The Eﬀect of Mortgage Broker Licensing On Loan Origination
Standards and Default: Evidence from U.S. Mortgage Market
Department of Economics
University of Washington
April 23, 2012
We study the U.S. origination-to-distribution mortgage ﬁnancing market from the mid 1990s
to the late 2000s. Mortgage loan brokers originated close to two thirds of the mortgage loans in
this period. We examine whether stricter licensing requirements of loan brokers raise lending
standards by i) admitting only higher quality brokers who beneﬁt more from a long-term career
and thus have greater incentives to protect borrowers and lenders’ long-term interests, ii) raising
entry costs and thus generating higher future rents that reduce brokers’ incentives to chase
short-term proﬁts, e.g., by lowering loan origination standards, that jeopardize their likelihood
of winning future business from borrowers and lenders. We exploit the cross-state and over time
variations in licensing requirements and ﬁnd that originated loans in states with more stringent
requirements had higher standards: FICO score were higher, and LTV and DTI were lower and
there were fewer negative amortization, interest only, balloon, ARM, Low Doc, and subprime
loans. The requirements on surety bonds and net worth, education, and oﬃce in state have
the greatest impact on loan origination standards. The education (and exam) requirements for
employees are more eﬀective than those for licensees. The eﬀect of licensing on loan origination
standards is greater for neighborhoods with greater minority percentages and lower income,
and for lenders that specialize in sub-prime lending. Corroborating ﬁndings on loan origination
standards, states with more stringent licensing requirements had lower default rates: Moving
from the 25th to the 75th percentile in licensing requirements is associated with close to 20
percent reduction from the mean of the 90 days or more delinquency rate. These ﬁndings point
to the value of broker licensing when lenders’ incentives to screen are compromised with the
securitization of mortgages.
I would like to thank Yoram Barzel, Bo Becker, Fahad Khalil, Jacques Lawarree, Erica Clower, Troy J. Scott,
Hendrik Wolﬀ, Rong Zhao, and participants at the University of Washington Economics Department brownbags for
their comments. Contact information: Lan Shi, UW Econ, Box 353330, Seattle, WA 98195. Email: email@example.com.
Key words: Mortgage; brokers; securitization; information asymmetry; moral hazard; incen-
tives; occupational licensing
JEL codes: D82; G21; G28; J44; L1
Much has been said about the origins of the U.S. 2008 credit crisis. One prominent development of
U.S. mortgage ﬁnancing market leading to the 2008 crisis is the securitization of mortgage loans.
Starting in mid 1990s, the U.S. mortgage ﬁnancing industry gradually moved from an origination-
and-keep to an origination-to-distribute model. Under the new system, lenders originate loans
and then can sell them in a secondary market. While the potential advantage is more eﬃcient
risk-sharing, many researchers (Keys et al., 2010; Purnanandam, forthcoming) documented that
lenders have less incentives to screen loans.1
According to a 2004 study by Wholesale Access Mortgage Research & Consulting, Inc.,
close to two thirds of all residential loans in the U.S. were originated by brokers. Few papers have
examined the role of mortgage brokers.2 . This paper focuses on brokers. In particular, given that
lenders have little incentives to screen applications and borrowers are less informed than mortgage
brokers, would regulation of loan brokers make a diﬀerence in the loan origination standards? We
examine whether regulation, in the form of licensing requirements, leads to greater eﬃciency in the
loan origination market.
There are several channels that licensing helps raise the standards of loan origination. First
is the selection eﬀect. More stringent requirements raise the bar of becoming a loan broker, and
thus the average quality of loan brokers. For higher quality brokers, a longer-term career is more
valuable, and this reduces their incentives to lower current standards to earn business. Putting it
another way, looser regulations make it easier for brokers aiming to proﬁt from short-term gains to
enter the market. These brokers care little about their reputation, and thus the loan origination
standard is lowered to generate fees by issuing as many loans as possible.3
A related yet distinct mechanism is that by raising the entry barrier, more proﬁt is gener-
ated, which raises the value of staying in the business for the long term and thus reduce brokers’
incentives to pursue short-term proﬁt that might jeopardize their prospect of long-term proﬁts.
Those risky actions include the brokers’ actions to take advantage of borrowers and lenders in the
short-term; over time, borrowers and lenders alike could ﬁnd out that the brokers were not acting in
It appears that the reputation mechanism is not fully functioning between lenders and issuers (the parties that
securitize the debt), and between issuers and investors.
Exceptions are Berndt et al. (2010) and Garmaise (2010).
Facing the pressure from brokers who take up a greater market share by lowering loan standards, other loan
brokers also lower standards to win business, especially in an environment where consumers reward agents with good
their best interests and are less likely to give the brokers future business or referrals. We therefore
predict that loan-origination standards will be higher where licensing is more stringent under the
originate-to-distribute mortgage ﬁnancing model.
We exploit the cross-state variations in licensing requirements to measure the ease of en-
tering the mortgage loan market. Commonly used licensing requirements include surety bonds,
requirements of net worth, work experience, education, and passing exams. We rely on the mea-
surement of licensing requirements by Pahl (2007) who carefully coded the requirements. Licensing
requirements evolve over time. It usually takes several years before an initiative becomes a piece
of law (Kleiner and Todd, 2009).
For the loan origination data, we use Home Mortgage Disclosure Act data, which contains
the details of each loan application, including whether it is originated, turned own, whether it is
sold by the lender, the name of the lender, characteristics of the loan, and characteristics of the
borrower. We also use tract-level Census data which includes various information on census tracts.
We ﬁrst test the eﬀect of mortgage broker licensings on the eﬃciency of mortgage loan
origination market at the national level. We ﬁrst document that loan brokers’ entry to the market
is aﬀected by the licensing regulation. Second, we ﬁnd states that sell a greater percentage of
their loans pass a greater percentage of loan applications, suggesting that lenders in states where
investors are further away from the loan-issuers use looser standards in issuing loans. Third, states
that have more stringent requirements pass a lower percentage of loan applications, and those loans
have a lower loan amount/applicant income ratio. Fourth, among the requirements, those on surety
bonds and net worth, education level of employees, and oﬃce in state have the greatest impact on
the standards of loan origination.
If the licensing is eﬀective in raising loan origination standards, it should lead to two things.
First, requirements to have a loan passed are more strict. We thus test whether FICO score, LTV
(loan to value ratio), and DTI (debt to income ratio) vary with the licensing stringency. Second,
brokers are less likely to use predatory terms, including interest only, negative amortization, balloon
payment, ARM, no/low-doc, and subprime loans, etc. We thus test whether the percentage of loans
having these terms is lower in states with greater licensing requirements. We ﬁnd evidence that
supports the predictions.
We then conduct tract-level loan origination analyses. First, we conﬁrm our ﬁndings
from the state-level analysis. Second, we test whether tracts that have more severe information
asymmetry problems experience lower loan standards, and whether the eﬀect of licensing is greater.
We measure the degree of information asymmetry between the broker and the borrower using the
tract-level percentage of minority population. Our ﬁndings conﬁrm the predictions.
We also conduct lender-level analyses. First, lender-level analysis conﬁrms our ﬁndings
from the state-level analysis. The Department of Housing and Urban Development maintains a list
of lenders that specialize in subprime lending. We ﬁnd that the eﬀect of licensing requirements on
loan origination standards is greater for lenders that specialize in subprime lending.
Since information on the loan application is not complete, one worries about the omitted
variable problem. We thus follow up the loan origination analyses with loan performance analyses.
If loan origination standards were higher, it should lead to better performance for loans in states
with more stringent licensing requirements. First, we show that loan performance varies with
the loan terms. Second, we show that loan performance is better; 90 days or more (90+ days)
delinquency rate is lower in states with more stringent requirements. The economic magnitude is
large; moving from the 25th to the 75th percentile in the licensing requirements is associated with
close to 20 percent decrease in the 90+ days delinquency rate from the mean. Third, the eﬀect of
licensing on loan performance is larger for ARM, low doc, and subprime loans.
We investigate whether the ﬁnding of lower origination rates in states with more stringent
licensing requirements is due to omitted state-level laws on mortgage lending. In particular, many
states adopted anti-predatory lending (APL) laws in 2000s to combat predatory mortgage lending.
We ﬁnd that introducing APL laws does not aﬀect our results.
The closest paper is Kleiner and Todd (2009), who examine the eﬀect of licensing require-
ments on the labor market outcome for brokers and loan quality for consumers. Their focus is on
the usefulness of diﬀerent licensing requirements, in particular, surety bonds. Our paper diﬀers
from theirs in two ways. First, we examine the impact of all types of requirements, and the impact
of licensing on diﬀerent levels of loan brokerage ﬁrms (the licensees or the employees). Second, our
data allow us to shed light on the mechanism of how licensing helps improve eﬃciency. Last, we
examine the loan performance. Our ﬁndings that surety bonds/net worth and education require-
ments matter (especially for employees) suggest that both the charter value mechanism and the
channel of raising the quality of players are at work.
Other related works include Dell’Ariccia et al. (2009) who study the relationship between
credit expansion and loan-issuance standards at the MSA level, Becker and Milbourn (2011) who
ﬁnd that the entry of a third credit-rating agency reduces the quality of credit-rating agencies,
Garmaise (2010) who examine the relationship dynamics between mortgage brokers and lenders,
and Jiang et al. (2010) who ﬁnd that brokers issue lower quality loans than lender employees.
Keys et al. (2009) show that the ﬁnding that securitization leads to lenders’ reduced incentives to
screen is mainly for subprime low-doc loans.
We provide background information in Section 2, develop hypotheses in Section 3, present
identiﬁcation strategy and econometric speciﬁcation in Section 4, introduce data in Section 5,
examine loan origination in Section 6, conduct analysis of loan performance in Section 7, explored
alternative hypotheses in Section 8, discuss further analyses in Section 9, and conclude in Section
2 Background; Institutional Details
2.1 Securitization of Loans by GSEs and Private Label Issuers
In the U.S., the Federal government created several programs, or government sponsored entities, to
foster home ownership. These programs include the Government National Mortgage Association
(known as Ginnie Mae), the Federal National Mortgage Association (known as Fannie Mae) and the
Federal Home Loan Mortgage Corporation (known as Freddie Mac). These government sponsored
enterprises (GSEs) purchase conforming loans from lenders, allowing lenders to originate more
The GSEs can securitize these loans, and these are known as mortgage-backed securities
(MBS). Whether or not a loan is conforming depends on the size and a set of guidelines. Non-
conforming mortgage loans which cannot be sold to Fannie or Freddie are either “jumbo” or
“subprime”, and can be packaged into mortgage-backed securities and sold to investors by private
label issuers, underwritten by investment banks. Since 1970s, markets developed for mortgage-
backed securities. For example, the securitized share of subprime mortgages increased from 54
percent in 2001 to 75 percent in 2006.
The development of secondary markets has many beneﬁts. First, conversion of mortgages
into mortgage-backed securities permits a better distribution of the risk of holding mortgages.
Second, mortgage-backed securities are also “liquid” while mortgages themselves are not. Several
important checks are designed to prevent mortgage fraud, the ﬁrst being the due diligence of the
issuer. In addition, the originator typically makes a number of representations and warranties about
the borrower and the underwriting process. When these are violated, the originator generally must
repurchase the problem loans.4
2.2 Players in the Market
The lenders are commercial banks, thrifts (which include savings and loan associations and savings
banks), credit unions, and mortgage banks in the United States. The ﬁrst three are depository
institutions. States and the federal government each issue bank charters. State-chartered banks
operate under state supervision. National banks are chartered and regulated by the Oﬃce of
the Comptroller of the Currency (OCC), a division of the Treasury Department. Oﬃce of Thrift
Supervision (OTS) charters and supervises thrifts. Mortgage banks specialize in originating and/or
servicing mortgage loans, and are state-licensed.
A mortgage company does not take deposits from consumers or businesses. Their two
primary sources of revenue are loan origination fees and loan servicing fees (provided they are a
Issuers underwrite the sale of securities backed by the pool of subprime mortgage loans to investors. An important
information asymmetry also exists between issuers and investors.
loan servicer).5 Mortgage companies dominated the US market towards the end of 2007. However,
many of the large mortgage companies, such as Wells Fargo Mortgage, are aﬃliated with large
commercial banks or bank holding corporations.
2.3 The Use of Employees and Brokers by Lenders
In the U.S., the process by which a mortgage is secured by a borrower is called origination. This
involves the borrower submitting a loan application and documentation related to his/her ﬁnancial
history and/or credit history to the lender. Loans are originated in two ways. One route is by
lender employees (loan oﬃcers), called retail lending. The other is that lenders set aside a wholesale
branch, to which independent brokers have access. A broker establishes wholesale relations with
several lenders. Brokers serve the role of bringing the applicants and collecting information, and
the lenders provide the funds. Most lenders have both employees and wholesale lending programs.
To expand business, it is often more expedient to use brokers.6
2.3.1 The Arrangement Between Lenders and Brokers
Brokers are compensated for their services in two ways. On the one hand they receive fees paid
directly by the borrower. These include the loan origination fee, credit fee, etc. In addition, the
broker is paid a yield spread premium by the lender. A yield spread premium is the fee paid by a
lender to a broker for higher-rate loans (Federal Reserve Press Reserve. Dec. 18, 2007).7
Lenders provide brokers diﬀerent compensations for diﬀerent types of loans. The types of
loans vary by the credit score of borrowers (prime/jumbo, Alt-A, or subprime), whether the rate
is ﬁxed (FRM vs ARM), the payment option (interest-only, balloon, negative amortization), and
documentation requirements (low doc vs full doc). Because of the higher rates and greater proﬁts
that the lenders can receive from certain types of loans in the secondary market, lenders often give
higher rebates to brokers for those types of loans. The fee schedule by the lenders give brokers
three kinds incentives. The ﬁrst is the incentive to charge above-market rates. The second is the
incentive to steer borrowers to the types of loans that give brokers the highest commission. The
third, of course, is to close many deals as possible since they are paid by commission.
By selling them shortly after they are closed and funded, they are eligible for earning a service release premium.
A third way to lend is via correspondent lending programs (other mortgage banks, commercial banks, Savings
and Loans, and credit unions.) Correspondent banks are usually small and receive funding from wholesale lenders.
They diﬀer from brokers in that they fund the loans.
Lenders pay brokers rebates on high-rate loans, and charge points on low-rate loans. Points are upfront payments
from borrowers to the lender, and rebates are upfront payments from the lender to the brokers. For example, on
November 7, 2007, wholesale lenders quoted the following prices to brokers for a 30-year ﬁxed-rate mortgage: 6% at
zero points, 5.75% at 1.25 points, 6.25% at 1 point rebate, and 6.50% at 2 points rebate. This means that the lender
wants to be paid 1.25% of the loan amount upfront for 5.75% loans, and will pay 1% or 2% rebates upfront for 6.25%
and 6.5% loans, respectively. Brokers can choose to use part of the rebates to cover borrowers’ closing costs.
Brokers do need to compete with other brokers and lenders to ﬁnd borrowers. Therefore
there is a trade-oﬀ. If the broker charges a higher rate, it is more proﬁtable for himself and the
lender, but he is less likely to obtain the business from the borrower. However, when the borrowers
do not shop around for rates, which is made diﬃcult by the various kinds of charges across loans, or
when they do not have the ability to truly understand the risks and costs of various loan products,
borrowers can be led by the broker into loans that gives the broker the highest commission instead
of what best meets her needs.
