The Role of Market and Regulatory Discipline in Mortgage Lender

Document Sample
The Role of Market and Regulatory Discipline in Mortgage Lender Powered By Docstoc
					         The Role of Market and Regulatory Discipline in Mortgage Lender Failures:
                   Bank versus Non-bank Failures in the Subprime Crisis

                                     Zsuzsa R. Huszár * and Wei Yu †

                       Very preliminary: Please do not quote without permission
                                          First draft: March 2011
We provide a unique analysis of the failures of non-bank lenders in comparison with bank
lenders to examine the role of market and regulatory discipline in the subprime mortgage
crisis. We find that for non-bank lenders, the market discipline instead of promoting market
stability rather impair it. First, lenders with high concentration of refinance and minority
loans are more prone to fail despite that the information on these potentially aggressive
lending practices has been public for years prior to the crisis. Second, competition increases
risk taking and the probability of failures because lenders in competitive environment with
finite number of qualified borrowers likely increase lending to less qualified borrowers to
maximize revenues. More importantly, the similarities in bank and non-bank lenders’ failures
cast doubt on the importance of bank regulatory disciplines. Focusing on the bank lender
subsample, we find that regulatory oversight, such as the capital reserve requirements and
loan reserve ratios were generally inadequate. Since the higher loan reserves are associated
with higher probability of failures, banks likely took on risky loans for higher profits while
increased reserves which provided insufficient protection. Overall, we find that liquid asset
ratios had the greatest economic impact in reducing the probability of failures suggesting that
in addition to strengthening the existing regulatory requirements more emphasis should be
put on alternative risk management tools such as liquid asset ratios.

Keywords: non-bank failures, bank failures, mortgage market, subprime lenders, market
discipline, banking regulation.

    Zsuzsa R. Huszár is from the Finance Department at the National University of Singapore (NUS). Mochtar
    Riady Building, 15 Kent Ridge Drive, Singapore 119245. Huszár is also affiliated with the Risk Management
    Institute (RMI) and the Institute of Real Estate Studies (IRES) at NUS. Email: Tel: +(65)
    6516 8017, Fax: +(65) 6779 2083
    Wei Yu is from the Finance, Real Estate and Law Department at the College of Business Administration,
    California State Polytechnic University, Pomona, CA. 3801 West Temple Avenue, Pomona, California 91768.

    The authors gratefully acknowledge the comments of Sumit Agarwal, David C. Ling, and Alexander
    Ljungqvist, Wenlan Qian and Weina Zhang.
In Watters vs. Wachovia, the Court ruled that the Office of the Comptroller of the Currency
(“OCC”) is permitted exclusively to regulate mortgage companies that operate as subsidiaries
of federally-chartered banks. In recognizing the OCC’s exclusive regulatory authority, the
Court’s decision effectively strips state regulators of their ability to enforce existing consumer
protection laws and hold bank mortgage lenders to the same standards for licensing, education,
and criminal background checks as mortgage brokers and other state-chartered entities.
                                          US Supreme Court: Vol. 550. Docket No. 05-1342.

1.        Introduction

          Banking crisis, clustered bank failures, periodically occur all over the world where a

handful of financial institutions emerge as phoenix from the ashes and benefit by overtaking the

weaker ones. In the 1980s, during the Saving and Loan Association (S&L) crisis, many heavily

regulated financial institutions failed as they lost competitive advantage to commercial banks

while commercial banks gained the market shares of thrifts. Despite many differences between

the S&L and the recent crises, two important similarities are the overexposure to real estate and

the near collapse of an industry segment. In the 1980s, thrifts almost disappeared, while in the

recent crisis, non-bank subprime mortgage lenders and mortgage brokers seem to wane.

          Among others, Mian and Sufi (2009) and Mian, Sufi, and Trebbi (2010) examine the

causes of the recent financial crisis, such as disintermediation, securitization, and the cooling of

the real estate market. They provide important insights about the geographic distribution of the

crisis by examining the geographic variation in the real estate price run-up and credit expansion,

but provide limited empirical insight about the cross-sectional differences in performance among

the different types of lenders. Some banks, such as Bank of America and Wells Fargo fared well

enough to benefit via “successful” acquisitions, while others have been acquired or are still in the

recovery process. 1 Similar situation can be observed in Australia, where the four big commercial

    Belratti and Stulz (2010) examine the cross-sectional variation in bank performance to reveal why some bank
    performed well in the crisis.

banks significantly increased their market shares as risk-averse clients moved back from non-

bank lenders to the safety of the Big Four despite higher interest rates.

