J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

E c o n o m i c F a c t o r s A f f e c t i n g H o m eM o r t g a g e D i s c l o s u r e A c t R e p o r t i n g

A u t h o r Michael LaCour-Li t t le

A b s t r a c t The release of the 2004–2005 Home Mortgage Disclosure Actdata raised a number of questions given the increase in thenumber and percentage of higher-priced home mortgage loansand continued differentials across demographic groups. Thispaper assesses three possible explanations for the observedincrease in 2005 over 2004: (1) changes in lender businesspractices; (2) changes in the risk profile of borrowers; and (3)changes in the yield curve environment. Results suggest that aftercontrolling for the mix of loan types, credit risk factors, and theyield curve, there was no statistically significant increase inreportable volume for loans originated directly by lenders during2005, though indirect, wholesale originations did significantlyincrease. The findings also reveal that the market price of riskincreased by about 15 basis points in 2005 versus 2004, implyingthat mortgage costs increased for all borrowers on a risk-adjustedbasis.

The Home Mortgage Disclosure Act (HMDA) was enacted in 1975 as a result ofpolicymakers’ concerns that financial institutions were failing to provide mortgagecredit in low income and center city neighborhoods, thereby accelerating urbandecline. Disclosure of home lending activity was intended to assist the public andpolicymakers in determining whether lending institutions were meeting thehousing finance needs of their local communities and to facilitate enforcement offederal fair lending laws. Over the past thirty years, HMDA has evolved to coveradditional elements of the mortgage lending process. The first major legislativeamendments occurred in 1989; these required the disclosure of application andloan-level information for home loans, including the disposition of applicationsand the income, sex, and race or ethnicity of individual loan applicants. Prior tothat time, HMDA disclosures were limited to summary totals of loans actuallyoriginated by census tract and roll-up categories. Analysis of the loan-level dataprompted concern about the fairness of mortgage lending decisions, as the datarevealed disparities in the rates of approval of loan applications across racial andethnic lines. Since that time, HMDA data has become an important screening toolin fair lending enforcement, though it is widely recognized that risk factors, whichplay a major role in both underwriting and loan pricing, are not contained in

4 8 0 � L a C o u r - L i t t l e

it.1 For a discussion of econometric issues related to attempting to estimatediscrimination effects from reduced form equations, see Phillips and Yezer (1996).

Additional changes to the Federal Reserve’s Regulation C, which implementsHMDA, occurred in 2002.2 A number of significant changes were made to thereporting requirements beginning with the 2004 data, increasing the types andamount of information made available about home lending. The most significantamong these was the requirement that lenders disclose pricing for higher pricedloans, defined as those with annual percentage rates (APRs) above specifiedthresholds.3 Policymaker goals in requiring this additional reporting were to learnmore about pricing in the relatively new subprime market, in which variousabusive practices had been alleged. For a review of the uses and limitations of thenew HMDA pricing data for fair lending analyses, see Staten (2005).

While the relationship is inexact, APR spread-reportable loan volume is oftentaken to be a proxy for subprime, or nonprime, lending, since subprime loanscarry rates that are generally several hundred basis points higher than conventionalconforming or prime loans. For example, LaCour-Little (2007) reports that theaverage subprime mortgage rate was about 270 basis points higher than theaverage conventional conforming rate for first lien home purchase loans duringcalendar year 2002, though increased competition may have reduced this spreadin recent years.4 A number of commentators on the HMDA data have interpretedspread-reportable loans as subprime loans (see, for example, Center forResponsible Lending, 2006 or Consumer Federation of America, 2006). Thismarks a change from the prior research practice of characterizing loans assubprime if they were originated by lenders whose predominant business wassubprime, based on a list developed by the Department of Housing and UrbanDevelopment (HUD).5 Calem, Gillen, and Wachter (2004) state that all studiesprior to that date, including their own analysis, used the HUD list to identifysubprime lending. The difficulty with the HUD list, of course, is that it fails toaccount for differences in the mix of prime and subprime business at many lenders.Other researchers have characterized HMDA loans for which rate spreads arereportable as ‘‘roughly equivalent to what industry sources call non-prime orsubprime loans,’’ (Apgar, Bendimerad, and Essene, 2007).6

The volume of mortgage lending covered by HMDA is extensive, including about8,800 lenders in both reporting years. For a previous analysis of HMDA coverage,see Berkovec and Zorn (1996). For 2005, lenders reported information on slightlyover 30 million home-loan applications, up from about 28 million applications in2004. Federal Reserve Board economists have analyzed and reported on patternsin both the 2004 and 2005 HMDA data (Avery, Canner, and Cook, 2005; andAvery, Brevoort, and Canner, 2006). In 2004, HMDA data showed that about15.5% of all originated loans were rate spread-reportable. By 2005, however, thepercentage of rate spread-reportable loans had increased over ten percentage pointsto 26.2%.7 In terms of loan counts, this amounts to an increase from about 2.2million spread-reportable loans in 2004 to about 4.1 million loans in 2005. This

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 8 1

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

occurred while the total market also grew, from about 14 million loans in 2004to almost 16 million loans in 2005.

Avery, Brevoort, and Canner (2006) identify three factors that may have producedchanges in the reporting of higher-priced (i.e., spread-reportable) lending: (1)changes in lender business practices; (2) changes in borrower credit risk profiles;and (3) changes in the interest rate environment. The objective here is to furtheranalyze these potential explanations and suggest additional factors that helpexplain the observed increase in reportable loans during 2005.

The plan for the balance of the paper is as follows. The second section presentsa review of the economic environment that existed during 2004–2005, notingmajor trends in the mortgage market during those years. In the third section, theimplications for HMDA reporting are described along with an outline of theempirical analysis used to test those implications. The fourth section describes thedata, the empirical analysis, and the results. The final section offers conclusionsand further research questions.

� T h e E c o n o m i c E n v i r o n m e n t 2 0 0 4 – 2 0 0 5

I n t e r e s t R a t e s , H o u s e P r i c e s , a n d H o u s e h o l d I n c o m e

Many commentators have noted the flattening of the yield curve over the courseof 2004 through 2005. Exhibit 1 illustrates this pattern, graphing the ratio of the10-year constant maturity Treasury to the one-month Treasury monthly. Most ofthe movement was caused by increases in short-term rates, as the 10-year remainedin a fairly tight band between 4.00% and 4.75% over the two-year period. As aresult, 30-year fixed-rate prime mortgage rates moved relatively little over thisperiod as well, fluctuating between 5.45% and 6.33%, based on the Freddie MacPrimary Mortgage Market Survey data. It was not until 2006 that prime mortgageinterest rates consistently exceeded 6.00. The yield curve remained flat andoccasionally slightly inverted during 2006.

A related pattern observable in the adjustable rate segment of the mortgage marketand related to the flattening of the yield curve is an increase in the initial termdiscount provided on these loans. The initial term discount is the differencebetween the initial term rate and the fully indexed rate, defined as the index plusthe margin at the time of loan origination. As a result, adjustable rate instrumentswould be assumed to experience larger increases at the time of re-set, affectingAPR calculation. APR calculations for adjustable rate mortgages (ARMs) will bediscussed later in the paper.

Though less obvious, there appears to be some evidence that the risk premiumincreased over this time period as well. Tabulation of AAA and BAA bond yields(not shown) monthly over the time period 2004–2005 shows a slight widening.To the extent that the flatter yield curve signaled greater risk of recession and

48

2�

La

Co

ur

-L

it

tl

e

Exhibi t 1 � Change in the Yield Curve: 2004–2005

0.00

1.00

2.00

3.00

4.00

5.00

6.00

Jan-

04M

ar-0

4M

ay-0

4

Jul-0

4Sep

-04

Nov-0

4Ja

n-05

Mar

-05

May

-05

Jul-0

5Sep

-05

Nov-0

5Ja

n-06

Mar

-06

May

-06

Jul-0

6Sep

-06

Nov-0

6

Date

Rat

io o

f 10

Yea

r to

th

e O

ne

Mo

nth

Tre

asu

ry

Slope

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 8 3

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

market participants anticipated future housing market downturns (as indeed didoccur during 2006–2007) then risk premiums might reasonably be expected tohave increased in advance. Using the empirical data described later in this paper,a test can be constructed to determine whether this overall increase in the marketprice of risk passed through to the mortgage market over this time period. Tobriefly preview results, the findings indicate about a 15 basis point increase in themarket price of risk over this time period.