2.4 Regulations in the Loan Origination Market
The ﬁve agencies — OCC, FRS, FDIC, OTS, and NCUA have regulatory and supervisory respon-
sibility over national banks and their subsidiaries, member state-chartered banks and their sub-
sidiaries and subsidiaries of bank holding corporations, non-member state-chartered banks, savings
and loans institutions and their subsidiaries, and credit unions, respectively. Compared with the
depository-taking lenders, non-depository mortgage lenders are lightly regulated, at the state level
(Keys et al., 2009).
2.4.1 Regulation of Loan Brokers
At the federal level, loan origination is regulated by laws including the Truth in Lending Act and
the Real Estate Settlement Procedures Act (1974). Credit scores are often used, and these must
comply with the Fair Credit Reporting Act. The extent of the regulation, often in the form of
licensing, depends on the state.
Commonly used methods are i) education requirements, e.g., number of hours of in-class
training, ii) work experience requirements, iii) passing exams, and iv) posting surety bonds, or
meeting net-worth requirements, or both. Surety bonds are a common practice. These bonds func-
tion as a guarantee from a Surety to a government and its constituents (obligee) that a company
(principal) will comply with an underlying statute, state law, municipal ordinance, or regulation.
The principal will pay a premium (usually annually) in exchange for the bonding company’s ﬁnan-
cial strength to extend surety credit. In the event of a claim, the surety will investigate it. If it
turns out to be a valid claim, the surety will pay it and then turn to the principal for reimbursement
of the amount paid on the claim and any legal fees incurred. Because of this, the most important
thing a surety looks for in the principal is the net worth. Some states require surety bonds, while
others require net worth. We thus create a variable that is the sum of both.8
A typical clause regarding education is “The applicant (for Mortgage Broker Individual license) must have
completed 24 hours of classroom education on primary and subordinate ﬁnancing transactions and the laws and
rules.” A typical clause regarding examination is “The applicant must pass a written test adopted by the department
that is designed to determine competency in primary and subordinate mortgage ﬁnancing transactions as well as to
test knowledge of the laws and the rules”.
Licensing is implemented at various levels of the mortgage broker ﬁrms. Financial re-
quirements are certainly at the ﬁrm level. For the qualiﬁcation requirements, some states impose
requirements only at the applicant (licensee) level, some include requirements for the managing
principal (who actively runs the brokerage ﬁrm), and some further include requirements at the
employee level. Pahl (2007) does a very careful job in coding the various requirements for each
party. We will use the sum of all requirements, the speciﬁc requirements, and requirements at
3 Hypothesis Development
Under an origination-and-sell business model, for loans that will be sold, lenders’ proﬁt depends
on the quantity and proﬁt rate of loans that will be sold to issuers who then package and sell them
to investors. Since lenders do not hold the loans, they care less about the re-payment risk. Indeed,
Keys et al. (2010) and Purnanandam (forthcoming) show that lenders have reduced incentives to
screen loans than in an origination-and-keep model.
There are two ways brokers may aﬀect the loan origination outcome. First, they can
inaccurately record or omit or misrepresent data, or intentionally falsify information. Second, they
can win the trust of the lenders and then reduce their loan screening eﬀorts. The eﬀect of broker
screening on loan origination outcome is particularly true for low-doc or no-doc loans. For in these
loans, documentation of income is not required by the lender; often, stated income is used. Yet,
brokers may have some or can elect to collect soft information on the borrowers.10 Their inactivity
or possible falsiﬁcations aﬀect the loan origination outcome.11
In the preceding Section 2.3.1, we examined the contracts between brokers and lenders.
Given the contracts, we predict that brokers have incentives to i) increase earnings by closing
more deals, which can be more easily achieved by lowering loan origination standards, and ii) steer
borrowers to the types of loan products that give them the greatest compensation instead of what
maximizes the borrowers’ interests. In such a environment, we argue that licensing can play a role
in improving eﬃciency in the loan origination market. There are two mechanisms that licensing
may improve eﬃciency.
First, stricter licensing prevents the less qualiﬁed individuals from entering the market.
While loan brokers are regulated, employees (loan originators) of regulated ﬁnancial institutions or insurance
companies, lawyers, and real estate agents were exempt. Licenses can be suspended or revoked.
Garmaise (2011) shows that brokers ﬁrst established certain reputation with the lenders and then decreased
their loan-screening eﬀorts. Over time, brokers present more loans to the bank, and these loans have higher default
Jiang et al. (2011) show that the broker channel is worse among low-doc loans. Correspondent brokers have
stronger reputation concerns due to their exclusive or long-term relationships with lenders. They also show that
lenders do not factor risk into pricing.
Less qualiﬁed brokers are less likely to survive, while high ability ones are more likely to survive.
Thus the latter beneﬁt more from a long-term reputation, which encourages them to pursue clients’
long-term interests, or prevents them from exploiting the clients and pursuing short-term proﬁts
in the form of charging excessive fees or rates, or using loan terms that facilitates closing without
regard to borrowers’ ability to repay.
Second, there is the eﬀect of concern for future proﬁts (charter-value argument by Keeley
(1990) and Hellmann et al (2000). With greater entry costs comes greater proﬁt and higher
future proﬁt staying in the industry. By lowering lending standards or engaging improper lending
practices, brokers may gain more business currently, but their lenders and clients would ﬁnd out
their practices (over time) and are less likely to fund loan applications or seek loans brokered by
them. This loss of future proﬁt is greater when competition in the loan origination market is
smaller and the equilibrium proﬁt greater. That is, more stringent licensing requirements improve
the brokers’ incentive to uphold borrowers’ long-term interest. Summing up, licensing enhances
the broker selection process, as well as improves the incentives for brokers.
We thus predict that the loan origination quality (standard) is higher in states with more
stringent licensing requirements. Among the various means of licensing, education and exam
requirements are likely more diﬃcult to satisfy, especially for employees. Therefore, to examine
whether the selection mechanism is at work, we shall test whether the eﬀect of education and
exam requirements for employees have a particularly greater eﬀect. Practitioners point out that
the greatest entry barrier is surety bonds/net worth. Therefore, to examine whether the charter
value mechanism is at work, we will test whether the eﬀect of licensing is particularly greater for
Since borrowers diﬀer in their susceptibility to fall for brokers’ self-serving behavior, we
predict that the eﬀect of licensing is greater for less informed borrowers. Along the same line of
reasoning, lenders diﬀer in their emphasis on sub-prime lending market. We therefore predict that
the eﬀect of licensing is greater for lenders targeting less informed borrowers — sub-prime borrowers.
4 Identiﬁcation and Econometric Speciﬁcation
The identiﬁcation strategy is the state-level variation in licensing regulations. Our baseline speci-
P ct originatedarea = β lic Licst + β i Xi area + β l Xl area + β h Xh area + β c Xc area + β h m Xh m + εarea ,
where i refers to the borrower, l is the loan, h is the house, c is the community level, st represents
state, h m refers to housing market, pct originated is the percent of loan applications that are
originated, Licst is the licensing requirements at state-level, Xi is the borrower-level variables,
including their race, gender, and income, Xl is the loan-level variables, including loan amount, the
loan purpose (home-purchase or re-ﬁnance), whether it is a ﬁrst lien loan, and the type of the loan
(guaranteed by government agencies or not), Xh is the house-level variables, including the census
tract it is in, whether it is a single home, multi-family home, or manufactured home, and Xh m are
variables capturing the local housing market, e.g., the average house price change, and the area
can be state or census tracts. If the control variables fully capture the relevant hard information
used by brokers in brokering the loan, brokers in states with greater licensing requirements will
use soft information to a large degree, which should lead to lower likelihood of origination. For
example, brokers in states with more stringent requirements are less likely to falsify documents for
no-doc loans, and thus loan origination likelihood is lower. That is, we test whether β lic < 0.
A further test will be to look at the loan performance. If loan origination standards were
lowered, it should lead to worse performance of the originated loans. We will examine loan defaults
in Section 7.
One concern with this identiﬁcation strategy is that states might use regulations to adjust
to changing economic environments. However, we take comfort in the fact that it usually takes
years for a legislative initiative to become law. Therefore, the licensing requirements are likely ex-
ogenous. In addition, if there is any remaining endogeneity, it is likely that states that experienced
predatory lendings (higher origination rate), for example, adopt more stringent licensing require-
ments, creating a possible positive correlation between the dependent variable and the licensing
variables. Yet our hypothesis is that the coeﬃcient on the licensing variable is negative. Therefore,
this mechanism, if present, biases us against ﬁnding β lic < 0.
It is possible that licensing requirements impacts the quality of loan applications; antici-
pating that brokers and lenders are more diligent in screening, applicants of lower quality are less
likely to apply. If there is such an eﬀect, this should lead to a positive relation between licensing
requirements and the likelihood of origination, i.e., this mechanism would bias us against ﬁnding
the hypothesized negative eﬀect of licensing on loan origination likelihood.
Diﬀerent lenders attract loan applications of diﬀerent quality. This consideration suggests
that we should include lender-type dummies to control for unobservable loan application quality. In
addition, the rich data allow us to further test the prediction by exploiting variation in borrowers’
degree of being informed proxied by the tract-level minority percentage.
We also test the prediction by directly examining the loan characteristics, including FICO
scores and LTV, and loan terms, including negative amortization, interest only, balloon payment,
and no-doc. More stringent requirements should lead to better loan characteristics and terms.
Finally, assuming state characteristics do not change over time, including state ﬁxed eﬀects would
take care of the unobserved heterogeneity across states.
5 Data and Summary Statistics
Our loan origination data set comes from HMDA (Home Mortgage Disclosure Act) 1996-2009,
which provides loan-application-level data. The Home Mortgage Disclosure Act (HMDA) was en-
acted by Congress in 1975 and is implemented by the Federal Reserve Board’s Regulation C. For
example, in 2010 almost 19.5 million loan records for calendar year 2009 were reported by 8,124
institutions. HMDA applies to most depository institutions (commercial banks, savings associa-
tions, and credit unions) with home or branch oﬃces in MSAs. The 2004 revision required more
information to be disclosed, including pricing if the treasury rate is exceeded by a certain amount.
HMDA also extends to mortgage and consumer ﬁnance companies, whether such companies are
independent, subsidiaries of banking institutions, or aﬃliates of bank or thrift institution holding
It is estimated that the more than 8,800 lenders currently covered by the law account
for approximately 80% of all home lending nationwide. Because of its expansive coverage, the
HMDA data likely provide a representative picture of most home lending in the United States.
Depository institutions account for the bulk of the reporting institutions, but mortgage companies
report the majority of the applications and originations. In 2005, for example, nearly 80% of the
8,850 reporting institutions were depository institutions, but together they reported only 37% of
all the lending-related activity. Mortgage companies accounted for 63% of all the reported lending;
70% of these institutions were independent and not related in any way to a depository institution.
Appendix A oﬀers more details on constructing the sample. The summary statistics show that
about one in ﬁve loan applications is denied, while about one-fourth of all loans are extended by
We control for changes in the economic environment for the relevant area by including
variables that have been shown to be good predictors of loan denial decisions at the individual
level (see Munnell et al., 1996). In particular, we include a measure of house price appreciation to
take into account the role of collateral.
For across-state variation in regulation, we use Pahl (2007). Data on the number of brokers
comes from the Occupation Employment Survey, which can be categorized by state. Census-level
data provide tract-level data on race, including the percent of minority population.
Appendix B shows how the variables are deﬁned. For example, for surety bond require-
ments, a requirement of bond amount over $50,000 is codes 3, $20,000-$30,000 is coded 2, under
$25,000 is coded 1, and none is coded 0. States diﬀer greatly in how strict their licensing re-
quirements are. The top panel of Table 1B lists the 5 states that score the highest in the various
requirements, while the bottom panel lists the states that score the lowest in the various require-
ments.12 For the overall requirements, FL and CA are among the top 5 most stringent states. For
Loan oﬃcers of lenders work under the umbrella license of their employers. Depository institutions are usually
surety bond/net worth, New Jersey and WI are among the top 5. For the exam requirement and
the education requirement, FL and CA are consistently the top 2 states. The mean value of the
indicator variable for surety bonds is 1.2, and the mean for net-worth is .46. In 166 out of the
561 state-years, the surety bonds requirement is absent. In 450 out of the 561 state-years, the
net-worth requirement is absent. Figure 1 shows the average licensing requirement over time. It is
apparent that states became more stringent in their requirements.
Coded licensing information is only available for 50 states and D.C. area and for years
1996-2006. We obtained HMDA data for 1997 and 2003-2009. Therefore the data that merged
HMDA and licensing data covers years 1997 and 2003-2006.
The loan performance data encompass all securitized loans issued by private label parties,
including commercial banks (Wells Fargo, Chase, etc.), mortgage companies (Countrywide), thrifts
(WaMu), and investment banks (Merrill Lynch, Bears Stern, etc.).
The sample for securitized loans diﬀers from that for non-securitized loans in potentially
two ways. First, having to meet the requirements for securitization, they likely have higher FICO
scores. Second, given that they can be securitized, lenders (and brokers) collect less information
on their repayment risk (since they will not keep the loans.) These two forces work in opposite
directions, therefore the bias from focusing on securitized loans only should be limited.
6 Econometric Analyses of Loan Originations
6.1 State-level Analyses of Entry of Loan Brokers and Oﬃcers
First, we examine whether looser licensing regulations lead to easier entry of loan brokers, partic-
ularly with respect to hot loan market. In California, for example, the number of newly licensed
brokers increased from 1, 000 in 1996 to 7, 000 in 2006, and then started to decrease (shown in
Figure 2). We obtain employment and earning data from the Occupation Employment Survey,
1999-2010. The data does not distinguish the speciﬁc occupations of loan brokers. Instead, it covers
occupations listed as loan counselling, loan interviewers and clerks, and loan oﬃcers. Loan counsel-
lors provide guidance to prospective loan applicants who have problems qualifying for traditional
loans. Loan interviewers and clerks interview loan applicants to elicit information. Specializing in
commercial, consumer, or mortgage loans, loan oﬃcers include mortgage loan oﬃcers and agents,
collection analysts, loan servicing oﬃcers, and loan underwriters. These occupations likely include
both loan brokers and loans oﬃcers working as employees in lenders.
It is well understood that entry and exit also respond to the house price change. We thus
more heavily regulated than mortgage broker ﬁrms. Often, mortgage ﬁrms are similarly regulated as mortgage broker
estimate an equation of the below form:
Emp grst,t = β 1 licst + h s price appreciationst,t−1 + εst ,
where emp gr is the growth in employment for the three occupations listed above, lic is the various
licensing requirements, hs refers to house, and st represents state. The regression results are in
Table 2A. We ﬁnd that the employment growth is negatively associated with the various require-
ments, especially those for education, exams, and surety bonds and net worth requirements. But
the estimates border on statistical insigniﬁcance. We also tried to use the lagged regulation vari-
ables, which appear to produce a better ﬁt. The most prominent variable in aﬀecting employment
growth is the bond and networth requirement.13 These pieces of evidence suggest that licensing
aﬀects entry.14 15
6.2 State-level Origination Rates
We examine the loan origination outcome at the state-level. We examine the percentage of loans
that are originated among applications. Speciﬁcally, we estimate an equation of the below form:
P ct originatedst, yr = β 1 licst, yr + β 2 selling rtst + β 3 loan amt/incomest,yr
+β 4 h s price changest,yr + αt + εst,yr ,
where lic is the various licensing variables, st refers to state, yr refers to year, selling rt is the
percent of loans that are not kept by the lenders, hs price change is the change of house price
over the past year, and αt are year ﬁxed eﬀects.16 The origination rate depends on many factors.
