematchedandunmatchedHMDAloanslooksimilaracrossallimportantdimensionsand,therefore,wescaleuptoreflectall42.9millionloanswhencalculatingthenumberofloansthathaveforeclosedorthatareseriouslydelinquent.


38/49


Our classifcation o housing market types is based on 2000-2005 state-level housing price indicesrom the Federal Housing Finance Agency. Table 1.3 lists the states included in each category.

Table 1.3: Classification of Housing Market Types

WeakMarketStates Indiana,Iowa,Kansas,Kentucky,Michigan,Mississippi,Nebraska,NorthCarolina,Ohio,Oklahoma,Tennessee,Texas

StableMarketStates Alabama,Arkansas,Colorado,Georgia,Illinois,Louisiana,Missouri,NorthDakota,SouthCarolina,SouthDakota,Utah,WestVirginia,Wisconsin

ModerateGrowthMarketStates Alaska,Connecticut,Idaho,Maine,Minnesota,Montana,NewMexico,NewYork,Oregon,Pennsylvania,Vermont,Washington,Wyoming

BoomMarketStates Arizona,California,Delaware,DistrictofColumbia,Florida,Hawaii,Maryland,Massachusetts,Nevada,NewHampshire,NewJersey,RhodeIsland,Virginia


39/49


appendIX 2: Tables of CoMpleTed foReClosURe and seRIoUs delInqUenCY RaTesfoR seleCTed MeTRopolITan aReas


40/49


Total

Total

African

American

African

American

African

American

Latino

Latino

Latino

Asian

Asian

Asian

No

n-Hispanic

White

Non-Hispanic

White

Non-Hispanic

W

hite

Detroit,MI

54.7

28.0

2.3

2.2

22.4

16.0

36.0

26.0

10.9

11.2

8.6

16.6

13.0

5.7

LasVegas,NV

51.7

5.8

17.3

8.2

16.1

14.3

16.4

20.3

20.2

14.7

13.5

15.2

18.6

15.6

Riverside-

SanBernardino,CA

36.1

5.7

30.4

5.5

13.6

11.0

14.5

17.0

15.7

11.1

9.9

12.5

12.6

11.8

Phoenix,AZ

59.5

2.6

17.5

2.8

13.0

11.0

15.4

20.0

15.5

9.3

8.2

11.5

13.1

9.1

Sacramento,CA

52.8

4.9

10.8

9.0

12.5

9.7

17.9

20.7

16.9

9.4

8.7

12.0

11.8

9.7

SanDiego,CA

50.4

2.7

17.2

7.9

9.5

7.0

11.6

16.5

10.1

8.5

7.2

11.2

12.2

8.5

Denver,CO

68.7

3.5

11.5

2.6

9.3

6.8

16.9

20.7

9.5

5.2

4.5

8.9

8.5

5.1

Atlanta,GA

51.3

25.0

4.7

4.0

9.2

6.5

14.6

14.7

7.9

9.6

6.5

15.8

13.4

7.0

Miami,FL

15.6

11.4

62.3

1.2

9.0

6.4

8.0

10.2

7.8

23.4

18.0

23.3

25.5

20.1

Oakland,CA

77.4

3.6

2.7

4.1

9.0

6.1

12.7

17.7

7.2

8.3

6.7

11.7

12.1

7.4

Minneapolis-

St.Paul,MN

39.2

6.9

14.4

19.6

8.8

7.2

22.0

22.3

17.0

5.8

5.2

11.6

9.5

7.9

Cleveland,OH

71.8

13.1

2.3

1.2

8.2

6.1

18.1

11.1

NA

10.5

8.7

20.1

13.1

6.4

LosAngeles,CA

32.2

7.2

29.7

10.5

6.8

4.7

7.6

9.7

6.4

8.9

7.1

11.0

11.4

7.5

Washington,DC

40.0

24.2

10.3

7.3

6.7

4.2

6.0

20.2

9.1

6.6

4.0

10.4

9.9

6.2

Tampa,FL

64.3

6.8

11.3

2.3

6.7

6.1

8.4

9.7

7.3

15.8

14.6

20.3

22.0

16.1

Houston,TX

47.5

11.5

lost ground

Documents