Over this same time period, house prices in the United States continued theirupward trajectory. For example, the Office of Federal Housing EnterpriseOversight (OFHEO) repeat sale House Price Index over the eight quarters of2004–2005 increased by about 33%. The mean and median prices can also becalculated from the Federal Housing Finance Board’s MIRS data described laterin the paper. Over the 24-month period from January 2004 to December 2005,the mean house prices as reported increased from $259,000 to $309,000 and themedian increased from $211,000 to $253,000. Of course, many metropolitan areasexperienced house price growth well in excess of the national rate, howevermeasured. Meanwhile, real median household income increased by only 1.1% in2005 over 2004 to $46,326. As a result, indicators of housing affordability, suchas the widely reported National Association of Realtors measure, showed decliningaffordability over this time period.

Across demographic groups, there continues to be substantial difference acrossracial and ethnic groups: Black real median income was $30,858; Hispanic realmedian income was $35,967; White Non-Hispanic real median income was$50,784, and Asian real median income was $61,094 (all income figures U.S.Census Bureau 2006).

M o r t g a g e M a r k e t T r e n d s

One of the most pronounced trends in the mortgage market over the 2004–2005timeframe was the increase in Alt-A lending. Moody’s (2007, a-c) reports thatAlt-A securitizations, which include option ARMs and other recent productinnovations, virtually doubled in 2005 versus 2004, reaching $425 billion. Basedon securitized volume, Alt-A outstripped the volume of jumbos for the first timein 2004 and was more than twice total jumbo volume in 2005. Since Alt-Aunderwriting standards are more flexible than traditional jumbo standards, whichtend to require prime quality credit and only loan amount as a variance from whatwould otherwise be conventional conforming underwriting standards, it seemsreasonable to infer that borrowers needing nonconforming size loans increasinglyswitched to Alt-A over the 2004–2005 time period.

Related to this is the trend toward consolidation in the mortgage industry and theentry of new firms, especially Wall Street investment houses seeking to establishvertically integrated mortgaged-backed securities production capabilities. Forexample, in 2005 Lehman Brothers, Bear Stearns, and Credit Suisse were all top-

4 8 4 � L a C o u r - L i t t l e

ten issuers of Alt-A mortgage-backed securities. Concurrently and perhaps in partrelated to the accounting scandals that tarnished both firms’ reputations, volumeand market share for Freddie and Fannie dropped significantly, falling to about40% versus 70% in 2003. This pattern can be readily seen in bond marketaggregate statistics compiled by the Federal Reserve (graph not included in thisversion of the paper but available from the author on request).

A continued trend over this same time period was the growth in so-called ‘‘non-traditional’’ mortgage products, including interest-only and payment-optionARMs. Such products have been labeled ‘‘affordability products’’ by some marketparticipants, since they allow households to borrow more relative to income byaccepting reduced (and possibly negative) amortization of their mortgage debt, atleast in the short run, and a greater degree of interest rate risk than would bethe case with the traditional fixed-rate fully amortizing loan. For a morecomprehensive review of the development of non-traditional mortgage products,see GAO (2006). GAO reports a tripling of the use of non-traditional products (asa percentage of all loan originations) over the period 2003–2005, with the use ofthese products concentrated in coastal states with relatively high cost housing.GAO also reports that such products became available to a wider segment ofthe borrowing population, including lower-income households and first-timehomebuyers.

The fundamental economic reason for the growth of these products is thecontinued pattern of house price growth outpacing income growth. Of course,existing home owners (about 69% of all households) benefit from rising houseprices and associated capital gains, so it is not the case that all households arestretching farther to afford home purchase. There is some evidence that theseaffordability products are in greater use among younger households; for example,a recent Wall Street Journal Online/Harris Interactive personal-finance pollindicated that younger borrowers have been more likely to use non-traditionalproducts. For example, the poll indicated that 23% of 18–34-year-old borrowershave an interest-only product, while only 7% of 45–54-year-old borrowers do.

According to the Mortgage Bankers Association (2007), interest-only (IO) loans,with both adjustable- and fixed-rates, and payment option loans that allow someamount of negative amortization, have become a significant part of the mortgagemarket In the second half of 2005 and the first half of 2006, IOs accounted forabout 25% of the dollar volume of originations. Payment option loans representstill another product innovation; typically allowing borrowers a choice of fourdifferent payments with each monthly statement. Borrowers may make fullyamortizing payments calculated over a 30- or 15-year term or they may make aninterest-only payment, which would not reduce the loan balance, or a minimumpayment less than the interest-only payment. This last choice would negativelyamortize the loan, as the interest due above the minimum payment would be addedto the loan balance. Typically, the maximum allowable negative amortization iscapped at 110% or 115% of the loan’s original balance, with balances higher thanthat level triggering a new amortization schedule requiring higher monthly

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 8 5

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

payments. Hence, if initial LTV ratios are moderate, say 80%, negative equity isunlikely unless house prices actually decline, a problem increasingly evidentduring 2006 and later.

� I m p l i c a t i o n s f o r H o m e M o r t g a g e D i s c l o s u r e A c tR e p o r t i n g

HMDA spread reporting is based on comparable maturity Treasury rates, yet loanpricing, generally, is not. Consider a very simple example of the effect of this facton APR spread reporting, focusing on 30-year loans for simplicity’s sake. If the30-year Treasury rate remained unchanged at, say, 5.00% over time, the thresholdrate will be an APR of 8.00%. Mortgages, however, represent callable debt, sincethe borrower can prepay at any point in time, though the prepayment penaltiescommon in subprime and some Alt-A loans are intended to discourage refinancing,at least during the first few years of the contract. Prepayment, as well asamortization itself, reduces the expected life of mortgage loans substantially. Forthirty-year fixed-rate mortgages, the expected life is 5–10 years. For adjustable-rate mortgages, it is even shorter, probably 3–5 years. For subprime and Alt-Aloans it is shorter still, approximately 2–4 years. Specific contract features mayfurther alter expected durations. For a more detailed discussion of prepaymentsand the duration of mortgages, see, for example, Hayre (2001).

Accordingly, loan originators typically price mortgages off of the point on theyield curve corresponding to expected duration, rather than stated maturity. Hence,a subprime or Alt-A loan is generally priced based on a spread above the 2- or3-year rate, even if the contract provides for 30-year maturity. Assume, to continuewith the example above, that the appropriate credit spread for prime loans is 200basis points and the approximate credit spread for riskier subprime loans is 500basis points. The question is spread above what?

If the prime loan is priced off a 10-year Treasury of, say 4.50%, the prime ratewill be 6.50% and the spread to the 30-year Treasury (assuming note rate equalsAPR for simplicity) will be 150 basis points and the loan will not be spread-reportable. If the yield curve is steep, say with the two-year rate at 2.00%, thenthe subprime rate will be 7.00% and the subprime loan will also not be spread-reportable, since it will be priced only 200 basis points (7.00%–5.00%) over the30-year Treasury. But, on the other hand, if the yield curve is relatively flat,implying a two-year rate of, say 4.00%, then the subprime rate will be 9.00% andthe subprime loan will be reportable, since the relevant 30-year benchmark of5.00% (assumed constant here for illustration purposes) would be a full 400 basispoints lower. If this scenario played out precisely over a two-year period and allloans were identical within category, then no loans would be reportable duringthe first year and all of the subprime loans would be reportable in the second year.Of course, reality is considerably more complex, as credit spreads may fluctuateover time and expected durations may vary based on continuously updated

4 8 6 � L a C o u r - L i t t l e

prepayment forecasts based on interest rate movements and other economicindicators.