For example, states that have more stringent requirements likely have lower quality loans, and
consequently should be turned down. Therefore, in all regressions, we include the state-level
average of loan amount/income.
We conduct the analyses in two ways. First, we run the regression every year. The results
for individual years are reported in columns 3-7 of Table 2B. Across the years, we ﬁnd that the
coeﬃcient on the percent sold is positive, which is consistent with the claim that lenders who do
not hold the loans have less incentives to screen the loans. For our key variables, i.e., the summary
This is consistent with an article on surety bonds being the true entry barrier for loan brokers: (“Largest Barrier
to Mortgage Licensing” by Integrity Mortgage Licensing).
We ﬁnd that the earning increase is higher in less regulated states. Since we also ﬁnd that entry is easier in less
regulated states, the ﬁnding that earning increases are higher in less regulated states suggests that the entry did not
totally drive away the proﬁts.
More stringent licensing requirements on brokers may lead loan professionals to become loan oﬃcers instead.
Thus, we think our results are a lower bound of the true eﬀect of licensing on broker entry.
We also included other state-level variables, like GDP per capita. We ﬁnd that the most signiﬁcant state-level
economic variable is house price change.
variable for licensing requirements, the coeﬃcient is consistently negative in each year of 2003-2006.
The coeﬃcient on the house price change is positive, although it is not always signiﬁcant.
We then pool all years of data together, and the results are in columns 1-2. For this
speciﬁcation, we add year ﬁxed eﬀects. Since standard errors are likely correlated over time within
a state, we cluster them at the state level. The coeﬃcient on the summary licensing requirement
variable is signiﬁcantly negative. We ﬁnd that the coeﬃcient on the summary licensing requirement
variable from 2003 to 2006 to be −0.0053 and statistically signiﬁcant. The 25th percentile value of
the summary licensing variable is 4 and the 75 percentile is 9. The estimated coeﬃcient suggests
that, compared with states that are at the 25th percentile value of the summary licensing variable,
states that are at the 75th percentile have .0053 ∗ (9 − 4) = 2.7 percentage point lower origination
We conduct a couple of robustness checks. For instance, CA and FL have stringent laws,
and their origination rates are low. We check whether CA and FL dominate our results, and ﬁnd
that they do not. Second, the eﬀect of licensing requirements might take time to show. We thus
use the lagged value of the summary licensing variable. The results are in column 2. We ﬁnd that
the coeﬃcient changes little. Third, we use the sample where all loan applications (including those
purchased by lenders) and use the percent originated or purchased as the dependent variables.
The results are in column 8; the coeﬃcient on the summary licensing variable is of slightly lower
6.3 Which Speciﬁc Requirements Matter?
Table 2C reports results examining the individual requirements. Instead of the summary require-
ments, we include the requirements for passing exams, education, work experience, continued
education, the sum of surety bonds and net-worth requirements, and ﬁnally the requirement that
a lender that operates a branch in another state must have an oﬃce in that state. All are indicator
variables, as deﬁned in Appendix B.
Column 1 reports results using all years from 2003 to 2006. We ﬁnd that the coeﬃcient
on education is −.021 and statistically signiﬁcant. The 25th percentile value of the education
requirement is 0 and the 75th percentile is 1. The estimated coeﬃcient suggests that, compared
with states that are at the 25th percentile value of the education requirement, states that are at
the 75th percentile have .021 ∗ (1 − 0) = 2.1 percent lower origination rates.17 We ﬁnd that the
coeﬃcients on the requirement of passing exams and work experience are negative but statistically
Education requirements include two kinds, one being a certain number, usually 24-40 hours of classroom educa-
tion, and the other being a degree in related ﬁelds, e.g., ﬁnance, real estate, business, etc. No states speciﬁcally require
earning of a formal degree; several states specify that a formal degree can be used as a substitute for requirements
of exams or work experience.
We also ﬁnd that the coeﬃcients on net-worth plus surety bonds requirements is −.0088
and statistically signiﬁcant. The 25th percentile value of the sum of the net-worth and surety
bonds requirements is 1 and the 75th percentile is 2. The estimated coeﬃcient suggests that,
compared with states that are at the 25th percentile value of the sum of net-worth and surety
bonds requirements, states that are at the 75th percentile have .0088 ∗ (2 − 1) = .88 percent lower
6.3.1 Licensing on Oﬃce in State
A recent phenomena is that there arose many internet ﬁrms that oﬀered loan brokerage services,
ﬁrms like Quiken Loan. Licensing entities have regulated such brokerage service branches. For
example, the state of IL requires brokers to have in-state oﬃces. One state regulates that the
requirement of in-state oﬃces can be waived if brokers put up extra bonds. In the state of WA,
for example, in 2010 close to half of all loan originators were from out of state.19
A requirement of in-state oﬃces increases the cost of doing business in a state for out of
state broker ﬁrms. It is likely this reduces entry of brokerage ﬁrms and raises the rent for current
brokers, and thus they have less incentives to pursue current proﬁts by reducing lending standards.
We include a dummy variable that takes the value of 1 if state laws require in-state oﬃces for
mortgage loan brokerage ﬁrms. The regression results are in column 7 of Table 2C. We ﬁnd that
the coeﬃcient on the variable is negative, yet its statistical signiﬁcance is borderline. The scale
of internet lending picks up over time, and we thus focus on latest year in the data — 2006. The
results are in column 8 of Table 2C. We ﬁnd that the coeﬃcient on the variable to be −.022 and
signiﬁcant at the 7% level. The estimate is economically signiﬁcant. The magnitude suggests that
compared with a state that is at the 25th percentile value of the requirement (0), a state that is
at the 75th percentile (1) has 2.2% lower origination rates.
We plan to examine the eﬀect of the amount of surety bond; instead of category variables,
we use the raw amount. In particular, we plan to use the ratio of surety bond over the local
6.4 Regulation on Licensee- or Employee-level
What is the most eﬀective way to regulate mortgage loan brokerage ﬁrms, licensing the applicants
(licensees), managing principals (the one who is actively running the ﬁrm), or employees?20 To
The coeﬃcient on house price change is .16. The 25th percentile of house price growth is .05 and the 75th
percentile is .11. The coeﬃcient suggests that moving from 25th to 75th percentile of the house price growth value
is associated with .16 ∗ (.11 − .05) = 1 percent higher origination rate.
The information is at http://raincityguide.com/2011/02/01/loan-originator-mortgage-broker-and-consumer-
Miami Herald published an article showing how criminals jump in to become loan orignators to feed the lending
frenzy. FL regulates the mortgage brokers only at the licensee level, not at the loan originator level.
examine this, we sum all requirements on each party, creating variables called licensee requirement,
managing principal requirement, and employee requirement. The results are in column 1 of Table
2D. We ﬁnd that the requirement for licensees has the greatest impact. The −.013 coeﬃcient
suggests that compared with states at the 25th percentile (0) of the licensee requirement, the state
that is at the 75th percentile (1) has a 1.3% lower origination rate.
We focus on the requirement for education, i.e., the education requirements for licensee,
managing principals, and employees.21 In columns 2-4, we report the coeﬃcient on the education
requirement at each level. Among the three speciﬁcations, the coeﬃcient on employee education
requirement has the greatest magnitude. The coeﬃcient on employee education (−.033) suggests
that one standard deviation in employee education (.34) is associated with a 1.1 percent lower
origination rate. We notice that the requirement for employee education and employee exams is
highly correlated (.67). For the requirement on exams, the eﬀect is greatest for licensee, although
all of them are insigniﬁcant.22
6.5 Working on Identiﬁcation
We explore the sensitivity of results to the alternative identiﬁcation strategy of using within-MSA
cross-state variation. The idea is that areas across two or more states yet within one MSA are
integrated economies and the omitted variable problem is minimized; variation in licensing across
states can then identify the eﬀect of broker licensing on loan origination standards. However, we
note that many MSAs have asymmetric presence across states; in many cases, the majority of the
MSA economy is in one state. Therefore comparing the economies across the two states (within
one MSA) could capture the city vs town eﬀect. We therefore control for all possible variables that
capture economic environment, like tract-level income, minority percentage, etc. We are currently
working on it.
6.5.1 Over-time Variation in Licensing
States vary in various and possibly unobservable dimensions; therefore the issue of omitted variables
is always a concern for any state-level analyses. The ﬁnding that the origination ratio is lower in
high licensing states could be because that loans in those states are worse than in other states,
and thus the origination ratio is low. Including state ﬁxed eﬀects helps address this. If for some
Summary statistics for speciﬁc requirements at diﬀerent levels of the ﬁrm are as follows. We create dummy
variables that take the value of 1 if a state has these requirements. From 1997 to 2006, for education requirements,
the mean value of the licensee requirement is .14, the employee requirement is .097, and the managing principal
requirement is .063. For work experience requirement, the mean value of the licensee requirement is .20, the em-
ployee requirement is .02, and the manager requirement is .127. For exam requirements, the mean value of licensee
requirement is .043, the employee requirement is .1, and managing principal’s requirement if .092.
The eﬀect of licensees’ work experience requirement on loan origination standards is mostly negative, while the
eﬀect of employee work experience requirement is signiﬁcantly positive.
reasons, states consistently have worse loans over time, state ﬁxed eﬀects can capture that. The
over-time change in licensing is then used to identify the eﬀect. The econometric speciﬁcation is
P ct originatedst, yr = β 1 licst, yr + β 2 selling rtst + β 3 loan amt/incomest,yr
+β 4 h s price changest,yr + αi + αt + εst,yr ,
where αi are state ﬁxed eﬀects.
Regression results are in Table 2D. We ﬁnd that the coeﬃcient on the summary licensing
requirement variable is mostly insigniﬁcant. We also ﬁnd that the coeﬃcient on the exam require-
ment is negative in OLS regression. The insigniﬁcant coeﬃcient on the licensing variable could be
due to the possibility that loans in states with higher licensing requirements began to have better
characteristics, which raises their likelihood of being originated.
The eﬀect of licensing shows in two ways. First, brokers would turn down loans that they
deem too risky, which shows up in lower origination rates. Second, those risky loans could only be
originated by using risky terms, like negative amortization (negam), ARM, no doc, interest only
(io) etc. In turning down those risky loans, the use of risky terms, in originated loans, should be
down. In the following subsection, we test this prediction using data on characteristics and terms
6.6 The Eﬀect of Licensing Requirements on Characteristics and Terms of Orig-
Data on loan characteristics and terms are available only for loans that are originated and se-
curitized by private label issuers. We ask whether characteristics of originated loans were worse
in states that had looser license regulation. Critical loan characteristics include many: Loan to
Value ratio (LTV), debt to income (DTI), FICO (credit) scores, whether the loan is ARM or ﬁxed
rate, whether the loan has negative amortization (negam), whether the loan is IO (interest only),
whether the loan has a balloon payment towards the end of duration, and whether the loan is
no-doc or full-doc.
The loan performance data provide information on various characteristics of loans and
their origination year. We focus on loans originated 2000-2007. For each characteristics, we use
as dependent variable the proportion of loans, in terms of originated loan value, that has the focal
characteristics, at state level for each year, i.e., we estimate an equation of the below form:
P ct loan value with f eatureit = βlicit + αt + αi + εit,
where i is state, t is origination year, αt are year ﬁxed eﬀects, and αi are state ﬁxed eﬀects where
Shown in Table 2E, the (origination value) weighted average state-level LTV ratio increased
in 2004-06. The higher the LTV ratio, the more likely a borrow can withstand a possible decline
in house price and continue to have equity in the house. We regress the state-level LTV ratio on
the state licensing requirements. Column 1 of Table 2F shows regression results. In results with
state ﬁxed eﬀects, states with higher summary licensing requirements have lower LTV ratio; work
experience requirement has the most signiﬁcant impact among all requirements.
The (origination value) weighted average FICO score dipped in 2004-06 (shown in Table
2E). The higher the FICO score, the less likely that the borrower will be delinquent in payment or
walk away from the mortgage and damage his/her credit score. We regress the state-level average
FICO score on the state licensing requirements. Column 2 of Table 2F shows regression results
with state ﬁxed eﬀects. States with higher summary licensing requirements had higher FICO (the
estimated coeﬃcient on FICO is .61). In particular, requirements on employees and the requirement
of oﬃce in state are associated with originated loans having higher FICO scores.
A debt-to-income ratio (abbreviated as DTI) is the percentage of a consumer’s monthly
gross income that goes toward paying debts, including principal and interest payment, certain
taxes, fees, and insurance premiums. The ratio depends on the loan amount, the term of the loan,
the interest rate, and the type of the loan. The higher the ratio, the more likely the borrower
faces the risk of being unable to repay. The mean of DTI for loans originated 2000-07 is 37.75.
Column 4 shows results from regressing the state-year value of (loan value) weighted DTI on the
licensing requirement variable and a house price appreciation variable. We ﬁnd that states with
more stringent requirements have lower DTI ratio for their originated loans.
The mean of the dummy variable for non negam is .96 across 2000-07. Column 5 shows
regression results using the speciﬁcation with state ﬁxed eﬀects. States with higher summary
licensing requirements have higher proportion of loans that are non negam; the estimated coeﬃcient
is .0023, signiﬁcant at 5-percent level. Individual dimension of licensing has insigniﬁcant impact.
The proportion of loans that are non-IO drastically decreased 2002-2007: It was .98 in
2002, .75 in 2005, and .67 in 2007. Regression results for IO, in column 6 of Table 2F, are unlike
what we ﬁnd for other loan characteristics.
Prior to 2004, ﬁve percent loans have the balloon payment feature. In 2005-06, more than
10 percent are balloon payment loans. Regression results, in column 7, show that states with higher
summary license requirements have higher percentage of loans that are non balloon. Oﬃce-in-state
requirement has signiﬁcant impact, so does the employee requirement. Licensee requirement in
fact raises the proportion of balloon loans, and so does the work experience requirement.
Loan can be originated by brokers or loan oﬃcers of lenders. The eﬀect of the broker
licensing should be greater for brokers. Unfortunately, eighty seven percent of loans have no data
on whether brokers or loan oﬃcers originated the loans. Below we examine other three important
loan terms in more details.
6.6.1 Eﬀect of Broker Licensing on Loans Being ARM
Fixed-rate mortgages and adjustable-rate mortgages (ARMs) are the two primary mortgage types.
Hybrid ARM is a form of ARM that has a ﬁxed interest rate for a couple of years and then after the
re-set date, start to have variable rates. Often 3/27 or 2/28 are used, and substantial interest rate
risk is imposed on borrowers. The rate is often lower than FRM since borrowers bear part of the
interest rate risk. The success of ARM with these terms hinge on the availability of re-ﬁnancing.
Option ARM allows the borrower to choose their own pay methods: minimum (negative
amortization, negam), interest only, balloon payment (30-year loan with 40 years amortization,
hence a large payment at the maturity date). The minimum payment on an option ARM can
jump dramatically if its unpaid principal balance hits the maximum limit on negative amortization
(typically 110% to 125% of the original loan amount). If that happens, the next minimum monthly
payment will be at a level that would fully amortize the ARM over its remaining term. In addition,
option ARMs typically have automatic “recast” dates (often every ﬁfth year) when the payment
is adjusted to get the ARM back on pace to amortize the ARM in full over its remaining term.
Borrowers often pay relatively less in initial years, which makes it a term that can be easily abused
in loan originations.
Table 2G shows results from examining the percent of loans that are ﬁxed rate. We ﬁnd that
in the speciﬁcation with state ﬁxed eﬀects, states with higher licensing requirements have greater
proportion of loans being ﬁxed-rate. Among the various requirements, the exam requirement raises
the ratio being ﬁxed-rate. Education requirement reduces it. Requirements on licensees raise the
percent of ﬁxed-rate loan more than the employee requirement. The results are robust to the
inclusion of the house price appreciation over the minimum of the past four years. In addition, the
eﬀect is also true for the percent of all loan counts (instead of loan amount) that are ﬁxed-rate.
The eﬀect is particularly strong for the period 2002-07.
The results using OLS speciﬁcation are diﬀerent. Upon examination, we ﬁnd that CA,
FL, and NV have both high license requirements and high ratio of ARM. Excluding CA, FL, and
NV in the OLS speciﬁcation makes the coeﬃcient on the license variable to be negative, consistent
with the results from the speciﬁcation with state ﬁxed eﬀects.
6.6.2 Eﬀect of Licensing on Loans Being Low Doc
Loans diﬀer in their level of documentation. They are categorized into full, low, or no doc. Full
doc loans have documentation on income and asset. Low doc loans have documentation on asset
only, and no doc loans have no documentation on either the asset or income and are rare. Low doc
loans are attractive to self-employed people whose income can be erratic. Lenders do check FICO,
and may charge higher rates.
For 2000-07, the sum of loan origination amount is $833 billion for full-doc, $1000 billion
for low-doc, and $50.7 billion for no doc loans. Therefore the true diﬀerence is between full doc and
low doc. Table 2E shows that the percentage of home purchase loans that are full doc decreased
from .68 in 2001 to .35 in 2007. There was no apparent increase in the percent of self-employed
people.23 One apparent explanation is the lowering of loan origination standards to get loans
Regression results with state ﬁxed eﬀects are in column 8 of Table 2F. The results are
consistent with the overall big picture: States with higher summary license requirements have
greater percentage of loans that are full doc. Oﬃce in state requirement has impact, and so does
exam requirement. Work experience requirement reduces the proportion of loans that are full doc,
6.6.3 Eﬀect of Licensing on Loans Being Sub-Prime
Loans can be roughly categorized into three types, prime (conforming prime & Jumbo), Alt-A, and
subprime. GSEs are the major party that buys conforming prime loans. Jumbo are prime loans
except that their amount are above the conforming loan criteria and GSEs are not allowed to buy
them. Alt-A loans are to borrowers who have good credit record but the loans are underwritten
with aggressive terms, like high LTV ratio or low doc, etc. Sub-prime loans are loans to borrowers
with poor credit records.
For loans originated in 2000-07 that were securitized by private label issuers, $1.14 trillion
were Jumbo or prime loans, $1.5 trillion were Alt-A, and $1.85 trillion were subprime. Years 2005-
06 were the peak years of securitization. There is often a quarter lag between the loan origination
and the securitization.
We examine whether licensing regulation leads to loan origination of higher quality — we test
whether the percent of subprime loans is less in states with more stringent licensing requirements.
The unit of observation is state-year. Column (9) of Table 2F shows the results.
Our dependent variable is the percent of originated loans in 2000-07 that are prime. The
year dummies are included, so are the house price appreciation over the minimum of the past 4
years and the year ﬁxed eﬀects. The coeﬃcient on the summary licensing requirement variable is
.0038, signiﬁcant at 5 percent level. The mean of the dependent variable is .26. That is, moving
from the 25th to the 75th percentile in licensing requirement is associated with .0038*5=.019, a 7.3
percent increase from the mean of the percent of loans that are prime; this is evidence that stricter
licensing of brokers is associated with granting loans to borrowers with greater creditworthiness.
Among them, CA and FL are among the states that experienced the largest reduction. We also ﬁnd that large
states, in particular, CA and FL, had lower than average (often lower than .50) full doc ratio, and they later had
higher delinquency rate. Note that these state experienced high increase in house price and therefore low housing
6.6.4 Eﬀect of Licensing on Loans Being Sub-Prime and Low Doc
Keys et al. (2010) show that the eﬀect of securitization on the lessened lenders’ incentives to
screen is true mainly for subprime low doc loans. Their ﬁndings suggest that it is for subprime
low doc loans that lenders have lessened incentives to collect soft information due to the ease of
While prime and low doc loans (ALT A) are given to borrower with good credit history
albeit with aggressive terms, non-prime and full doc loans give the lenders some protection by
checking for information, sub-prime and low doc loans are loans that are potentially rife with
manipulations, omission, misrepresentation, or even forgery. The percent of private label securitized
loans that are sub-prime and low doc increased from being 10 percent in 2000 to 20 percent in
2005-2006 before falling back to 10 percent in 2007.
Columns 10-11 of Table 2F show regression results where the state-year value of the per-
cent of securitized loans that are subprime and low doc is regressed on the state-year licensing
requirement variable, a house price appreciation variable, and state ﬁxed eﬀects. For 2000-07, the
coeﬃcient on the licensing variable is -.0022. This means that moving from 25th (4) to 75th (9)
percentile in licensing is associated with .0022*5=.011 percentage point reduction in the percent of
subprime and low-doc loans, a 8 percent decrease from the mean of the dependent variable (.14).
The results for 2003-06 are very close.
We note that using pooled OLS mainly leads to positive or insigniﬁcant coeﬃcient on the
summary licensing variable. We conduct preliminary county-level analysis. We ﬁnd that counties
in states with more stringent experience requirement have lower delinquency rate.
In sum, the previous sub-sections provide state-level analyses on the relation between loan
origination standards and state licensing strictness from various angles. However, we are concerned
that possible omitted variables can explain the correlation. We thus explore further variations that
can help tell apart our hypothesis from alternative explanations. In the two sub-sections below,
we explore the variation ﬁrst at the tract-level, and then at the lender-level.
6.7 Tract-level Analyses of Loan Origination
Our hypothesis predicts that the eﬀect of mortgage broker licensing on loan origination standard
is greater when the client (the borrower) is less able to evaluate the broker’s suggestions. The pay
arrangement between loan brokers and lenders is such that brokers’ pay increases with the number
of loans closed and the interest rate charged. Borrowers who are less informed are at a greater
danger of being misled by brokers to loan products that maximize the brokers’ pay instead of the
borrowers’ best interests. We hypothesize that licensing is more helpful for borrowers who do not
have the ability to see through loan brokers’ recommendations.
We argue that borrowers who are less educated are less able to protect their own interests.
However, the borrower-level education variable is not present in HMDA data. Education level is
closely correlated to minority status.
HMDA provides information on the tract that the borrower resides in, and tracts diﬀer in
the minority percentage. We thus measure the degree of un-informedness by using several variables
at the tract-level: the minority percentage in the tract, and the percentage of low income families
in the tract. The tract-level data are available from the 2000 Census.
Table 3A provides summary statistics of tract-level data from 2007. There were 65,515
census tracts; the median number of population is 4, 043, the median minority percent is 19 percent,
and the median HUD income is $59, 100. Our current analysis uses the sample that includes loans
that were purchased, so we examine the percent of loans that were originated or purchased, termed
We ﬁrst estimate an equation of the following speciﬁcation:
P ct issuedtract, year = β 1 licensing rqmtstate, year + β 2 percent soldtract,year +
β 3 pct minoritytract, year + β 4 licensing rqmtstate, year ∗ pct minoritytract, year + εtract,state,yr .
Results are in Table 3B. All standard errors are clustered at state county level. Column
1 introduces only the three speciﬁc licensing requirements: those on education, exam, and work
experience. Column 2 further introduces the tract-level information, column 3 introduces the
interaction term, and column 4 uses the summary licensing requirement variable and its interaction
with the minority percentage variable.
Across the columns, we conﬁrm the ﬁnding from state-level analysis that more stringent
requirements, particularly those for education, are associated with lower issuance rates.24 Tracts
that have higher percentages of minorities have lower percentages of loan applications being is-
sued. For the variable of interest — licensing rqmtstate, year ∗pct minoritytract, year , the coeﬃcient
is insigniﬁcant. It is likely that tracts diﬀer in unobservable ways. In the following analysis, we
therefore examine the over time change in issuance rates.
6.7.1 Tract-level Change in Issuance Rate
We examine the change in issuance rate at the tract-level. The percent issued in 2004 was .51 and
for 2005 was .49 (Table 1). Conﬁrming this, Table 3A shows that across tracts, the median of the
growth rate is .96.
A diﬀerent way to examine the credit expansion is to look at the growth of the originated
amount.25 Table 3C reports the results. The results are for home purchase loans in 2004-05. We
We also ﬁnd that the requirement for exams is now negatively associated with the issuance rate.
It is worth pointing out that it could be eﬃcient to have disproportional credit expansion in low income neigh-
borhoods, but it should not be at the cost of lowering loan standards.
ﬁnd that i) high minority percentage tracts experienced greater increase in loan originations, and
ii) this eﬀect is smaller in states that have more stringent licensing requirements.26
6.8 Lender-level Analyses of Loan Origination
We conduct lender-level analyses. Table 4A provides some summary statistics. We observe that
the number of lenders increased from 1997 to 2005, and then decreased. Second, there are huge
variations in the size of the lenders. For example, row 1 shows the number of lender-level loan
applications. In year 2005, for example, the mean is 1, 843, and the median is 69. Row 2 shows
the mean and median value of lender-state. On average, a lender has presence in 5 states, and the
median value is 2 states.
It appears that lenders of diﬀering sizes have diﬀering origination rates. For example, in
2005, lenders that had a larger than median number of loan application had .74 origination rate,
and lenders that had a smaller number of loan applications than the median had a .83 origination
It is also possible that diﬀerent licensing requirements attract certain lenders. We therefore
use an econometric speciﬁcation that includes lender ﬁxed eﬀects; therefore the coeﬃcient on
the policy variable captures the eﬀect of within-lender across-state licensing requirements on the
loan application outcome. We cluster the standard error at the lender level. That is, our basic
econometric speciﬁcation is
Ylender,state,year = β L Licensestate + β X Xlender,state,year + αlender + εlender,state,year .
The results are in Table 4B. Column 1 includes the summary licensing variable only, and
the coeﬃcient on the variable is signiﬁcantly negative. Column 2 introduces the four categories of
licensing requirements: passing exams, education, work experience, and the sum of bond and net
worth. We ﬁnd that the coeﬃcients on exams and education are signiﬁcantly negative, conﬁrming
the evidence found in state-level analysis. The coeﬃcient on the sum of bond and networth is
negative, but its statistical signiﬁcance is borderline. Column 3 introduces the other two licensing
variables, continuing education (contedu) and branch in-state requirements, and also includes other
across-lender variables, like the proportion of applicants who are male, the proportion of ﬁrst lien
loans, and the proportion of houses that are owner occupied. We ﬁnd that the coeﬃcient on
contedu is signiﬁcantly negative and the coeﬃcient on the branch-in-state requirement is negative
Column 4 uses the indicator variable for the employee exam requirement, and column 5 uses
the indicator variable for the licensee exam requirement. It appears that the eﬀect of exams mainly
comes from the employee requirement, conﬁrming our ﬁndings in tract-level analysis. Column 6
These ﬁndings are also present in 2003-04.
reports results from year 2005 that uses the summary licensing variable. The results are utterly
the same as in 2006.
Other variables are reasonably estimated. For example, the states that have a higher
holding rate see higher standards within a lender. States that have greater house price increases
see a lower standard within the same lender. Overall, ﬁndings at the lender-level are consistent
with those from the state-level.
6.8.1 Subprime Lenders vs Non-subprime Lenders
HUD (Housing and Urban Development) maintains a list of subprime lenders from the mid-1990s
to mid-2000s. We merge the yearly list of subprime lenders with the yearly HMDA data. There
were 211 lenders on the list. When we merge the list with the HMDA data set, we ﬁnd that 189
of the listed subprime lenders appear in the HMDA data set. These lenders on average originate
in 20 states. The median lenders originate in 14 states.
We examine whether the eﬀect of licensing on loan origination standards is stronger for
the list of subprime lenders. Our current analysis uses the sample that includes loans that are
purchased, so we examine the percent of loans that are originated or purchased (issued). Results
for years 2004 and 2005 are in Table 4C.27 The dependent variable is the percent of loans that
are originated or purchased (loans issued) among all loan applications at the lender level. We
ﬁrst introduce the dummy variable for subprime-lender status, and then the interaction variable of
subprime-lender and the licensing variables. Since being a subprime lender or not is a time-invariant
characteristics of a lender, we drop lender ﬁxed eﬀects. We ﬁnd that i) the education requirement
is negatively associated with the issuance rate, ii) being a sub-prime lender is associated with a
lower issuance rate (possibly because of the lower quality loans that the lenders tend to attract),
and iii) the eﬀect of the education requirement on loan origination standards is stronger for lenders
that are on the HUD subprime lender list.28
While the preceding results are consistent with higher origination standards in states with
more stringent licensing requirements, omitted variable bias remains a concern — could the lower
origination rate in a state a reﬂection of the worse loan applications, which prompts the state to
strengthen its licensing? If our hypothesis that licensing raises loan origination standards is right,
loans originated in heavily licensed states would perform better. Yet if the alternative hypothesis
is right, heavily licensed states would have worse loan applications and worse originated loans, and
thus worse performance. The following section examines loan performance.
We are working on including results from other years.
We conduct robustness checks. First, we are concerned that the list is incomplete; therefore using a dummy
variable and implementing an interaction term might not be the best way. We try focusing on data that belong to
subprime lenders. We ﬁnd that the coeﬃcients on the summary licensing variable, or education, passing exams, and
work experience are similar to the baseline results, and are slightly higher. Second, we exclude lenders that issue
loans soly in one state and the results change little.
7 Econometric Analyses of Loan Performance
7.1 Data on Loan Performance
The loan performance summary reports report performance of mortgage loans that are securitized
by private label issuers, i.e., non-agency securitizers. GSEs, enjoying implicit backing from the
U.S. government, have advantage in securitizing prime loans. Shown in Table 1A, a very high
percentage of loans were securitized by private label issuers. For example, as of Dec. 2008, 53
percent of all home purchase loans originated in 2005 were securitized by private label issuers.
7.2 Statel-level Panel Analysis of Loan Performance and Licensing
We ﬁrst examine whether loan performance varies with loan characteristics. We measure loan
performance using the percent of loan value that is OTS 90 days or more (90+) delinquent (past due
with no payments being made) as of December 2008 for loans originated 2000-2007. Table 5A shows
the 90+ delinquency rate varies with the presence of various loan characteristics. ARM loans have
10 percentage point higher delinquency rate than ﬁxed rate loans. Loans with negative amortization
have close to 6 percentage point higher delinquency rate than loans without it. Interest-only loans
have close to 3 percentage higher delinquency rate than non-IO loans. Loans with balloon payment
have close to 6 percentage higher delinquency rate. And low-doc loans have two percentage point
higher delinquency rate than full-doc loans.