A second important implication of the growth in Alt-A and non-traditionalmortgage products is an increase in the volume of ARMs.8 Because future interestrates are uncertain, calculation of the APR on an ARM requires assumptions. Inparticular, the APR calculation assumes that the rate on the loan will increase atits first adjustment to the fully indexed rate (the index in effect at the time of loanorigination plus the margin established by the terms of the contract) and thenremain unchanged for the balance of the term. It was mentioned earlier that fullyindexed rates on ARMs increased relative to the initial rate over the period 2004–2005.

An example may help illustrate this point. Suppose an ARM indexed to the one-year Treasury rate was originated as in April 2007. The one-year Treasury rateon that date mid-month was 4.97% and, according to the Freddie Mac PrimaryMortgage Market Survey; one-year ARMs originated during that month had anaverage starting rate of 5.45% with a margin 2.76%. Hence, to calculate the APRon the loan, one would assume that the initial one-year rate would be 5.45% andthat one year later the rate would increase to 7.75% (equal to the fully indexedrate of 4.97 � 2.76 at time of origination). The APR calculation further assumesthe loan will be held to maturity, so one must calculate the implied yield for oneyear at 5.45% and 29 years at 7.75%.9 This produces an APR of 7.53, ignoringany points or other upfront fees that might be included in the calculation.

From this example, it can be seen that APR will tend to increase with both theindex and the margin and decrease with the length of the initial (discounted) termto first reset, at least in the usual case (and in the example provided) where theinitial start rate is discounted well below the fully indexed rate. Later it will beenseen that the growth of ARM lending over the period 2004–2005, together withthe increase in the initial term discount, likely played a significant role in thegrowth of APR spread-reportable lending.

The growth in junior lien lending, especially simultaneous close seconds over the2004–2005 period, has already been noted. Indeed, examination of the tablescontained in Avery et al. (2005) and Avery et al. (2006) indicates that about422,000 additional loans secured by junior liens were spread-reportable in 2005compared to 2004. Several contract features of second liens are relevant here.First, they are typically shorter than 30 years in maturity, often 15-year; hencethey will trigger reporting thresholds at lower rates, even though the reportingstandard is 500 basis points, rather than 300. Second, they are typically adjustablerate. While expected durations for these loans is not readily available, it seemslikely that the same factors affecting Alt-A and subprime would have tended tomake these loans relatively higher priced in 2005 compared to 2004, increasingthe probability of spread reporting.

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 8 7

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

� D a t a a n d E m p i r i c a l M e t h o d o l o g y

The data compiled for this research covers calendar years 2004 and 2005, as wellas the first half of 2006, though the main focus is 2004–2005.10 Three distinctdata sources are used. The first is the servicing and securitization data fromLoanPerformance, a well-known commercial provider of mortgage loan data tothe industry, which claims coverage of approximately 78% of the mortgagemarket.11 The servicing data contains both securitized and non-securitized loansand information is available at certain aggregate levels only. The securitizationdata includes loan-level information, but only for loans that serve as collateral forprivate label mortgage-backed securities (jumbo, Alt-A, and subprime); hence,Freddie and Fannie issues are not included. The LoanPerformance data sets, whilebroadly representative of the entire mortgage market, do not contain many HMDAdata fields or the APR on the loan. Exhibit 2 provides a comparison the LPservicing data and its correspondence to overall HMDA data for 2004–2005.Depending on the category and whether one matches to loan counts or loanbalances, the LP data covers 68%–75% of the total HMDA market.

The second data source is a proprietary data set assembled under the direction ofthe Consumer Mortgage Coalition, an industry trade group. This data set includesmany, but not all, HMDA fields, including whether the loan was rate spread-reportable, and important supplemental information, such as whether the loan isfixed or adjustable rate, whether it was originated directly or indirectly (e.g.,through a mortgage broker), and several key risk factors, such as credit score andloan-to-value ratio. The data represents a random sample of all participatinglenders origination volume and includes multiple unidentified lenders.Accordingly, it is expected to be broadly representative of large volume financialinstitutions who were engaged in both prime and non-prime lending activity duringboth calendar years. These two data sources will be referred to as the LP dataand the CMC data. Both are large data sets. In the LP data, there are approximately9.3 million loan records over the two years and in the CMC data, there areapproximately 385,000 loan records, including loans from the first six months of2006.

The third data source is the previously mentioned Federal Housing FinanceBoard’s Monthly Interest Rate Survey (MIRS) data for calendar years 2004 and2005, which contains survey data on approximately 724,000 home purchase loans.The survey provides monthly information on interest rates, loan terms (includingLTV), and house prices by property type, by loan type (fixed- or adjustable-rate),and lender type (savings associations, mortgage companies, commercial banks,and savings banks). To conduct this survey, the FHFB asks a sample of mortgagelenders to report the terms and conditions on all single-family, fully amortized,purchase-money, non-farm loans that they close during the last five business daysof the month. The survey excludes FHA-insured and VA-guaranteed loans,multifamily loans, manufactured housing loans, and loans created by refinancing

48

8�

La

Co

ur

-L

it

tl

e

Exhibi t 2 � LP Servicing Dataset Compared to Aggregate HMDA Data: 2004–2005

Loan Category 2004 2005

Purchase Sum of Total Count HMDA 6,452,860 7,404,202LP-Serv 4,496,691 4,816,533[HMDA–LP-Serv] 1,956,169 2,587,669Percent Difference 30% 35%

Sum of Total Bal HMDA $1,154,349,288,000 $1,381,992,788,000LP-Serv $820,465,687,946 $946,042,864,546[HMDA–LP-Serv] $333,883,600,054 $435,949,923,454Percent Difference 29% 32%

Purchase Sum of Total Count 12,905,720 14,808,404

Purchase Sum of Total Balance $2,308,698,576,000 $2,763,985,576,000

Refinance Sum of Total Count HMDA 7,606,211 7,121,455LP-Serv 5,513,571 4,650,349[HMDA–LP-Serv] 2,092,640 2,471,106Percent Difference 28% 35%

Sum of Total Balance HMDA $1,340,447,044,000 $1,401,667,305,000LP-Serv $1,007,227,615,395 $969,274,959,151[HMDA–LP-Serv] $333,219,428,605 $432,392,345,849Percent Difference 25% 31%

Refinance Sum of Total Count 15,212,422 14,242,910

Refinance Sum of Total Balance $2,680,894,088,000 $2,803,334,610,000

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 8 9

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

Exhibi t 3 � Descriptive Statistics: CMC Data 2004 and 2005 Originations

Variable

2004Originations

N Mean Std. Dev.

2005Originations

N Mean Std. Dev.

Loan is spread-reportable 63,101 0.15 0.36 63,470 0.26 0.44

Term (in years) 63,101 26.3 6.67 63,470 26.7 6.58

Loan amount (thousands) 63,101 197.7 181.5 63,470 207.4 195.8

Junior lien 63,101 0.03 0.18 63,470 0.05 0.23

Not owner occupied 63,101 0.08 0.28 63,470 0.10 0.30

Manufactured housing 63,101 0.02 0.13 63,470 0.01 0.11

FICO credit score 60,573 699 72 62,384 697 74

LTV ratio 63,101 76.9 18.5 63,470 77.1 18.5

Adjustable rate (ARM) 63,101 0.36 0.48 63,470 0.38 0.48

Refinancing 63,101 0.53 0.50 63,470 0.51 0.50

Home improvement 63,101 0.03 0.17 63,470 0.05 0.21

Government 63,101 0.07 0.25 63,470 0.05 0.21

Wholesale origination 63,101 0.29 0.46 63,470 0.27 0.44

Slope of yield curve 63,101 3.85 0.98 63,470 1.52 0.35

Note: Reportable loans over-sampled to approximate HMDA proportions.

another mortgage. This publicly available data has been widely used byresearchers in the mortgage field for many years so it will not be further described.

Exhibit 2 compares the LP data to HMDA totals for the calendar years underconsideration. Exhibit 3 provides descriptive statistics for the CMC data.

The general methodology is to determine the effect membership in particular loancategories has on the probability that the loan is spread-reportable under HMDA.This analysis uses loan level data so that the APR can be observed and todetermine whether the loan was reportable under HMDA given the interest rateenvironment in the particular year of origination. Concurrently, the trend inparticular loan categories is examined for the entire market over the time horizonof interest, calendar years 2004 and 2005. If, for example, ARMs are more likelyto be reportable than FRMs, then an increase in the percentage of total ARMlending in the market is likely to increase the overall level of loans with APRspread reporting. Ultimately, we will quantify those relationships using regressiontechniques and producing equations that may be used, together with assumptionsabout volume and mix, to predict HMDA results out-of-sample.

4 9 0 � L a C o u r - L i t t l e

But to begin, changes in the risk profile of borrowers over time are assessed. Ifthe pool of borrowers in 2005 were systematically riskier than in 2004, a higherfraction of APR spread-reportable loans would be expected, given the risk-basedpricing that has become prevalent in the mortgage market.

B o r r o w e r R i s k P r o f i l e s

Default risk in mortgage lending has been a topic of research for many years [seeVandell, (1993) for a survey, and Epley, Liano, and Haney (1996) on borrowerequity issues]. Empirically, risk in mortgage lending has been shown to be relatedboth to borrower credit (as measured, for example, by the FICO credit score) andthe level of borrower equity in the property, as generally measured by the loan-to-value ratio (LTV). Indeed, many rate sheets used by lenders explicitly price onthese two factors. In addition, the overall level of borrower indebtedness, asmeasured by the debt-to-income ratio, is also thought predictive, and borrowerswith excessive levels of debt (even those with excellent credit) are often limitedto subprime or Alt-A products.

Beginning with LTV, Exhibit 4 depicts average LTV ratios calculated for homepurchase mortgage loans originated over 2004–2006 using the MIRS data. Asimilar pattern results when using the LP data (not shown). Values fluctuate around75%. In 2006, however, values are consistently above 75% and appear to beheading toward 80%. A similar pattern results when using the LP data (notshown).

The use of simultaneous close second mortgages (sometimes called‘‘piggybacks’’) is reported to have increased during 2004–2005. Such loans wouldproduce effectively higher overall LTV ratios, potentially not reflected in Exhibit4. Good data on simultaneous close seconds are not available from any of thethree data sets, but industry publications suggest levels of 25%–30% of all homepurchase loans and increasing over time, with some significant variation relatedto housing price levels.

Exhibit 5 presents the distribution of borrower credit score over time for homepurchase loans over the period 2004–2006. Using the LP data, the percentage ofnewly originated loans of all types (prime and subprime) is graphed to borrowerswith credit scores less than 600, 600–720, and over 720. Below 600, mostborrowers would be in the subprime category; above 720 they would clearly bein the prime category, with 600–720 representing an intermediate range.

No obvious shift in the mix of borrower credit scores is apparent over the timeperiod 2004–2005. All three lines are virtually flat, suggesting that the mix ofborrower credit types did not change over the two-year period. Borrowers in thelowest score category (under 600) consistently represent about 10% of borrowers.As with LTV, however, the picture changes somewhat in 2006. Here a slightdecline is observed in the fraction of borrowers in the highest score category,along with an offsetting increase in the fraction in the intermediate category. The

Ec

on

om

ic

Fa

ct

or

sA

ff

ec

ti

ng

Ho

me

Mo

rt

ga

ge

�4

91

JR

ER

�V

ol

.2

9�

No

.4

–2

00

7

Exhibi t 4 � Mean Home Purchase Loan LTV: 2004–2006 (MIRS Data)

70.00

75.00

80.00

85.00

90.00

Jan-04 Apr-04 Jul-04 Oct-04 Jan-05 Apr-05 Jul-05 Oct-05 Jan-06 Apr-06 Jul-06 Oct-06 Jan-07

Average LTV

49

2�

La

Co

ur

-L

it

tl

e

Exhibi t 5 � Distribution of FICO Scores for Purchase Loans Originated 2004–2006

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Jan-04 Apr-04 Jul-04 Oct-04 Jan-05 Apr-05 Jul-05 Oct-05 Jan-06 Apr-06 Jul-06 Oct-06 Jan-07

Date

Per

cen

t o

f A

ll L

oan

s

Pct<600

Pct 600-720

Pct>720

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 9 3

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

fraction in the lowest category stays relatively constant at about 10% during 2006.Similarly stable patterns, though at slightly lower overall credit score levels, wouldresult from graphing loans for refinancing purposes. Such loans commingle cashout refinancing, considered relatively risky especially if for the purpose of debtconsolidation, with relatively lower risk rate-term refinancing, which tends to bedriven by interest rates.

In summary, no evidence is found of an increase in borrower LTV or a decreasein borrower credit score over the 2004–2005 period. On the other hand, theincrease in the use of non-traditional products and some of those, notably thosereduced documentation products not requiring verification of borrower income,would entail more risk. Likewise, products that incorporate little, or negativeamortization, might be characterized as riskier per se. Accordingly, Exhibit 6shows the percentage of loans in which negative amortization is allowed using LPdata. This illustration shows a sharp increase over the late 2004 and early 2005,followed by relative stabilization at around 25% of loan volume, followed by aslight decline in 2006, though data on this field is limited in the second half usingthe LP data source the graph is truncated at midyear 2006. The initial pattern maybe a proxy for the growth of the Alt-A segment of the market over this timeperiod; decline in the latter part of 2006 may reflect tightening of underwritingstandards in reaction to increases in default and housing market declines andregulatory pronouncements.

O r i g i n a t i o n Vo l u m e M i x C h a n g e s

The effect of the industry shift toward Alt-A loans is now examined, along withthe mix of ARM to FRM in the market. As discussed, a sharp increase in Alt-Alending is apparent in Exhibit 6. Tabulations (not presented) of all loans, bothprime and nonprime, that were originated as ARMs, monthly, over the 2004–2005time period shows no clear pattern. There was an initial increase from a level ofabout 30% in early 2004 to a level approaching 45% over much of 2004, followedby a decline to under 35% by the end of the two year period.12

This pattern of ARM usage is initially puzzling. Most of the published academicresearch on ARM usage has stressed the relative cost advantage of ARMs overFRMs when the yield curve is steep (Brueckner and Follain, 1988; Dhillon,Shilling, and Sirmans, 1989; and Stanton and Wallace, 1995, 1999). These papers,however, focus on the mortgage market prior to the development of the nonprimesegment, in which a significant fraction of loans are ARMs. Borrowers in thesubprime category, in particular, are generally thought to be relatively less ratesensitive and the spread between FRM and ARM rates in that segment of themarket tend to be smaller than in the prime market. If nonprime lending, includingAlt-A, garnered a higher market share over the course of the time period studiedand if a disproportionate fraction of nonprime loans are ARMs, then the ARMshare pattern observed makes sense. Moreover, the pattern observed may reflecta reaction to the movement of FRM rates over this time period, which initiallyincreased, but then fell.

494�LaCour-Little

Ex

hib

it6

�Loans

Originated

Allow

ingN

egativeA

mortization:2004

–2006

0%

10%

20%

30%

40%

50%Jan-04

Feb-04

Mar-04

Apr-04

May-04

Jun-04

Jul-04

Aug-04

Sep-04

Oct-04

Nov-04

Dec-04

Jan-05

Feb-05

Mar-05

Apr-05

May-05

Jun-05

Jul-05

Aug-05

Sep-05

Oct-05

Nov-05

Dec-05

Jan-06

Feb-06

Mar-06

Apr-06

May-06

Jun-06

Jul-06

PctN

egAm

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 4 9 5

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

E f f e c t o f t h e Y i e l d C u r v e , R i s k , a n d O t h e r E c o n o m i cF a c t o r s

This discussion begins with a simple analysis of the effect of the yield curve onrate spread reporting, ignoring all other factors, including the question of whethercomparable maturity treasuries used are appropriate benchmarks. This simple,mechanical approach is followed by a more comprehensive regression analysis,which takes into account the mix of loan volume and potential changes in riskfactors across loan segments. Finally, there is a separate analysis of the change inthe market price of risk over time. CMC data are used for all three analyses.