Subprime loans have much poorer performance than Alt-A and prime loans; the 90+ days
delinquency rate (as of Dec. 2008) for securitized loans that were originated 2000-07 was 2.7
percent for prime loans, and 19.3 for subprime loans.29 No-doc subprime loans are the worst, at
22.3 percent. These numbers suggest that whether the loan is subprime or not plays a bigger role
in loan performance than whether the loan is full or low doc.
In addition, regression results in Table 5B show that loans with higher FICO scores and
lower DTI have lower delinquency rate. FICO scores have larger impact than LTV or DTI on
loan performance. Having shown that loan performance depends on loan terms and borrower
characteristics, which loan broker licensing has an impact on (shown in Section 6.6), below we test
whether broker licensing aﬀects loan performance.
The dependent variable is percent of loans that are 90+ delinquent as of December 2008
for loans originated 2000-07. The econometric speciﬁcation is as follows:
loan perofrmancest = β p hs pr ch heightst + β u pct unemployst + β l licen sin g reqst + εst ,
where s refers to state, t refers to year, hs pr ch height is percent change in state-level Oﬃce of
Federal Housing Enterprise Oversight (OFHEO) price index of the current price relative to the
These statistics correspond well with Ashcraft and Schuermann (2008).
maximum in the past 4 years, pct unemploy is percent unemployed in the state as of December
Results are in Table 5C. OLS regression (in column 1) yields positive sign on the summary
license requirement variable. States might diﬀer in unobservable ways, which might aﬀect the
license requirements and the dependent variable. One way to capture it is to include state ﬁxed
eﬀects; the over-time variation is then exploited.30
We focus on years 2003-06. During this period, 45 percent of the states experienced a
change in the licensing regulation, and the changes happened on average in January 2005. The
summary licensing requirements in states that experienced a change had a median value of 6 prior
to the change, and so did states that had no change.31 This is the ﬁrst piece of evidence that states
that did change the requirements do not diﬀer from those that did not in observable ways.
To examine whether stayers and changers have diﬀerent trend prior to the change, we ﬁrst
use 90+ delinquency rate in Dec. 2008 as the loan performance measure. We ﬁnd that changers
had somewhat higher delinquency rates than stayers (those that did not change in licensing). This
is true for the same year loan performance (performance for loans originated in the same year).
These patterns show up both in graph and in tests for equality.
Important to us is that changers do not exhibit a reduced upward trend in delinquency
rates than stayers. Therefore we do not think the found estimate is due to pre-existing reduced
upward trend for changers. Our speciﬁcation is as follows:
loan perofrmancest = β p hs pr ch heightst + β u pct unemployst + β l licen sin g reqst + αs + αt + εst .
Shown in Table 5C, results of panel regressions with state ﬁxed eﬀects show that increases
in license requirements are associated with reduced delinquency rate, measured as 90 days or more
late in payments. The estimated coeﬃcient on the summary licensing requirement variable is −.30.
The 25th percentile value of summary licensing variable is 4 and the 75 percentile is 9, and the
mean delinquency rate 90+ is 12.3 percent for loans originated during 2003-06. That is, moving
from the 25th to the 75th percentile in licensing is associated with .30 ∗ 5 = 1.5 percentage point
reduction in delinquency rate, which is 12 percent reduction from the mean.
For a sub-sample analysis, we examine house purchase loans originated in 2003-2006. The
coeﬃcient is −.46. The 25th to the 75th percentile of the summary licensing variable is 4 to 9.
The mean dependent variable value is 12.6 percent. That is, moving from the 25th to the 75th
percent is associated with .46∗ 5 = 2.3 percentage point decrease in delinquency rate, an 18 percent
decrease from the mean.32
More than half of the states changed their licensing regulation. Most of the changes took place around 2002,
before the housing and lending market boom took oﬀ.
The value is 8 for states after the change.
For all loans, including home purchase, reﬁ, and home equity loans during 2003-2006, the coeﬃcient for the
Mortgage loans, if they default, usually do within 4-5 years after origination, therefore
a downside of using delinquency rate as of December 2008 for loans originated in years 2000-07
is that loans originated earlier may have passed their most active years of default. We therefore
focused on loans originated 2003-06. An alternative approach is to use the loan performance for
loans originated within a certain period.
We therefore use as the dependent variable the year end performance for loans originated
in the same year. The mean of the dependent variable for loans originated during 2000-07 is
1.45 percent. The results are in Table 5D. The coeﬃcient is −.052, which implies that moving
from the 25th (3) to the 75th (8) percentile of the summary licensing requirements translates into
.052 ∗(8 −3) = .26 percentage point reduction, an 18 percent decrease from the mean. If we restrict
us to loans originated during 2003-06. The coeﬃcient is −.04.This implies that moving from the
25th (4) to the 75th (9) in licensing requirements is associated with .04 ∗ (9 − 4) = .20 percentage
point reduction in 90+ delinquency rate, a 22 percent decrease from the mean (.90 percent).
Results on the eﬀect of individual licensing requirements are in columns 4-5. We see that
education and experience requirements matter. The coeﬃcient on education requirement is −.39,
i.e., moving from 0 (the 25th percentile) to 1 (the 75th percentile) in education requirement is
associated with .39 percentage point reduction in delinquency 90+ rate, which is a 26.9 percent
reduction from the mean (1.45 percent). Requirements on employees appear to matter. The
coeﬃcient on the employee requirements is −.13, i.e., moving from the 25th percentile (0) to the
75th percentile (2) is associated with .13∗2 = .26 percentage point reduction in the 90+ delinquency
rate, an 18 percent decrease from the mean (1.45 percent).3334
7.3 Eﬀect of Licensing Is Ampliﬁed for Certain Loan Terms
7.3.1 Full vs Low Doc
If licensing matters, brokers in states with more stringent requirements, due to both the selec-
tion and the incentive eﬀects, are less likely to abuse no-doc loans. No-doc loans in those states
would experience better performance than those in states whose brokers abuse the terms to proﬁt
themselves without regard to borrowers’ or lenders’ interests.
We thus test whether the eﬀect of licensing is greater for loan terms where brokers have
greater room in using to entice borrowers, in particular, low or no doc loans. We use house purchase
summary licensing requirement variable is -.30. The mean of dependent variable is 12.3. That is, moving from the
25th to the 75th percentile in summary licensing requirements is associated with .30*5=1.5 percentage point decrease
in the delinquency rate, a 12 percent decrease from the mean.
Results are better for delinquency 90+ and seriously delinquency (90 or more days past due or currently in
foreclosure ) than for foreclosure (not shown).
We note in regression without state ﬁxed eﬀects, the negative eﬀect mainly comes from the work experience
loans during 2003-06 and include HPI and unemployment rate as control variables. We ﬁnd that
for low-doc loans, the coeﬃcient on the summary licensing variable is negative at −.61, signiﬁcant
at 1-percent level. The magnitude of the coeﬃcient suggests that moving from the 25th to the 75th
percentile in licensing is associated with .61∗5 = 3.05 percentage point lower 90+ delinquency rate,
a 22 percent reduction from the mean of the dependent variable (13.7 percent). For full-doc, the
coeﬃcient is -.31 and signiﬁcant at 5-percent level. That is, the eﬀect of licensing on delinquency
rates for home purchase loans is particularly large for low doc loans.
7.3.2 ARM vs Fixed Rate Loans
ARM is a term where brokers can more easily steer borrowers into closing a loan, often by using
other potentially predatory terms or terms that borrowers do not fully or bother to comprehend
including negative amortization, interest only, and balloon payment options. The licensing of
brokers is hypothesized to reduce the abuse. We thus hypothesize that ARM loans in more stringent
states would perform better than ARM loans in less stringent states.
We use the current year 90+ delinquency rate data for loans originated in the same year.
Results are in columns 6-9 of Table 5D. The coeﬃcient on the summary licensing requirement
variable for ARM loans is −.08 and statistically signiﬁcant. The mean of the dependent variable
is 2.15 percent. That is, moving from the 25th (3) to the 75th (8) percentile for the licensing
requirements is associated with .08 ∗ (8 − 3) = .4 percentage point reduction in delinquency rate,
a 19 percent decrease from the mean. The coeﬃcient on the licensing variable for ﬁxed rate
loans is insigniﬁcant. In summary,we ﬁnd that the negative coeﬃcient on the summary licensing
requirement variable is mainly driven by the eﬀect for ARM loans. Shown in columns 8-9, the
eﬀect of individual requirements are also greater for ARM loans than for all loans together (in
7.3.3 Subprime vs Non-subprime
We test whether the eﬀect of licensing on loan origination standards is greater for subprime loans.
Columns 1-4 of Table 5E reports regression results. Column 1 shows OLS results and columns
2-4 include states ﬁxed eﬀects. The unit of observation is state-year-loan type. Columns 1-2
cover all loans while columns 3-4 only subprime loans. We ﬁnd that i) subprime loans have much
higher 90+ delinquency rate, and ii) states with more stringent licensing requirements have lower
90+ delinquency rate for subprime loans, especially in 2003-06. The coeﬃcient on the summary
licensing variable is −.40, which implies that moving from the 25th to the 75th percentile in
licensing requirements is associated with .40 ∗ 5 = 2 percentage point lower delinquency rate, a 9.5
percent decrease from the mean of the dependent variable (.21).
7.3.4 Subprime Low Doc Loans
The last two columns of Table 5E report regression results for the sample of subprime and low
doc loans. For the period 2003-06, the coeﬃcient on the summary licensing variable is -.40, which
implies that moving from the 25th to the 75th percentile in licensing requirements is associated
with .40 ∗ 5 = 2 percentage point lower delinquency rate, a 8.3 percent decrease from the mean of
the dependent variable (.24).
The licensing of brokers is hypothesized to change the selection and incentives of brokers.
We thus use the data to exam whether the eﬀect of licensing is stronger for broker-originated loans
than for retail loans. Unfortunately, a major part of the data miss information on the origination
channel. Nevertheless, for the limited data that has information on origination channel, results are
shown in the last column of Table 5C. We ﬁnd that for 2003-2006, the eﬀect is negative for broker
originated loans. It was absent for retail loans (not shown).
In summary, this section provides evidence on the eﬀect of broker licensing on the loan
performance. We ﬁnd that more stringent licensing requirements are associated with better per-
formance of originated loans, and the channel is that fewer loans with features that are positively
correlated with defaults were originated.
8 Alternative Explanations
8.1 Federal and State Regulations of Anti-Predatory Lending
Federal agencies have applied provisions of laws to seek redress for consumers who have been
victims of predatory lending. Among the most frequently used laws are TILA, HOEPA, the Real
Estate Settlement Procedures Act (RESPA), and the FTC Act. Congress has also given certain
federal agencies responsibility for writing regulations that implement these laws. For example, the
Board writes Regulation Z, which implements TILA and HOEPA, and HUD writes Regulation X,
which implements RESPA.
Truth in Lending Act (TILA), which became law in 1968, was designed to provide con-
sumers with accurate information about the cost of credit. In 1994, Congress enacted the HOEPA
amendments to TILA in response to concerns about predatory lending. First, it places restrictions
on loans that exceed certain rate or fee thresholds, which the Board can adjust within certain
limits prescribed in the law. For these loans, the law restricts prepayment penalties, prohibits
balloon payments for loans with terms of less than 5 years, prohibits negative amortization, and
contains certain other restrictions on loan terms or payments. Second, HOEPA prohibits lenders
from routinely making loans without regard to the borrower’s ability to repay. RESPA, passed in
1974, seeks to protect consumers from unnecessarily high charges in the settlement of residential
mortgages by requiring lenders to disclose details of the costs of settling a loan and by prohibiting
certain other costs. The FTC Act, enacted in 1914 and amended on numerous occasions, provides
the FTC with the authority to prohibit and take action against unfair or deceptive acts or practices
in or aﬀecting commerce.
However, with the exception of loans covered under HOEPA, there are no federal statutes
that expressly prohibit making a loan that a borrower will likely be unable to repay. In response
to the lack of protection of consumers in mortgage lending, many states adopted anti-predatory
lending laws, which are often more stricter than those at the federal level.
In 1999, North Carolina passed the ﬁrst comprehensive state law that was modeled after
the federal HOEPA (mini-HOEPA law). Prompted by growing concerns over the explosion in
subprime lending, many other states also enacted anti-predatory lending laws. As of 2007, 29
states and the District of Columbia had mini-HOEPA laws in eﬀect and another 14 states had
some types of older anti-predatory lending laws that were still in eﬀect, which were adopted prior
to 2000 and restricted prepayment penalties, balloon payments, or negative amortization for all
mortgages only (Bostic et al., 2008a).
These laws, which could potential correlate with state licensing of brokers, may aﬀect the
loan origination, potentially causing a spurious relation between the broker licensing and loan
origination standards. In the loan performance regression, we thus include the coded APL law
data from Ding et al. (2010), and also the house price change from the loan origination year to
Loan perfit = β a AP Lit + β p h p 2008 over orig yrit + β l licit + εit
Results are in Table 6. State-level panel analyses ﬁnd that APL has insigniﬁcant eﬀect on
loan performance, and states with stronger licensing requirements have lower delinquency rate,
i.e., including APL laws does not aﬀect our results.
9 Further Analyses
9.1 Cost-beneﬁt Analysis
According to the Board of Federal Reserve, the residential mortgage loan outstanding in 2007
is $11.9 trillion. Moving from the 25th to the 75th percentile of licensing requirements leads to
.30*5=1.5 percentage decrease in 90+ delinquency rate,35 i.e., if the whole nation moved from the
25th to the 75th percentile in licensing requirements, the nation would have 11.9*1.5/100= $.18
trillion= $180 billion fewer loans that are 90 days or more delinquent.
Using 2004 data, US has 13m loan applications, 132k being the average loan amount and
.68 being the origination rate. This means in year 2004, 13m ∗ 132k ∗ .69 ∗ 1.5/100 = $18 billion
The number .30 is from column 3 of Table 5C.
can be saved from being 90+ delinquent if the nation moves from the 25th to the 75th percentile
in licensing requirements. Considering the loan outstanding is an accumulation of loans made over
time, these two numbers are roughly consistent.
According to a 2004 study by Wholesale Access Mortgage Research & Consulting, Inc.,
there are approximately 53,000 mortgage brokerage companies that employ an estimated 418,700
employees and originate 68 percent of all residential loans in the U.S. In 2005, the per capita
income was $35, 452, i.e., $17.7/hr. Assume the in-class room education is 40 hours, the maximum
lost productivity (and earning) is 418, 700 ∗ 40 ∗ 17.7 = $296m = $.3b per year. This back-of-the-
envelope calculation suggests that there is huge welfare gain from using more stringent licensing
10 Discussion and Concluding Remarks
This paper examines whether licensing regulation of mortgage loan brokers improves loan origi-
nation outcomes when the lenders have little incentives to screen loans in an origination-to-sell
mortgage ﬁnancing model. The mechanisms are i) more stringent licensing requirements screen
out low quality brokers, who place a lower value on future proﬁts, and are more inclined to pursue
current proﬁts by using lax standards in issuing loans, and ii) more stringent licensing requirements
block entry, raising the long-term proﬁts, which reduces the brokers’ incentives to take risks, by
lowering lending standards in current periods, that jeopardize their rate of survival into the future.