The most direct way to consider the effect of the yield curve on APR spreadreporting in 2005 is to ask the question: If the same loan was originated underthe 2004 yield curve environment, would the loan have been reportable? To answerthis question, the 2004 yield curve data (the comparable maturity table providedby the FFIEC for calendar year 2004) is simply loaded onto the 2005 data andthe APR spread for each loan is re-calculated to determine whether that re-calculated spread would have been reportable under HMDA. For example, woulda loan with a 30-year term and APR of 7.75 originated on February 15, 2005 havebeen reportable had it been originated February 15, 2004? In this example, the30-year comparable maturity rate was 4.92% in February 2004 so a loan with anAPR of 7.75 would not be reportable, since its rate spread is 7.75 � 4.92 �2.83 � 3.00. In contrast, as of February 2005, the 30-year comparable maturityrate was 4.55%, so a loan with an APR of 7.75% would be reportable, since7.75 � 4.55 � 3.20 � 3.00.

Results of repeating this process for all loans in the CMC sample are shown inPanel A of Exhibit 7.13 Overall, 23.2% of all loans would have been reportable,had the 2004 yield curve applied, rather than 26.0% actually reported. Thisdifference (of 2.8 percentage points) is somewhat higher than the estimate offeredby Avery et al. (2006) of 2.0 percentage points. To explain the source of thedifference, note that these authors had to assume that all reportable loans are 30-year term since loan term is not a required data element in HMDA. In fact, itappears that about 25% of HMDA spread-reportable loans are, in fact, shorter interm than 30 years, as shown in Panel B of Exhibit 7.

Returning to Panel A of Exhibit 7, the process can be repeated to isolate the effectof the yield curve on various loan types: ARM vs. FRM, high credit score versuslow credit score, etc. Results show considerable variation in the effect of the yieldcurve, with the greatest effect (5.6 percentage points) on low credit score loans(defined as FICO less than or equal to 620) and the smallest effect on junior liens(0.1 percentage points). Since any individual lender will have its own mix of loantypes and the yield curve appears to vary in its effect depending on loan type, itwill be hard to generalize on the overall effect of yield curve for any individuallender and interpretation of an increase, or decrease, in 2005 over 2004 will bedifficult.

49

6�

La

Co

ur

-L

it

tl

e

Exhibi t 7 � Effect of the Yield Curve on HMDA Spread Reporting 2004–2005

TotalRetailOnly

WholesaleOnly

ARMOnly

FRMOnly

FICO�620

FICO620–720

LTV�90

FICO�620& LTV�90

Loan�$100k

SecondLiens

Panel A: Yield Curve Simulation

%Spread-reportable in 2005 26.0 25.0 28.8 45.5 14.4 70.0 34.3 30.6 46.0 40.7 22.0

%Reportable Under 2004 Yield Curve 23.2 22.5 25.0 40.8 12.7 64.4 29.7 27.1 41.6 37.8 21.9

Difference 2.8 2.4 3.9 4.7 1.7 5.6 4.6 3.6 4.5 2.9 0.1

Notes: Comparable maturities for calendar year 2004 are used to recalculate the APR spread for loans originated in 2005. Difference measures pure effectof interest rate changes on reporting during 2005.

Ec

on

om

ic

Fa

ct

or

sA

ff

ec

ti

ng

Ho

me

Mo

rt

ga

ge

�4

97

JR

ER

�V

ol

.2

9�

No

.4

–2

00

7

Exhibi t 7 � (continued)

Distribution of Loan Term for HMDA Reportable Loans: 2004–2005

Percentile Term in Years

Panel B: Distribution of Loan Term

90% 30

75% 30

50% 30

25% 20

10% 15

5% 10

1% 6

0% 1

Notes: All spread-reportable loans. The source is CMC data.

4 9 8 � L a C o u r - L i t t l e

Next, a regression approach is employed that uses the CMC data to assess theeffect of the yield curve, risk, and other economic factors on whether a loan isspread-reportable under HMDA. The technique is the well-known logit model inwhich the dependent variable is binary, equal to one, if loan is HMDA spread-reportable and zero otherwise. Both the broad categories of loan type and thespecific characteristics of the loan itself are controlled for. Loan-specificcharacteristics include loan term (since APR spread reporting is based off ofcomparable maturity Treasury) and loan amount (to account for the tendency ofsmaller loans to be priced slightly higher to reflect the economies of scale in loanorigination). The slope of the yield curve is also controlled for at the time of loanorigination by using the metric shown in Exhibit 1.

As previously noted, descriptive statistics for the two years are shown in Exhibit3. Average loan size is $198,000 in 2004 and $207,000 in 2005. These values areslightly higher than the mean loan sizes for all HMDA loans implied by Exhibit2.14 All major categories are represented, though less than 10% of the sampleis in non-owner occupied, government-insured, manufactured housing, homeimprovement, or junior liens categories. A little more than half of all loans arefor refinancing purposes in each year. A little less than one-third of all loans wereoriginated through indirect, wholesale channels, which include mortgage brokers,certain correspondent lending relationships, builder programs, and the like.

Since the incidence of reportable loans was lower in the sample compared tooverall HMDA levels, non-reportable loans were randomly discarded so as toproduce an incidence of reportable loans approximately equivalent to that in theaggregate HMDA data, as reported by Avery et al. (2006).15 After deletion ofoutliers and missing values and re-sampling to match aggregate HMDA patterns,the final data set consists of 126,571 loans about equally divided between 2004and 2005 originations. As mentioned, a little less than one-third of these loanswere originated through indirect channels. In such cases, a third party (such as amortgage broker) sets the pricing for the loan while the lender funds and closesit.16 Since a third party is setting the pricing on these loans, it seems appropriateto analyze these separately. Exhibit 8 depicts the aggregate growth of spread-reportable loans volume over the two-year time period, calculating percentagesfrom the CMC data after the adjustments described.

The overall objective is to determine what factors are most important indetermining whether a loan is HMDA spread-reportable in each of the two yearsand to what extent there was a real increase in 2005, after controlling for thosefactors. To do this, six logistic regressions were estimated by dividing the sampleinto the two origination years and whether the loan was originated throughwholesale or retail channels. Exhibit 9 shows results, with Panel A using retailoriginations and Panel B using wholesale originations. Results are highlyconsistent across sub-samples and measures of concordance are very high (over0.90), suggesting successful capture of the main factors that determine whether aloan is HMDA spread-reportable. In the vast majority of cases, estimates arehighly statistically significant.17

Ec

on

om

ic

Fa

ct

or

sA

ff

ec

ti

ng

Ho

me

Mo

rt

ga

ge

�4

99

JR

ER

�V

ol

.2

9�

No

.4

–2

00

7

Exhibi t 8 � Growth of Spread-Reportable Volume: 2004–2005

0%

5%

10%

15%

20%

25%

30%

35%

Jan-04 Mar-04 May-04 Jul-04 Sep-04 Nov-04 Jan-05 Mar-05 May-05 Jul-05 Sep-05 Nov-05

Date

Per

cen

tag

e

Source: CMC data.

50

0�

La

Co

ur

-L

it

tl

e

Exhibi t 9 � Logistic Regressions: Dependent Variable Loan is Spread-Reportable

2004

Parm. Est.

Retail

Std. Error Chi-Sq

2005

Parm. Est.

Retail

Std. Error Chi-Sq

Combined

Parm. Est. Std. Error Chi-Sq

Panel A: Retail Originations

Intercept 12.13 0.27 2,020.5 13.47 0.25 2,969.1 11.79 0.18 4,310.8

Loan amount �0.01 0.00 1,439.7 �0.01 0.00 1,445.8 �0.01 0.00 2,737.8Term �10 2.98 0.09 1,151.6 1.92 0.07 713.2 2.36 0.05 1,857.7Term 10–20 1.83 0.06 1,046.0 1.20 0.05 632.7 1.45 0.04 1,652.2

Second Lien �1.69 0.09 328.0 �1.74 0.07 612.6 �1.65 0.06 900.2Not owner occ. 0.35 0.08 17.3 1.17 0.06 414.8 0.90 0.05 376.7Manf. housing 0.17 0.20 0.8 0.49 0.15 11.1 0.28 0.12 5.6

FICO �0.02 0.00 4,195.2 �0.02 0.00 5,760.6 �0.02 0.00 10,120.8LTV 0.03 0.00 568.2 0.04 0.00 1,488.4 0.04 0.00 2,024.3ARM 2.47 0.05 2,402.2 1.88 0.04 2,273.7 2.06 0.03 4,621.8Refi. 1.16 0.05 451.9 1.26 0.04 861.0 1.20 0.03 1,310.7Home imp. 2.01 0.10 424.8 1.99 0.07 725.1 1.97 0.06 1,152.3Government �5.11 0.35 209.7 �4.09 0.18 502.6 �4.35 0.16 736.8Yield curve slope �0.37 0.02 344.9 �1.74 0.07 678.5 �0.47 0.02 645.1Year 2005 0.07 0.05 2.1R2 0.43 0.46 0.44Max-rescaled R2 0.71 0.68 0.68c-Statistic 0.96 0.94 0.95

Ec

on

om

ic

Fa

ct

or

sA

ff

ec

ti

ng

Ho

me

Mo

rt

ga

ge

�5

01

JR

ER

�V

ol

.2

9�

No

.4

–2

00

7

Exhibi t 9 � (continued)

Logistic Regressions: Dependent Variable Loan is Spread-Reportable

2004

Parm. Est.

Wholesale

Std. Error Chi-Sq

2005

Parm. Est.

Wholesale

Std. Error Chi-Sq

Combined

Parm. Est. Std. Error Chi-Sq

Panel B: Wholesale Originations

Intercept 16.20 0.55 861.5 15.98 0.44 1,335.3 12.54 0.31 1,608.1

Loan amount �0.01 0.00 467.4 �0.01 0.00 690.1 �0.01 0.00 957.9Term �10 1.57 0.50 9.7 0.95 0.54 3.1 1.29 0.37 12.3Term 10–20 0.72 0.11 41.4 0.64 0.10 37.6 0.71 0.07 93.0

Second Lien NA NA NA NA NA NA NA NA NANot owner occ. 1.346 0.119 127.9 1.728 0.077 508.1 1.530 0.061 624.0Manf. housing 0.599 0.156 14.8 0.484 0.179 7.4 0.703 0.115 37.2

FICO �0.029 0.001 1,379.5 �0.024 0.001 2,048.0 �0.024 0.000 3,587.4LTV 0.042 0.003 182.6 0.046 0.002 465.2 0.041 0.002 590.0ARM 0.613 0.084 53.3 1.896 0.060 1,007.5 1.371 0.046 893.3Refi. 0.279 0.089 9.8 0.291 0.062 22.0 0.228 0.049 22.0Home imp. 0.765 0.315 5.9 0.706 0.179 15.5 0.671 0.151 19.8Government �20.371 199.000 0.0 �6.090 0.524 135.1 �7.054 0.534 174.4Yield curve slope �0.410 0.036 131.8 �3.150 0.115 757.0 �0.573 0.032 320.0Year 2005 0.887 0.078 128.4R2 0.30 0.48 0.41Max-rescaled R2 0.62 0.68 0.65c-Statistic 0.96 0.94 0.94

Note:* There were no second loans originated through wholesale channels, so this parameter cannot be estimated.

5 0 2 � L a C o u r - L i t t l e

Reviewing results, note that larger loans are less likely to be HMDA spread-reportable. This fact alone helps explain the differential levels of spread reportingin the public HMDA data across racial and ethnic groups. Lower incomehouseholds will tend to demand smaller loans and those loans will be more likelyto be spread-reportable, other factors held constant. Moreover, as we have seen,race/ethnicity is correlated with income, with Black and Hispanic borrowershaving significantly lower incomes and Asian borrowers having significantlyhigher incomes, relative to Non-Hispanic Whites.

Turning to the interest rate situation, the sign of the coefficient on the slope ofthe yield curve is negative, implying that a steeper yield curve makes loans lesslikely to be reportable, other factors held constant. This result is consistent acrossorigination channels and year of origination, suggesting that yield curvemovements play an important role both within, and across, origination years.Short-term loans are also more likely to be reportable, consistent with the lowerthresholds resulting from use of shorter comparable maturity treasuries.

Loans on non-owner-occupied housing, manufactured housing, for refinancing, orhome improvement purposes are all relatively more likely to be reportable.Focusing on the key risk factors of FICO and LTV, loans with higher FICO scoresare less likely to be reportable and loans with higher LTVs are more likely to bereportable. Finally, adjustable rate loans are significantly more likely to bereportable, after controlling for other factors. This pattern persists across bothyears. This is consistent with the story about the effect of the assumptions requiredsimply to compute the APR for an adjustable rate instruments. The one result thatis not consistent with expectations is second liens, which appear to be less likelyto be reportable compared to first liens.18

The third specification of the model continues to segment by origination channelbut combines the two years of data and adds a dummy variable for an originationin calendar year 2005. The sign and statistical significance of this variable willreflect whether reportable loans increased in 2005, after controlling for the mixof loan types, risk characteristics, and yield curve movements. While the sign ofthe coefficient on the year 2005 origination indicator variable is positive in bothcases, it is not statistically significant in 2005 for retail originations, whereas it isstatistically significant for wholesale originations. Thus, all of the real growth inreportable loans in 2005 over 2004 was attributable to wholesale origination, aftercontrolling for risk and other economic factors, including the yield curve slope.

Finally, the risk-adjusted spread over time is examined by utilizing moreappropriate Treasury rates that are intended to approximate lender expectationsabout expected duration. In particular, the 10-year Treasury for all fixed rate loans;the 5-year Treasury for adjustable rate loans to borrowers with credit scores higherthan 620 and for all junior liens; and the 2-year rate for all loans to borrowerswith credit scores less than or equal to 620. Given these more relevant risk-freerates, an adjusted risk spread that is equal to the APR less the relevant rate canbe computed.

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 5 0 3

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

Exhibi t 10 � Test of Whether Risk-Adjusted Spread Increased in 2005

OLS Model of Adjusted Spread

Variable Parm. Est. Std. Error t-Statistic

Intercept 11.46 0.0425 269.4

Loan Amount �0.002 0.0000 �83.8

Second Lien 1.76 0.0168 104.6

Not Owner Occupied 0.32 0.0112 28.6

Manufactured Housing 0.48 0.0267 17.9

Fico Score �0.01 0.0001 �274.4

Loan-to-Value Ratio 0.01 0.0002 35.0

Loan is ARM 0.71 0.0073 98.1

Refinancing Purpose 0.28 0.0075 37.8

Home Improvement Purpose 0.73 0.0180 40.5

Government-Insured Loan �0.71 0.0161 �44.3

Loan Originated in 2005 0.15 0.0064 24.0

Notes: Adjusted R2 � 0.58, Root MSE � 1.12, and N � 122,953. Adjusted spread is calculatedby subtracting duration matched Treasury rate from APR for all FRM, 10-year Treasury; for PrimeARM, 5-year Treasury; for FICO�620, 2-year Treasury; and for all second liens.

In Exhibit 10, the adjusted spread is regressed on the set of risk factors availablein the CMC data set, including an indicator variable for loan origination incalendar year 2005. The value of the parameter estimate for the year 2005 dummyshould indicate the average increase in the market price of risk over the two-yearperiod. In addition, the average predicted values for each month of 2004–2005can be generated to observe the pattern in this adjusted risk spread measure.