We ﬁnd that states that use more stringent licensing requirements have higher lending standards
and better loan characteristics and terms that facilitate better loan performance. Two require-
ments, one for bonding and net worth, and the other for education (and exam) requirements, have
the greatest eﬀect on loan origination standards. The education (and exam) requirements are most
eﬀective at the employee level. The eﬀect of licensing is greater in neighborhoods with higher
minority percentages, and for lenders that focus on sub-prime lending. Finally, states that have
more stringent requirements have lower default rates for their loans.
Federal legislation started to license mortgage brokers at national level. The Secure and
Fair Enforcement for Mortgage Licensing Act of 2008 (SAFE Act), enacted on July 30, 2008,
sets minimum licensing and registration standards for individual mortgage loan originators. Also,
the ﬁnancial reform legislation passed in 2010—the Dodd-Frank Wall Street Reform and Consumer
Protection Act—prohibits YSPs. The evidence shown in this paper provides support to these
regulatory actions. Remaining questions related to the role of brokers abound. For example, how
does the relation between brokers and lenders evolve? Should brokers have ﬁduciary duty to their
Note that in normal business cycles, there is variation in lending standards, as higher house prices in booming
periods make reﬁnancing easier in the future, which reduces current lending standards (Rajan, 1994). In my study,
the major factor is the securitization of mortgage loans.
borrower customers? We plan to take up these issues in future research.
Ashcraft, Adam B. & Til Schuermann (2008), “Understanding the Securitization of Subprime
Mortgage Credit,” Federal Reserve Bank of New York, Staﬀ Reports no. 318.
Becker, Bo, and Todd Milbourn (2011), “How Did Increased Competition Aﬀect Credit Rat-
ings?” Journal of Financial Economics, 101 (1).
Berndt, Antje, Burton Holliﬁeld, and Patrik Sandas (2010), “The Role of Mortgage Brokers in
the Subprime Crisis,” Working paper, Carnegie Mellon University, Tepper School of Business.
Board of Governors of the Federal Reserve System, “Mortgage Debt Outstanding”
Campbell, John and Joao Cocco (2011), “A Model of Mortgage Default,” Harvard Econ
Choi, Hyun-Soo (2011), “The Impact of Anti-Predatory Lending Laws on Mortgage Volume,”
Working paper, Econ. Dept., Princeton University.
Dell’Ariccia, Giovanni, Deniz Igan, Luc Laeven (2009), “Credit Booms and Lending Standards:
Evidence from the Subprime Mortgage Market,” European Banking Center Discussion Paper No.
Ding, Lei, Roberto G. Quercia, Carolina Reid, and Alan White (2010), “The Impact of State
Anti-Predatory Lending Laws on the Foreclosure Crisis”
Federal Reserve Press Release. Dec 18, 2007.
United States General Accounting Oﬃce (GAO) (2004), “Consumer Protection: Federal and
State Agencies Face Challenges in Combating Predatory Lending,” GAO 04-280.
Garmaise, Mark (2010) “After the Honeymoon: Relationship Dynamics Between Mortgage
Brokers and Banks,” mimeo, UCLA Anderson.
Haggman, Matthew, Rob Barry, and Jack Dolan, “Thousands with criminal records work un-
licensed as loan originators,” Miami Herald.
Hellmann, Thomas F., Kevin C. Murdock, and Joseph E. Stiglitz (2000), “Liberalization, Moral
Hazard in Banking, and Prudential Regulation: Are Capital Requirements Enough?,” American
Economic Review, 90(1): 147—165.
Ho, Giang and Anthony Pennington-Cross (2006), “The Impact of Local Predatory Lending
Laws on the Flow of Subprime Credit,” Journal of Urban Economics 60: 210—228.
Ho, Giang and Anthony Pennington-Cross (2007), “The Varying Eﬀects of Predatory Lending
Laws on High-Cost Mortgage Applications” St. Louis Federal Reserve Review, January/February
2007, 89(1), pp. 39-59.
Jiang, Wei, Ashlyn Aiko Nelson, and Edward Vytlacil (2010), “Liar’s Loan? Eﬀects of Origi-
nation Channel and Information Falsiﬁcation on Mortgage Delinquency”. February.
Keeley, Michael C. (1990), “Deposit Insurance, Risk, and Market Power in Banking,” American
Economic Review, 80 (5), 1183-1200.
Keys et al. (2009), “Financial Regulation and Securitization: Evidence from Subprime loans”
Journal of Monetary Economics, 56: 700—720.
Keys, Benjamin J., Tanmoy Mukherjee, Amit Seru, Vikrant Vig (2010), “Did Securitization
Lead to Lax Screening? Evidence from Subprime Loans,” Quarterly Journal of Economics, 125
Kleiner, Morris M. & Richard M. Todd (2009), “Mortgage Broker Regulations That Matter:
Analyzing Earnings, Employment, and Outcomes for Consumers,” NBER Chapters, in: Studies of
Labor Market Intermediation, pages 183-231 National Bureau of Economic Research, Inc.
Mian, Atif and Amir Suﬁ (2009), “The Consequence of Mortgage Credit Expansion: Evidence
from the U.S. Mortgage Default Crisis” Quarterly Journal of Economics, 124 (4): 1449-1496.
Munnell, Alicia H.; Geoﬀrey M. B. Tootell; Lynn E. Browne; James McEneaney (1996), Mort-
gage Lending in Boston: Interpreting HMDA Data, The American Economic Review, Vol. 86, No.
1. (Mar.): 25-53.
Pahl, Cynthia (2007), “A Compilation of State Mortgage Broker Laws and Regulations, 1996-
2006.” Federal Reserve Bank of Minneapolis Community Aﬀairs Report, No. 2007-2.
Purnanandam, Amiyatosh K. (forthcoming), “Originate-to-Distribute Model and the Sub-prime
Mortgage Crisis,” Review of Financial Studies.
Rajan, Raghuram G. (1994), “Why Bank Credit Policies Fluctuate: A Theory and Some Evi-
dence,” The Quarterly Journal of Economics, 109 (2): 399-441.
Shapiro, Carl (1986), “Investment, Moral Hazard, and Occupational Licensing,” The Review of
Economic Studies, 53 (5): 843-862.
Stein, Jeremy (2002), “Information Production and Capital Allocation: Decentralized versus
Hierarchical Firms,” Journal of Finance, 57: 1891—1921.
A. Data Issues
A1. Sample Construction
HMDA requires lenders to report the outcome of loan applications, including denied, approved (by
lenders) but not accepted (by borrowers), originated, or purchased. For loans processed by loan oﬃcers
of a lender, the reporting is straightforward. Loans arranged by independent loan brokers are recorded
by the party which provides the funding, and therefore are reported by the lenders, since brokers do not
provide funding. Loans originated by correspondent lenders, which fund the loans and often immediately
sell to the lenders which usually provides a line of credit for the correspondent lenders, are reported by the
correspondent lenders as originated and as purchased by the warehouse lenders in the HMDA dataset. The
last type of loan is therefore double-counted in the HMDA dataset. In our main analysis, we use observations
excluding records that show the loans to be purchased.
It is possible that some correspondent lenders do not meet the HMDA’s requirements to report, and
thus the last type of loans go unreported. In those cases, those loans are likely reported by the lenders that
purchase them since the buyers are usually large institutions. For example, in 2006 3.3 million records show
that the lender purchased the loans. Yet only 1.8 mil records show that they are originated and then sold to
the categories of purchasers which consist of commercial banks/savings, credit unions, and aﬃliates. These
numbers suggest that the loans made by correspondent lenders are under-reported. Therefore, to examine
the sensitivity of our results, in robustness checks we include all records showing loans as purchased, and
use as the dependent variable the percentage of loans originated or purchased.
A2. Authenticity of Data
Since lenders input and submit the data to FFIEC (Federal Financial Institution Examination Council),
one concern is the authenticity of the data. One particular suspect is the reported income of the applicant.
This authenticity of this information can be altered by actions of the borrowers, brokers, and lenders.
HMDA data in 2007, compared with previous years, started to include tract-level data from the Census. In
particular, the census-level income and minority percentage data are provided. We thus compare the median
of the reported applicant income from HMDA with the median census income from Census.
In 2007, there were around 66k census tracts. The median tract-level (median) family income was
53.75k. The median tract-level (median) applicant income was 69k. There appears to be some exaggerations
since family income is usually less than applicant (individual) income. However, it is possible that the buying
family, in order to be able to aﬀord the houses in a neighborhood, are slightly richer than the current residents
who may have purchased the house earlier at a lower price.
A3. Mortgage Applications that Are Submitted More Than Once (to Diﬀerent Lenders).
It is likely that applicants will seek alternative lenders after their mortgage loan applications are denied.
We think this data issue will not aﬀect our analysis as long as it does not systematically correlate with the
state licensing of mortgage brokers.
B. Variable deﬁnitions (from Pahl, 2007)
REG = Licensing/registration of entities, sole proprietors, and individuals acting as mortgage brokers
(Licensed/registered = 1; None=0)
LIC-EDU=Speciﬁc education requirement for licensing/registration (Required of many principals=2;
Required of one principal=1; None=0)
LIC-EXP=Speciﬁc experience requirement for licensing/registration (Required of many principals=2;
Required of one principal=1; None=0)
LIC-EXAM=Examination required to obtain license/registration (Required of many principals=2; Re-
quired of one principal=1; None=0)
LIC-CONT-EDU=Examination required to obtain license/registration (Required of many principals=2;
Required of one principal=1; None=0)
NET WORTH=Net worth requirement for licensing/registration (Net worth over $50,000=3; Net worth
$25,000—$50,000=2; Net worth under $25,000=1; None=0)
BOND=Bond requirement for licensing/registration (Bond over $50,000=3; Bond $25,000—$50,000=2;
Bond under $25,000=1; None=0)
MAN-LIC=Regulation of managing principals (Licensed/registered as individual mortgage broker=2;
Licensed/registered as employee=1; None=0)
MAN-EDU=Speciﬁc education required for managing principal status (Required=1; None=0)
MAN-EXP=Speciﬁc experience required for managing principal status (Required=1; None=0)
MAN-EXAM=Examination required for managing principal status (Required=1; None=0)
MAN-CONT-EDU=Continuing education for managing principal (Required=1; None=0)
BRANCH-BOND=Bond requirement for licensing/registration (Bond over $50,000=3; Bond $25,000—
$50,000=2; Bond under $25,000=1; None=0)
BRANCH-IN-STATE=In-state oﬃce required for licensing/registration (Yes =1; No=0)
BRANCH-MAN-LIC=Regulation of branch managers (Licensed/registered as individual mortgage bro-
ker=2; Licensed/registered as employee=1; None=0)
BRANCH-MAN-EDU=Speciﬁc education requirement for branch manager status (Required=1; None=0)
BRANCH-MAN-EXP=Speciﬁc experience requirement for branch manager status (Required=1; None=0)
BRANCH-MAN-EXAM=Examination required for branch manager status (Required =1; None=0)
BRANCH-MAN-CONT-EDU=Continuing education for branch manager (Required =1; None=0)
EMP-REG=Employees regulated (Employees regulated=1; Employees not regulated=0)
EMP-EXP=Speciﬁc experience requirement for licensing/registering employee (Required=1; None=0)
EMP-EDU=Speciﬁc education requirement for licensing/registering employee (Required=1; None=0)
EMP-EXAM=Examination required for licensing/registering employee (Required=1; None=0)
EMP-CONT-EDU=Continuing education for employee (Required=1; None=0)
SUMMARY CODE=Summation of all variables
C. Tract-level Change in Keeping Rate and Securitization Rate
HMDA records whether loans originated or purchased are kept or sold, and the entity of the purchaser.
To inquire whether the disproportionate increase in origination rates is due to the ease of securitization (and
the decreased incentives to screen), we examine whether high minority tracts experience disproportionate
increases in securitization rates. We ﬁnd that high minority-percentage tracts experienced lower keeping
rates, greater rate increase of loans sold to private securitizers, and greater rate decrease of loans sold to
government agencies. Note that government agencies follow relatively stricter issuance guidelines.
D. Further Analyses
We did a brief analysis of the eﬀect of federal preemption of state APL laws on lenders’ loan origination
standards. We ﬁnd that national banks increased their originations more dramatically starting in 2004 when
OCC succeeded in exempting national lenders from state APL laws. Meanwhile, the market share of state-
chartered banks shrunk, and so did the independent mortgage companies. We are investigating whether the
expanded market share of national banks was mainly attributed to activities of their subsidiaries.
E. Robustness Checks
Our main dependent variable is an indicator variable for whether the loan is originated. We plan to use
an alternative dependent variable — whether a loan application is denied. Second, it is possible that it takes
time for broker licensing to aﬀect the behavior and selection of mortgage broker; we thus plan to explore
using lagged instead of current licensing variables.
F. Findings on Loan Riskiness.
We examine whether the stricter licensing requirements decrease the riskiness of the originated loans.
We measure the riskiness of the quality using the ratio of loan volume divided by borrowers’ income. If our
variables capture the information used by lenders, the higher the ratio is, the less likely that the borrower
can aﬀord the payment for the loan, and the more risky the loan is. The results from state-level analysis are
in column 9 of Table 2B. yWe regress the state-level mean loan amount/income for all observations on the
same set of variables. We ﬁnd that the estimated coeﬃcient on the summary requirement variable is −.013
and signiﬁcant at the 10 percent level.
G. Regulation on Mortgage Banks
Another concern is that there exists diﬀerence in laws for consumer loan companies. Currently, under
state laws, mortgage banks need to be licensed before they can do business in a state. In WI, for instance,
the net worth requirement is the same as that for mortgage brokers, and the surety bond requirement for
mortgage bankers is $50k and that for brokers is $10k. In MN, mortgage banks and brokers are together
licensed as mortgage loan originators. Across the states, it appears that the variations in licensing on
mortgage banks are correlated with those on mortgage brokers.
More lenient requirements on mortgage banks invite more entry of mortgage companies, which
causes greater competition in the loan origination market. This may increase their incentives to win market
share by lowering loan origination standards. It also results in less charter value, therefore mortgage compa-
nies are more likely to take risks in issuing low-standard loans either by lowering loan-origination standards
or by originating riskier loans. These mechanisms might confound the eﬀect of the broker licensing on loan
To assess the existence of these mechanism, we pick two states that are similar to each other yet
diﬀer vastly in their regulation of brokers and mortgage companies: MN and WI. We ﬁrst check whether
the percentage of mortgage banks in all loan applications (loan originations) is smaller in WI, the state with
more stringent requirement on mortgage companies than MN. We ﬁnd that it is true that, in 2008, 2006 and
in 2004, the percent loan applications or loan originations by mortgage companies was higher in MN than in
WI. More investigations are being conducted to disentangle the eﬀect of licensing of mortgage broker from
that of mortgage companies. One way is to exploit the cross-state variation in mortgage banker licensing as
opposed to mortgage broker licensing.
Table 1A: Summary Statistics of HMDA Data and Loan Performance Data
# HP loan Pct Pct orig Pct kept Loan Total hp HP amt Amt Ratio Mean Mean
appli- orig or amount, loan amt secure- secure- secure delq 90+ delq 90+
cations pur- median ($ (billion tized by tized by -tized as of Dec as of the
(millions) chased thousands) $) PLI PLI 2008 end of
(billion (billion orig year
2000 8.2 .58 106 64 114 12.4 1.9
2001 7.6 .64 115 85 238 11.6 1.2
2002 7.3 .69 125 112 350 7.7 .70
2003 10.5 .75 .53 .21 173 542 5.0 .55
2004 12.7 .71 .51 .18 132 855 360 802 .42 10.6 .63
2005 15.2 .69 .49 .19 133 991 526 1100 .53 14.8 1.0
2006 14.7 .69 .458 .18 135 909 454 996 .50 18.9 1.4
2007 141 354 12.6 4.2
Notes. The sample includes loans for home purchase (HP) only. The sample includes observations where loans are purchased by
lenders. Percent originated or purchased is the percent of loan applications that are either originated or purchased by the lenders.
The percent kept is the percent of loan applications that are originated and kept (not sold) by the lenders. From 2000 to 2002
where data comes from Urban Institute HMDA, data is for home purchase of 1-4 family units. Data on loan performance are for
loans that are securitized by private label issuers (PLI). Delq 90+ refers to the percent of loan value whose borrowers are 90 days
or more behind in their mortgage payments.
Table 1B: Licensing Requirements for Mortgage Brokers, 1996-2006
licensing bond+ Exam Education
States requirements States Net-worth States requirement States requirement
Florida 14.5833 Tennessee 5.583333 Florida 3 Florida 3
New Jersey 13 New Jersey 5 California 2 California 2
California 10 Arkansas 4 New Jersey 2 Indiana 1.166667
Ohio 8.91667 Idaho 3.5 Ohio 1 Kentucky 1.166667
Nevada 8.75 Wisconsin 3.333333 Arizona 1 Alabama 1
S. Dakota 1.75 California 0 Minnesota 0 Alaska 0
Wyoming 1 Colorado 0 Colorado 0 Virginia 0
Minnesota 0.75 chusetts 0 Hawaii 0 Hawaii 0
Colorado 0 South
Dakota 0 Idaho 0 Minnesota 0
Alaska 0 Minnesota 0 S. Carolina 0 Tennessee 0
Notes. The top panel of this table lists the five states that score the highest for the relevant licensing requirements. The
variable “summary licensing requirements” is the sum of all licensing requirements: education, work experience, exam,
surety bond and net worth, and office in state. Bond+networth is the sum of the bond requirement and the networth
requirement. Bond requirement is coded as follows: Bond over $50,000=3; Bond $25,000-$50,000=2; Bond under
$25,000=1; None=0, where bond is the surety bond requirement for licensing/registration. Networth requirement is coded as
follows: Net worth over $50,000=3; Net worth $25,000-$50,000=2; Net worth under $25,000=1; None=0, where net worth is
the requirement for licensing/registration. Education requirement is the sum of specific education requirement for
licensing/registration (Required of many principals=2; Required of one principal=1; None=0), specific education required for
managing principal status (Required=1; None=0), specific education requirement for branch manager status (Required=1;
None=0), and specific education requirement for licensing/registering employee (Required=1; None=0), where specific
education requirement is, for example, hours of classroom education. The definition for the exam variable is the same as that
for education except that the requirement is passing of an exam.
Table 2A: Entry of Brokers as a Function of Licensing Regulation
Gr_employment for Gr_emp for all loan Gr_employment Gr_emp for all
loan officers only workers for loan officers loan workers
Bond+net worth -.0084* -.03
lagged (.0051) (.02)
Bond_all+net worth -.0082 -.014
lagged (.0050) (.010)
No obs 447 449 447 449
R-squared .11 .03 .11 .03
Notes. Employment of loan officers is from Occupation Employment Survey, 2002-2006. Gr_employment_loan_officers is the
change in the number of loan officers. All loan workers include loan counseling, loan interviewers and clerks, and loan officers.
Bond, net worth, education, and passing exam are defined in notes of Table 1B. Bond_all is the sum of bond requirement for the
licensee and the branches. Included as explanatory variables are also lagged education requirement variable, lagged work
experience requirement variable, and lagged exam requirement variable. All columns include year fixed effects. Standard errors,
in parentheses, are clustered at the state level.
Table 2B: Panel and Yearly Analysis of Loan Origination at State-level, 1997 and 2003-06
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Pct_ orig Pct_ orig Pct_ orig Pct_ orig Pct_ orig Pct_ orig Pct_ orig Pct_issue Loan
Summary -.0053*** .002 -.0045*** -.0053*** -.0058*** -.0056*** -.0041*** -.013*
lic. req. (.0013) (.004) (.0013) (.0015) (.0016) (.0016) (.0010) (.007)
lic req, lag (.0014)
Ratio kept -.08 -.007 -.09 -.18 -.10 -.06 -.00083 -.21** .007
(.10) (.005) (.19) (.12) (.10) (.11) (.0090) (.10) (.023)
House .11 .13* .21 .24* .09 .094 -.04 .09 2.48***
price gr (.08) (.08) (.51) (.14) (.14) (.10) (.03) (.07) (.43)
Loan_ .003 .012 .17*** .01 .008 -.008 .13 .006
income (.024) (025) (.05) (.03) (.03) (.03) (.18) (.019)
No. of obs. 204 204 51 51 51 51 51 204 204
Period 03-06 03-06 1997 2003 2004 2005 2006 2003-06 2003-06
R-squared .33 .324 .27 .32 .26 .23 .21 .35 .33
Notes. The data are for home purchase only unless otherwise noted. Data excluded loans that were purchased by lenders, unless
otherwise noted. The dependent variable is percent of loans that were originated at the state-level. Pct_issue is the percentage of
loans that were originated or purchased by lenders. Ratio kept is the percent of loans that were not sold by the lenders (in the
secondary market) among all originated loans at the state-level. The summary licensing requirement variable is defined in notes
of Table 1B. House_price gr is the percentage change of home price over the last year. The house price index is from FHA.
Loan_income is the state average of loan amount/applicants_ income. All columns include year dummies for 2003-2006. Note
that in column 2, the state-level loan amount/income variable is the mean of all observation on loan amount/income, not the mean
of loan amount/income for originated loans. Estimation method is OLS. In all columns, standard errors, in parentheses, are
clustered at state level. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 2C: Panel Analysis of Loan Origination, 1997 & 2003-06: Specific Licensing Requirements
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Pct_orig Pct_orig Pct_orig Pct_orig Pct_orig Pct_orig Pct_orig Pct_orig Pct_orig
Ratio kept -.0062 -.07 -.19 -.098 -.05 -.0005 -.04 -.003 -.006
(.0055) (.20) (.13) (.11) (.13) (.011) (.11) (.011) (.006)
Exam-passing -.0032 Y -.002 -.0009 -.002 -.011* -.0036 -.0099 --
requirement (.0053) (.008) (.007) .007 .006 .0052 .006*
Education -.021*** Y -.013 -.017* -.024*** -.025*** -.018** -.025*** --
requirement (.008) (.010) (.009) (.008) (.009) (.007) (.009)
Experience -.0032 -.02 -.005 -.0013 -.006 Y -.0034 .003 --
requirement (.0051) (.02) (.006) (.006) (.007) (.0056) (.008)
Continuing -.0097* -.05* -.010* -.017** -.01 -.0082 -.0077 -.005 --
education (.0042) (.03) (.0055) (.007) (.007) (.0074) (.0046) (.007)
Licensee req -- -- -- -- -- -- -- -- -.013**
Man principal -- -- -- -- -- -- -- -- -.0068*
Employee req -- -- -- -- -- -- -- -- -.0046
Bond+networth -.0088*** .0040 -.006 -.009** -.0092** -.0089** -.0079** -.0096*** -.0092
(.0034) (.012) (.004) (.004) (.0038) (.0038) (.0033) (.0036) (.013)
Office in state -.013 -.022* -.007**
(.011) (.012) (.003)
House price gr .16** .44 .22 .20 .14 .20 .073 .22 .11
(.08) (.59) (.15) (.13) (.11) (.19) (.071) (.18) (.08)
Loan_income -.004 .21*** .01 -.02 -.024 -.064* .017 -.07** .005
(.03) (.06) (.03) (.04) (.03) (.035) (.023) (.03) (.02)
No. of obs. 204 51 51 51 51 51 204 51 204
Years covered 2003-06 1997 2003 2004 2005 2006 2003-06 2006 2003-06
R-squared .39 .29 .36 .36 .31 .34 .29 .37 .39
Note. Data are for home purchase loans only. Data excludes loans that were purchased by lenders. The dependent variable is the percent of loans
that were originated at state-level. Year fixed effects are included where applicable. The variables ratio_kept, requirements on exam, education,
surety bonds and net-worth, house price gr, and loan_income are defined in the notes of Tables 1B. Experience is the sum of requirement of work
experience at the licensee-, managing principal-, branch managers-, and employee-level. Contedu is the sum of requirement of continuing
education at the licensee-, managing principal-, branch managers-, and employee-level. The variable “Office in state” is an indicator variable for
the requirement that a branch needs to have an office in the state for it to legally operate in the state. Licensee requirement is the sum of all
requirements, such as education, work experience, exam, etc. that are at the licensee level. Similarly, managing principal requirement is the sum
of all requirements at the managing principal level. And employee req is the sum of all at the employee level. Estimation method is OLS.
Standard errors, in parentheses, are clustered at state level. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 2D: Over-time Variation in Licensing Requirements
(1) (2) (3) (4)
Pct_ orig Pct_ orig Pct_ issue Pct_ issue
Summary .00053 .0008 --
lic. req. (.00094) (.0008)
Work experience req. .02*** .008**
Education req. .007* .014***
Exam req. -.013*** -.015***
Cont edu req. .002 .005
Net worth req. .0004 Y
Office in state req. -.02 Y
HPI growth over past yr .17*** .16*** .06* .05
(.04) (.04) (.03) (.03)
Ratio kept -.05 -.05 -.08 -.09
(.09) (.08) (.06) (.06)
Loan_ Income .06*** .06*** .05*** .06***
(.02) (.02) (.02) (.02)
R-squared .11 .10 .06
Within .71 .58 .62
Between .01 .02 .01
Notes. Data are for 2003-06. State fixed effects are included in all columns. The number of obs. is 204. Variables are
defined in Table 2C. Robust standard errors are in parenthesis. *: significant at 10% level. **: 5% level, ***:
Table 2E: Summary Statistics of Loan Characteristics During 2000-07
Sum of FICO LTV DTI Sub- Pct ARM Neg Inte- Balloon Pct sub-
all loans prime full amorti rest prime
doc zation only low doc
0 651 78.25 38.5 0.46 0.41 0.03 0.03 0.04 0.10
01 233 672 75.97 36.4 0.31 0.68 0.37 0.01 0.04 0.03 0.08
02 343 672 75.72 36.5 0.35 0.63 0.49 0.01 0.12 0.02 0.11
03 509 672 76.94 36.4 0.41 0.58 0.55 0.01 0.17 0.01 0.15
04 786 664 79.16 37.6 0.46 0.52 0.73 0.06 0.32 0.01 0.18
05 1090 668 78.92 38.5 0.45 0.44 0.73 0.13 0.38 0.04 0.20
06 994 666 78.86 39.5 0.44 0.37 0.68 0.16 0.34 0.15 0.20
07 353 690 77.24 38.5 0.23 0.35 0.57 0.16 0.46 0.09 0.09
Notes. The sum of all loan value in column is for loans that are securitized by private label securitizers (in billions of dollars.) LTV is the ratio of
loan value over the house value. FICO is the borrower’s credit score. DTI is the ratio of monthly payment divided by monthly income. Subprime
refers to borrowers who credit history is poor. Loan documentation has three main categories: full-, low-, or no- documentation. ARM refers to
adjustable rate mortgage (vs. fixed-rate ones). Negative amortization is a payment option where monthly payments can not cover principal and
interest payment such that the principal balance builds up. Interest only is a payment option where monthly payment covers interest only. Balloon
payment is an option where 30-year loans are amortized as if it is a 40 year loans such that there is a large payment due at the maturity of the
Table 2F: The Impact of Licensing Regulation on Loan Characteristics of Originated Loans:
State-Year Panel Analysis
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Loan and borrower characteristics Payment options Doc. Borrower credit worthiness
LTV FICO FICO DTI Not Not Not Full Pct Pct sub- Pct sub-
nega interest balloon doc prime prime prime
tive only pay- low doc low doc
Summary lic. -.044 .62*** .60*** -.07*** .0023** .0006 .0014* .0026* .0038** -.0022** -0026**
Req (.046) (.23) (.22) (.03) (.0010) (.0030) (.0007) (.0014) (.0016) (.001) (.0011)
HPI growth over -- -- 4 -- -- -- -- -- -- -- --
past yr (11)
HPI gr. over -- -- -- .15 -- -- -- -- -.08** -- --
min. past four yr (.52) (.04)
No obs 408 408 408 408 408 408 408 408 408 408 204
Range 00-07 00-07 00-07 00-07 00-07 00-07 00-07 00-07 00-07 00-07 03-06
R-sqrd .14 .28 .29 .55 .48 .61 .78 .53 .53 .41 .08
Within .61 .79 .79 .76 .71 .82 .85 .91 .88 .77 .63
Between .0007 .0003 .003 .002 .08 .02 .0009 .06 .11 .01 .01
Mean of dep var 77.5 671 671 37.75 .96 .84 .95 .62 .26 .14 .17
Notes. LTV is the ratio of loan value over the house value. FICO is the borrower’s credit score. DTI is the ratio of monthly payment divided by
monthly income. Negative amortization is a payment option where monthly payments can not cover principal and interest payment such that the
principal balance builds up. Interest only is a payment option where monthly payment covers interest only. Balloon payment is an option where
30-year loans are amortized as if it is a 40 year loans such that there is a large payment due at the maturity of the loans. Doc. level refers to three
main categories: full-, low-, or no-documentation. Prime refers to borrowers whose credit score qualify them for prime loans. Subprime refers to
borrowers who credit history is poor. HPI refers to house price index (FHFA). Year fixed effects included in all columns. Consistent standard
errors are in parenthesis. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 2G. The Impact of Licensing Regulation on Being ARM for Originated Loans: State-Year Panel Analysis
(1) (2) (3) (4) (5) (6) (7)
Pct of fixed Pct of fixed Pct of fixed Pct of fixed Pct of fixed Pct of fixed Pct of fixed
rate in loan rate in loan rate in loan rate in loan rate in loan rate in loan rate in loan
amount amount amount amount amount count count
Summary lic req .00092 .0044* .0039*** -- .0047*** .0014 .0043***
(.002) (.0024) (.0015) (.0018) (.0017) (.0015)
Work experience .008
Education req. -.004
Exam req. .02
Office in state -.02
HPI over min. price -.12*** -.06*** -.17 -.13*** -.19***
of last 4 yrs (.04) (.01) (.03) (.03) (.04)
State FE N N Y Y Y N Y
No obs 408 48 408 407 306 408 408
Range 00-07 2005 00-07 00-07 02-07 00-07 00-07
States All Exclude CA, All all all All All
R-sqrd .75 .07 .74 .74 .57 .64 .63
Within -- -- .92 .92 .85 -- .85
Between -- -- .04 .04 .00 -- .04
Notes. Pct of fixed rate in loan amount is the percent of loan amount that is fixed rate (vs. ARM). Pct of fixed rate in loan count is the percent of
loan counts that is fixed rate. Mean across 432 state-yr obs of pct fixed rate in loan counts is .53. Year fixed effects are included. Consistent
standard errors are in parenthesis. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 3A: Loan Origination. Summary Statistics on Tract-level Data, 2007
Population number Minority pct HUD median Tract HUD inc
N 65515 65488 65550 65351
Median 4043 19 59100 97
Notes. Minority pct is the percent of population that is not-white. HUD median income is the median income at the tract-level.