Turning first to the regression, signs and magnitudes of coefficients seem quitereasonable. Loan amount has a negative sign, indicating that larger loans requiresmaller risk spreads. Second liens carry risk spreads that are about 176 basis pointshigher than first liens. Non-owner occupied properties carry premiums of 32 basispoints and manufactured housing premiums of 48 basis points. The risk spreadincreases with LTV and decreases with borrower credit score. ARMs carry riskpremiums that are 71 basis points higher than FRMs. Loans for refinancing carryrisk spreads about 28 basis points higher and home improvement loans spreadsabout 73 basis points higher than the hold-out category of home purchase loans.In contrast, government insured loans carry risk spreads about 71 basis pointslower, consistent with the insurance guarantee provided by FHA or othergovernmental entities. Model fit is reasonably good, with an Adjusted R-squaredvalue of 0.58.

5 0 4 � L a C o u r - L i t t l e

Finally, and of particular interest here, loans originated in calendar year 2005 carrya risk spread about 15 basis points higher, compared to loans originated in 2004.The implication of this finding of an increase in the market price of risk is thateven had the mix of loans not changed at all in 2005 versus 2004, the higheroverall pricing of risk would have increased the number of loans subject to HMDAreporting. Replication of this exercise (not reported) using 2006 data producessimilar results, implying that duration-matched credit spreads in the mortgagemarket increased by about 30 basis points over the 2004–2006 period. This is aninteresting result and will be the subject of further research.

R o b u s t n e s s Te s t s a n d O t h e r E c o n o m e t r i c I s s u e s

A number of tests were conducted to ensure the robustness of the regression resultsreported above. Among these were redefining the slope of the yield curve to usedifferent maturity points and using the difference rather than the ratio of the yields,re-running the models presented by lien type, restricting the samples to specificmaturities (e.g., 30-year loans only), and altering the random selection algorithmto produce distinct comparator groups of loans that are not spread-reportable.Results were highly consistent across all of these tests, suggesting that overallresults are robust. These robustness tests are not reported here in the interest ofbrevity but may be obtained from the author by request.

Another econometric issue is that ARM loan choice is endogenous, meaning thatborrowers choose ARMs over FRMs, at least in part, because of the interest ratesituation; hence, it may not be appropriate to include an indicator for ARMs as apredictive variable in the logistic regression models. To address this issue, onemay estimate a separate model of ARM choice (which may be done using eitherthe CMC or MIRS data) and include the predicted value, rather than the actualvalue, in the model. Results (also not reported) indicate that loan amount,geography, and the level of the FRM rate are all predictive of ARM choice andthat inclusion of a predicted value, in place of the actual value, does not materiallyaffect overall model results. Again, these results are available from the author byrequest.

G e o g r a p h i c Va r i a t i o n i n K e y F a c t o r s

Avery et al. (2006) note the wide variation in the percentage of loans that arespread-reportable by geographic area and attribute this both to variation in housingcosts and credit scores across regions of the country. In this section, this findingis explained further by identifying the geographic variation in ARM volume andits relationship to house prices. Since ARMs are much more likely to be spread-reportable, it follows that geographic areas with larger percentages of ARM loanswill tend to have higher fractions of loans that are spread-reportable, other factorsheld constant. ARM percentage can also be shown to be affected by the level ofhouse prices.

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 5 0 5

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

Based on the MIRS data, California is prominent among states with over 50% ofits home purchase loans estimated to be ARMs. In general, states with higherpriced housing tend to have a greater percentage of ARMs. Broadly, states on theEast and West Coast have higher percentage of ARMs as compared to Midwesternand Southern states. The percentage of home purchase loans that were ARMs in2004 and the percentage in 2005 are highly correlated, with a bivariate correlationof roughly 0.9. Similar geographic patterns would be evident for the use of non-traditional products, such as option-ARMs, consistent with results reported byGAO (2006).

� C o n c l u s i o n

This paper reports that a variety of factors affected the increase in the level ofAPR spread reporting during 2005 over 2004. First, the yield curve flatteningproduced changes in the reporting triggers across loan types, with the segmentsof the mortgage market that are most rapidly growing the most affected. Both Alt-A and ARMs fall into this category. Moreover, junior lien financing, particularlysimultaneous-close second mortgages used to stretch affordability on homepurchase transaction may have played a major role, though regression results didnot show that junior liens were more likely to be spread-reportable.

Overall, one might characterize this trend as a change in borrower product choiceresulting in a change in the mix of loan types, rather than a change in lenderbehavior. Of course, to the extent product choice is driven by lender pricing andmarketing strategies, one might also characterize this as a change in lenderbehavior. In addition, new lenders became involved in the market during 2005,especially in the fast growing Alt-A segment, so the organizational structure ofthe primary mortgage market played a role here, too. There was also an estimationthat there was an increase in the market price of risk in the amount of about 15basis points in 2005 compared to 2004, which would have increased the price ofmortgage credit to all borrowers over this time period.

In contrast, no evidence was found that borrower risk, as measured by LTV andFICO, were substantially changing over this time period, except to some extentwith the growth in simultaneous-close second mortgages, and the shift towardinherently riskier products, such as interest-only and payment-option ARMs,particularly in higher cost housing markets. Affordability issues, however, appearto have mainly affected first-time home buyers. Households with existing homeequity benefited from the run-up in house prices and appear to have been able tofund large downpayments for trade-up housing with those gains, keeping averageLTV ratios within normal ranges. Recent house price declines will tend to affectsuch households to a lesser extent so the increase in defaults recently observed inthe market is, unfortunately, likely to be concentrated among those householdswho had to stretch the farthest in order to achieve home ownership.

The regression results presented here indicate that the principal factors affectingHMDA spread reporting in both 2004 and 2005 were loan size, term, purpose,

5 0 6 � L a C o u r - L i t t l e

property type, loan type (especially whether the loan was an ARM), the key riskfactors of credit score and LTV, and the slope of the yield curve. In fact, aftercontrolling for all of these factors, there was no statistically significant increasein the probability of a loan being spread-reportable during 2005 for loansoriginated directly by lenders. For loans originated through indirect, wholesalechannels, however, there was a statistically significant increase in the probabilityof a loan being reportable during 2005.

Policymakers sought to better understand higher-priced mortgage lending whenthey implemented the HMDA price reporting standards that took effect in 2004.Unfortunately, a wide range of economic factors affect whether loans exceed thereporting thresholds and these factors are not constant over time. As a result, itis easy to misinterpret patterns in the data over time and incorrect inferences maybe drawn when those economic factors are not fully taken into consideration. Thenext round of revisions to HMDA represent an opportunity to address these issuesto make resulting measures of higher cost lending more meaningful and stableover time. This will be particularly important if additional consumer protectionsare to be established based on HMDA spread-reportable criteria.

� E n d n o t e s1 See House Financial Services Committee Hearing on HMDA July 25, 2007, for

statements by regulatory agencies summarizing recent actions in fair lendingenforcement; see also Courchane, Nebhut, and Nickerson (2000) for a discussion of fairlending exam processes.

2 See Home Mortgage Disclosure Act (12 U.S.C. §§ 2801-11), Regulation C (12 C.F.R.pt. 203), and accompanying staff commentary (12 C.F.R. pt. 203, Supp. I).

3 The rate spread reporting threshold is 3 percentage points for a first lien loan and 5percentage points for a second lien loan. In calculating the rate spread, the lender usesthe Treasury yield for securities of a comparable maturity as of the fifteenth day of agiven month. Lenders use the fifteenth day of a given month for any loan on which theinterest rate was set on or after that day through the fourteenth day of the next month.The applicable date is the date the interest rate on the loan was determined, generallythe rate lock date. The APR used is the one disclosed to the consumer for purposes ofthe Truth-in-Lending Act (Regulation Z).

4 For example, examination of Hilltop Lending Corporation’s online rate sheet on May 3,2007 (a random selection) showed a prime rate of 6.25% for a 30-year FRM (for aborrower with FICO � 700, 80% LTV, and no late payments) and a subprime rate of8.20% for a 30-year FRM (for a borrower with a FICO � 620, an 80% LTV, and up toone 90-day delinquency in the last 12 months), implying a spread of 195 basis points.

5 See, for example, Scheessele (2002) for an example of HUD analysis.6 Apgar et al. (2007) analyze public HMDA data from 2004 augmented with Census

information and focus on racial differentials in the incidence of APR spread-reportableloans as a function of origination channel, defined as credit union, regulated bank orthrift, or independent mortgage company.

7 Avery et al. (2006), page 144.

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 5 0 7

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

8 Most (over 75%) of subprime loans are ARMs and most non-traditional products areARMs, except for the newer 40- and 50-year fully amortized FRMs.

9 The APR is essentially the bond yield or the internal-rate-of-return (IRR), which maybe defined as that rate which makes the value of the initial cash flow (loan amount lessany points and fees) equal to the present value of all future payments discounted at theIRR.

10 In an earlier version of this paper, the 2006 results were forecasted, estimating that28.3% of loans would be spread-reportable when final figures are released in Fall 2007.

11 LoanPerformance states that their coverage is approximately 80% of the prime market,including all Freddie and Fannie loans, and 50% of subprime market. See www.loanperformance.com for further details on this data source.

12 Note that former Federal Reserve Chairman Alan Greenspan made remarks that werewidely construed as endorsing the use of ARMs during a speech given on February 23,2004.

13 For simplicity, only standard maturities (5, 10, 15, 20, 25, and 30-year) are used for thisexercise. Collectively, these account for over 99% of all loans in the CMC data.

14 Average loan size for home purchase loans reported under HMDA was $179,000 in2004 and $187,000 in 2005 (Exhibit 2).

15 This procedure is equivalent to over-sampling the event of interest, a common approachin logit modeling. Resulting slope coefficients, except for the intercept, are unbiased anda standard method for correcting the intercept is available (Maddala, 1983, page 91).

16 Loans closed by a third party and subsequently purchased by a lender are not subjectto APR spread price reporting requirements.

17 Manufactured housing is not statistically significant in 2004 for retail originations andgovernment-insured is not statistically significant in 2004 for wholesale originations.The year 2005 indicator is not statistically significant in retail originations.

18 Examination of credit score distributions by lien status indicates that second lienborrowers had slightly higher credit scores that first lien borrowers, which may helpexplain this result. Another possibility is that the high correlation between lien statusand loan term swamps any independent junior lien effect.

� R e f e r e n c e s

Apgar, W., A. Bendimerad, and R.S. Essen. Mortgage Market Channels and Fair Lending:An Analysis of HMDA Data. Harvard University, Joint Center for Housing Studies, April25, 2007.

Avery, R., K. Brevoort, and G. Canner. Higher-Priced Home Lending and the 2005 HMDAData. Federal Reserve Bulletin, 2006, Summer, 123–66.

Avery, R., G. Canner, and R. Cook. New Information Reported under HMDA and ItsApplication in Fair Lending Enforcement. Federal Reserve Bulletin, 2005, Summer, 344–94.

Berkovec, J. and P. Zorn. How Complete is HMDA? HMDA Coverage of Freddie MacPurchases. Journal of Real Estate Research, 1996, 11:1, 39–56.

Brueckner, J.K. and J.R. Follain. The Rise and Fall of the ARM: An Econometric Analysisof Mortgage Choice. Review of Economics and Statistics, 1988, 70, 92–102.

5 0 8 � L a C o u r - L i t t l e

Calem, P.K. Gillen, and S. Wachter. The Neighborhood Distribution of Subprime MortgageLending. Journal of Real Estate Finance and Economics, 2004, 29:4, 393–410.

Center for Responsible Lending. 2006. Unfair Lending: The Effect of Race and Ethnicityon the Price of Subprime Mortgages. Report issued May 2006.

Courchane, M., D. Nebhut, and D. Nickerson. Lessons Learned: Statistical Techniques andFair Lending. Journal of Housing Research, 2000, 11:2, 277–95.

Consumer Federation of America. 2006. Subprime Locations: Patterns of GeographicDisparity in Subprime Lending. Report issued September 2006.

Dhillon, U., J.D. Shilling, and C.F. Sirmans. Choosing Between Fixed and Adjustable RateMortgages. Journal of Money, Credit, and Banking, 1987, 19, 260–67.

Epley, D.R., D. Liano, and R.L. Haney, Jr. Borrower Risk Signaling Using Loan-to-ValueRatios. Journal of Real Estate Research, 1996, 11:1, 71–86.

Government Accountability Office (GAO). Alternative Mortgage Products, a Report to theChairman, Subcommittee on Housing and Transportation, Committee on Banking, Housing,and Urban Affairs, U.S. Senate, September 2006.

Hayre, L. (Editor.) Guide to Mortgage-Backed and Asset-Backed Securities. New York:John Wiley & Sons, 2001.

House Financial Services Committee, July 25, 2007. The Subcommittee on Oversight andInvestigations. Hearing on the Home Mortgage Disclosure Act. Available online: http: / /www.house.gov/apps/ list /hearing/financialsvcs dem/ht072507.shtml.

LaCour-Little, M. The Home Mortgage Preferences of Low-and-Moderate IncomeHouseholds. Real Estate Economics, 2007, 35:3, 265–90.

Maddala, G.S. Limited-Dependent and Qualitative Variables in Econometrics. EconometricSociety Monographs in Quantitative Economics. Cambridge: Cambridge University Press,1983.

Moody’s Investor Services. 2006 Review and 2007 Outlook: Alternative-A RMBS, 2007a.

——. 2006 Review and 2007 Outlook: Home Equity ABS, 2007b.

——. 2006 Review and 2007 Outlook: Private Label Jumbo RMBS, 2007c.

The Mortgage Bankers Association (M. Fratantoni, Principal Author). 2007. TheResidential Mortgage Market and Its Economic Context in 2007.

Phillips, R.F. and A. Yezer. Self-Selection and Tests for Bias and Risk in MortgageLending: Can You Price the Mortgage If You Don’t Know the Process? Journal of RealEstate Research, 1996, 11:1, 87–102.

Scheessele, R.M. Black and White Disparities in Subprime Mortgage Refinance Lending.Washington, D.C. Department of Housing and Urban Development, Office of PolicyDevelopment and Research, 2002.

Stanton, R. and N. Wallace. Arm Wrestling: Index Behavior and Prepayment of AdjustableRate Mortgages. Real Estate Economics, 1995, 23, 311–45.

——. The Anatomy of an ARM: Index Dynamics and Adjustable Rate Mortgage Valuation.Journal of Real Estate Finance and Economics, 1999, 19:1, 49–67.

Staten, M. The New Home Pricing Data: What Can They Tell Us About Pricing Fairness?Credit Research Center, Washington, D.C., May 2005.

U.S. Census Bureau. 2006. Current Population Reports P60-231. Income, Poverty, andHealth Insurance Coverage in the United States: 2005.

E c o n o m i c F a c t o r s A f f e c t i n g H o m e M o r t g a g e � 5 0 9

J R E R � V o l . 2 9 � N o . 4 – 2 0 0 7

Vandell, K.D. Handing Over the Keys: A Perspective on Mortgage Default Research.Journal of the American Real Estate and Urban Economics Association, 1993, 21:3, 211–46.

This paper was previously presented at the American Real Estate and UrbanEconomics Association Midyear Meeting in Washington, D.C. in May 2007. Iacknowledge helpful comments from Tony Yezer, Michael Staten, Mike Hollar, JohnWeicher, Bob Aaronson, Glenn Canner, and Bob Avery. The research was supportedby the Consumer Mortgage Coalition. I thank LoanPerformance for allowing accessto its Servicing and Securities data for this project.

Michael LaCour-Little, California State University–Fullerton, Fullerton, CA 92831or mlacour-little@fullerton.edu.