Tract_hud_inc_percentage is the tract-level median income as a percentage of the MSA median income.
Table 3B: Loan Origination: Tract-Level Analysis, 2007
Pct_issued Pct_issued Pct_issued Pct_issued
Ratio kept -.19 -.20 -.20 -.21
(.01) (.01) (.01) (.01)
Summary licensing -.0033
Exam requirement -.01 -.005 -.006
(.004) (.003) (.0025)
Education requirement -.026 -.02 -.018
(.004) (.003) (.0024)
Experience .007 .01 .011
requirement (.004) .(003) .(0035)
Population 3.8e-06 3.80e-06
Tract mino_pct -.002 -.0018
Tract HUD income pct .0005 .0005
Exam*mino pct .00005
Edu*mino pct -.0001
# obs 66317 -- 64245 64245
R-squared .09 .36 .36 .35
Standard error 3183 clusters 3132 clusters 3132 clusters 3132 clusters
Notes. Data are for home purchase loans in year 2007 only. Data includes loans that were purchased by lenders. The dependent
variable is the percent of loans that were originated or purchased by lenders at census-tract-level. The control variables are
defined in the notes of Tables 4B, 4A, and 1B. The variable “tract mino_pct” is the percent of minorities in a tract. Tract HUD
income pct is the tract-level median income as a percentage of MSA median income. The “Y” means that the variable is included
but the coefficients are not reported. Estimation method is OLS. Standard errors, in parentheses, are clustered at state-county
Table 3C: Changes in Issued Loan Amount at Tract Level, 2004-2005
Growth of issued Growth of issued Growth of issued Growth of issued
loan amt loan amt loan amt loan amt
Tract Minority pct .0043*** .0049*** .0055*** .0004***
(.0004) (.0004) (.0012) (.0001)
Tract MSA income pct -.00016
Summary licensing .01** .0004
Requirement (.005) (.0005)
Summary lic. req.* -.00004***
minority pct (.000015)
# obs 64924 64294 64294 64294
Notes. Data are for home purchase loans in year 2004-05. Data includes loans that were purchased by lenders. The dependent
variable is the change in total amount for loans that were originated or purchased by lenders from 2004 to 2005 at census-tract-
level. The control variables are defined in the notes of previous tables. The minority percent is from 2007 HMDA. Estimation
method is OLS and standard errors are robust in column 1. The specification in column 2 includes MSA FE. The adjusted R-
squared is .0031 and .0022 in columns 1 and 2. Robust standard errors are in parentheses. *: significant at 10% level. **: 5%
level, ***: 1% level.
Table 4A: Loan Origination: Summary Statistics at Lender-level
N Mean Median P1 P25 P75 P99
No. of lenders 7625 1052 71
Lender_loan_number 8244 1544 72 1 21 248 26636
Lender_number_state 8244 5 2 1 1 4 50
Lender_loan_number 8272 1843 69 1 21 261 29820
Lender_number_state 8272 5.3 2 1 1 4 50
Lender_loan_number 6561 1644 80 1 20 288 19751
Lender_number_state 6561 3.84 1 1 1 3 34
Notes. Lender_loan_number is the number of loan applications that a lender receives. Lender_number_state is the
number of states in which a lender originates or purchases loans. P1 represents percentile 1, and so on.
Table 4B: Loan Origination: The Effect of Mortgage Broker Licensing Requirements for
2004 2005 2005
Pct_issued Pct_issued Pct_issued
Ratio kept .049 .007 .067
(.011) (.009) (.10)
Exam requirement -.004 -.008 -.005
(.0015) (.001) (.0015)
Education requirement .0018 -.0018 .0015
(.0016) (.0015) (.0015)
Work Experience .0025 .0015 .00036
requirement (.0015) (.0015) (.0014)
Dum_subprime -.30 -.31 -.30
(.02) (.18) (.02)
Dum_subprime* -.015 -.0090
education requirement (.005) (.0059)
Dum_subprime* -.0016 -.0063
exam requirement (.0044) (.0048)
# obs 34697 36817 36817
R-squared .16 .50 .15
OLS method OLS Lender FE OLS
Notes. Data are for home purchase loans in year 2006. Data includes loans that were purchased by lenders. The dependent
variable is the percent of loans that were originated or purchased at the lender-level. The control variables are defined in the notes
of previous tables. Dum_subprime is an indicator variable for the lender being in the list of subprime lenders of the HUD list.
Standard errors, in parentheses, are clustered at lender level. The number of lenders is a bit above 8,000.
Table 4C: Lender-level Analysis of Loan Origination, 2006
(1) (2) (3) (4) (5)
Pct Pct Pct Pct originated; Exam Pct originated; exam
originated originated originated req. is for employee req. is for licensee
Summary licensing -.0025***
Exam requirement -.011*** -.0092 -.012*** -.0008
(.0014) (.0014) (.0026) (.004)
Education -.0091*** -.0075 -.009*** -.012***
requirement (.0016) (.0017) (.0016) (.002)
Experience .0010 .0042** .005*** .0031*
requirement (.0015) (.0017) (.002) (.0017)
Continuing -.0029* -.0025**
education req. (.0012) (.0011)
Bond+networth -.0015* -.00080 -.0006 -.0009
requirement (.00086) (.00085) (.0009) (.0009)
Office in state -.0064*** -.0059** -.0064**
requirement (.0025) (.0025) (.0026)
Rate kept -.027*** -.025*** -.017** -.017** -.017**
(.008) (.008) (.008) (.008) (.008)
Loan_amt/ income -.0013 -.0012 -.00097 -.0010 -.0010
(.0019) (.0010) (.00076) (.0008) (.0008)
HPI growth over .21*** .25*** .24 .21*** .21***
past year (.03) (.03) (.03) (.03) (.03)
No. of obs 34011 34011 34011 34011 34011
R-squared .0074 .0050 .0906 .0925 .0956
Notes. Data are for home purchase loans in year 2006. Data excludes loans that were purchased by lenders. The dependent
variable is the percent of loans that were originated at the lender-level. The exam requirement variable in column (4) refers to
exam requirement for employee while the exam requirement variable in column (5) refers to exam requirement for licensees. The
variables are defined in the notes of previous tables. In columns 3-5, the following additional control variables are included: Pct
no guarantee, pct owner occupy, pct male, pct Hispanic, pct single family, pct white, pct with first lien. “Pct no
guarantee” is the percent of loans that were not guaranteed by government agencies, like FHA, etc. “Pct owner occupy” is the
percent of loan applications for homes that were occupied by owners instead of for investment purposes. “Pct male” is the
percent of males among loan applicants. “Pct Hispanic” is the percent of Hispanics among applicants. “Pct single family” is the
percent of homes that are single family instead of manufactured homes among loan applicants at the lender-level. “Pct white” is
the percent of whites among applicants for the lender. “Pct with first lien” is the percent of first lien loans. In columns 4 and 5,
the exam requirement is superseded by an indicator variable for employee exam requirement and licensee requirement,
respectively. Estimation method is OLS that includes lender FE. The number of lender is close to 8200. Robust standard errors
are in parentheses. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 5A: Loan Performance as a Function of Loan Characteristics: Summary Statistics
Features Rate Payment options Documentation
ARM Fixed Negam Not IO Not IO Balloon No Full doc Low No doc
negam balloon doc
Mean of 90+ 17.3 7.5 15.6 10.0 9.5 12.3 16.8 11.2 11.2 13.2 11.4
Origination channel Borrower credit worthiness
Features Retail Broker Whole- No-data Prime Alt A Sub- Subprm Prime Subprm
sale on orig. prime low doc low doc full doc
Mean of 90+ 13.9 16.7 20.2 11.5 2.7 8.5 19.3 22.3 3.9 18.0
Notes. Each entry reports the average percent of loan value originated 2000-07 that is 90 or more days late in mortgage payment as of
Dec. 2008 for loans possessing the focal feature.
Table 5B: Loan Performance as a Function of Loan Characteristics, Cont’d
90+ delinquency rate 90+ delinquency rate 90+ delinquency rate
FICO -.14*** -.09** -- -.062**
(.03) (.04) (.028)
LTV -.06 -.13 -- -.04
(.14) (.27) (.11)
DTI -- -- .65** .47
HPI change from orig -- -- -18*** -15
yr to 2008 (2) (2)
Unemployment rate -- -- .51* .41
Spec OLS State FE OLS OLS
Obs 406 406 406 406
r-sqrd .53 .51 .64 .66
Notes. The dep. var. is the percent of loan balance that is 90+ days delinquent as of Dec. 2008 for securitized loans that were originated
in years 2000-07. Year fixed effects are included. Consistent standard errors are in parenthesis. *: significant at 10% level. **: 5% level,
***: 1% level.
Table 5C: Loan Performance and Licensing
Dep. Var.: 90+ All All All All All HP HP Refi HP low HP, Broker
Delinquency rate loans loans loans loans loans doc Full
Summary .23* -.17 -.30** -.46*** -.21 -.61*** -.31** -.48
licensing req. (.13) (.13) (.145) (.15) (.16) (.21) (.13) (.33)
Lag lic. req. -.34*** -.47***
Average of -.44***
current and lag (.16)
Yrs covered 00-07 00-07 03-06 03-06 03-06 03-06 03-06 03-06 03-06 03-06 03-06
State FE N Y Y Y Y Y Y Y Y Y Y
No of obs. 406 406 204 204 204 204 204 204 203 203 185
R-sqrd .65 .57 .68 .66 .63 .58 .58 .71 .53 .55 .33
Within r-sqrd -- .71 .85 .85 .85 .83 .83 Y .78 .83 .53
Between r-sqrd -- .28 .27 .22 .16 .13 .12 Y .11 .14 .02
Mean of dep var 11.7 11.7 12.3 12.3 12.3 12.6 12.6 11.7 13.7 11.9
Notes. The dep. var. is the percent of loan balance that is 90+ days delinquent as of Dec. 2008 for loans originated in years 2000-07 or otherwise
specified. The HPI change from the year of loan origination to 2008 and the state unemployment rate are included as control variables. Consistent
standard errors are in parenthesis. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 5D: Loan Performance at the End of the Year of Origination 2000-07, by Fixed vs ARM
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Dep var: 90+ Delinquency All loans All loans All loans All loans All loans ARM Fixed ARM ARM
rate loans rate loans loans loans
Summary licensing req. .01 -.052** -.04* -- -- -.08** -.026 -- --
(.02) (.023) (.02) (.03) (.019)
Dummy for ARM loans 1.2*** 1.2 .84 1.2*** 1.2 -- -- -- --
(.06) (.07) (.07) (.07) (.07)
Education requirement -.39** -- -.495* --
Experience requirement -.20* -- -.39** --
Exam requirement .14 -- .15 --
Total employee req. -- .-.13* -- -.22**
Licensee req. -- .15* -- .19
Bond+networth -.02 -.05 -.02 -.06
(.11) (.11) (.16) (.15)
Office in state .12 -.25 .23 -.27
(.35) (.24) (.56) (.37)
Yrs covered 00-07 00-07 03-06 00-07 00-07 00-07 00-07 00-07 00-07
State fixed effects N Y Y Y Y Y Y Y Y
No of obs 800 800 402 798 798 405 395 404 404
R-sqrd .44 .20 .43 .42 .48 .20 .47 .45
Within r-sqrd -- .53 .31 .53 .53 .62 .28 .62 .62
Between r-sqrd -- .02 .01 .005 .09 .02 .004 .0006 .11
Mean of dep var (percent) 1.45 1.45 .90 1.45 1.45 2.15 .74 2.15 2.15
Notes. The dep. var. is the percent of loan balance that is 90+ days delinquent as of the end of the year of the origination. Mean of dep var is 1.45
(# obs is 810). The median is .80. The “Y” means that the variable is included but the estimated coefficient is not reported. Consistent standard
errors are in parenthesis. *: significant at 10% level. **: 5% level, ***: 1% level.
Table 5E: Dec. 2008 Performance for Loans Originated in 2000-07, by Subprime or Not
Dep var: 90+ All loans All loans Subprime Subprime Subprime Subprime Subprime
Delinquency rate loans loans loans loans low doc
Summary licensing req. .20* -.14 -.21* -.40*** -- -- -.40**
(.12) (.18) (.13) (.15) (.22)
Dummy for subprime 13.8 13.9 -- -- -- -- --
loans (.49) (.48)
Education requirement -- -- -- -- .14 -- --
Experience requirement -- -- -- -- -.82 -- --
Exam requirement -- -- -- -- -2.1 -- --
Total employee req. -- -- -- -- -- 1.0 --
Licensee req. -- -- -- -- -- -.02 --
Bond+networth -- -- -- -- -.88 -.36 --
Office in state -- -- -- -- 1.6 -7.9** --
HPI change over past 4 -13 -13.7 -19 -41 -41 -42 -48
yrs (2) (2.3) (2.5) (5) (5) (5) (6.4)
Yrs covered 00-07 00-07 00-07 03-06 03-06 03-06 03-06
State fixed effects N Y Y Y Y Y Y
No of obs 1104 1104 405 204 204 204 204
R-sqrd .59 .57 .49 .63 .61 .56 .58
Within r-sqrd -- .62 .70 .84 .84 .83 .74
Between r-sqrd -- .03 .08 .29 .27 .24 .31
Dep var mean (percent) 10.7 10.7 19.3 21 21 21 24.2
Notes. The dep. var. is the percent of loan balance that is 90+ days delinquent as of the end of the year of the origination. Mean of dep var is 1.45
(# obs is 810). The median is .80. The state unemployment rate is included as a control variable. Consistent standard errors are in parenthesis. *:
significant at 10% level. **: 5% level, ***: 1% level.
Table 6: Effect of Anti-predatory-lending (APL) Laws on Loan Performance, 2000-07
Dep var: 90+ (1) (2) (3) (4) (5) (6)
APL in effect .81 -.21 .76 -- -- .67
(.96) (.89) (.76) (.97)
HPI change from yr -- -18 -14 -- -15 --
orig. to Dec. 2008 (1.9) (2.6) (2.6)
Summary licensing var -- .25 -.15 -.26** -.164 -.252*
(.12) (.14) (.13) (.134) (.131)
State FE N N Y Y Y Y
r-sqrd .41 .64 .58 .35 .58 .36
Within -- -- .71 .62 .71 .62
Between -- -- .31 .10 .30 .10
Notes. Year fixed effects are included in all columns. The unit of observation is state-year 2000-2007. The number
of obs is 406 for all columns. The dep var is the percent of loans that are 90 days + delinquency as of Dec. 2008 for
loans originated in years 2000 to 2007. APL in effect is from Ding et al. (2010). HPI change is the HPI in Dec 2008
over the HPI in the year of origin. Summary licensing variable is the licensing requirements in the year of loan
origin. Robust standard errors are in parenthesis. *: significant at 10% level. **: 5% level, ***: 1% level.
Figure 1: Summary of Licensing Requirements, 1996-2006
The graph plots the across state average of the summary licensing requirements (code), the requirements on
surety bond amount (bond_ttl), and the requirements on surety bond and net-worth (bond_ttl_networht).
For more details on the definition, refer to notes of Table 1B.
Figure 2: Grant of Licenses in California, 1999-2009
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012
The graph plots the number of mortgage loan broker licenses issued each year in California.