           Another strand of studies examine bank failures with the objective to provide guidelines

for establishing better early warning systems about bank failures (especially clustered events)

and aid regulatory supervision. Cole and Wu (2010) propose a simple dynamic hazard model

with time-varying covariates as a bank failure early warning model, by showing that the model

significantly outperforms regulatory probit models with and without the macroeconomic

variables. Boyd, De Nicoló, and Jalal (2010), testing Boyd, De Nicoló, and Jalal’s (2009)

banking model, find that with increased competition the probability of bank failures decline, but

warn regulators not to over-emphasize competition as a self-regulatory tool because competition

tends to increase loan-to-asset ratios.

           As the media and academics focus on the commercial bank failures, the failures of

hundreds (if not thousands) of smaller non-bank lenders went relatively un-noticed after the big

failures Ameriquest Mortgage Co, and New Century Financial Corp. 2 But, the role of non-bank

lenders cannot be ignored as they provided a major portion of the subprime, now troubled,

loans. 3 These non-bank lenders are especially important because with the use of mortgage

brokers they heavily promoted disintermediation and potentially contributed to the loosening of

credit standards. Also, these lenders, without deposit base as a source of founding, heavily relied

on securitizations by non Government Sponsored Enterprises (GSEs) and involved other

financial institutions (e.g., investment banks) in their lending business. According to Iyer and

    In general, we refer to non-bank lenders as non-depository institutions, financial institutions that are not under the
    regulatory oversight of FDIC.
    These two subprime lenders originated more than 160 billion in loans in a couple of year prior to 2008. The two
    other important subprime lenders, Countrywide and Indymac can be considered banks as they operate banking
    units and engage in deposit taking with FDIC guarantee.

Peydró (2010) these links between the affected financial institutions were likely crucial in the

spread of the crisis.

        In this study, we aim to contribute to the literature on bank failures and provide the first

comprehensive analysis of non-bank mortgage lenders failures in comparison with bank failures.

Specifically, we examine the role of market and regulatory discipline in conjunction with lending

practices in explaining the recent large number of lender failures. Generally, in the unavailability

of financial data on non-bank lenders hinders a comprehensive analysis of non-bank mortgage

lenders. We address this issue, using lending characteristics (i.e., geographical concentration,

pricing, and securitization) and loan issuance (market size measure) information from the Home

Mortgage Disclosure Act (HMDA) raw Loan Application Register (LAR) file.

        It is important to note that the recent financial crisis is term of time frame is quite

different than the S&L crisis, where institutions failed over a course of several years. In our

study, since the high concentration of lender failures around 2009 does not support any duration

or risk hazard model, we use standard logistics model in predicting lender failures. The

governmental intervention, via bad loan purchase and interest rate reductions, also contaminate

the data, making it uninformative to distinguish between two failures: the one in June 2009 from

the one March 2010. Thus, we treat all failures in the recent crisis as one event after 2007, and

use public information (on lending practices and from financial statements) from 2003 to 2007 in

the models.

        Our findings suggest that lending practices play a key role in lender failure. Interestingly,

despite the significant differences in institutional details and regulatory requirements, bank and

non-bank lenders are driven by the same factors. For both types of lenders, higher concentration

of high credit quality borrower and low average loan amount are associated with significantly

lower probability of failure. In case of non-bank lenders, competition and transparency are

expected to aid the self-regulation of the new industry but proved to be harmful instead. The

higher concentration of competing lenders in the area is associated with higher probability of

failure. Possibly some lenders relax their lending standards to attract enough borrowers for their

business survival in strong competitive environment where the limited number of qualified

borrowers do not provide enough revenues for all lenders. Transparency on lending practices is

found to be under-utilized, as the evidence of aggressive lending to minorities already surfaced

before the crises, which could have been used to identify problem institutions well in advance.

       Concerning bank lenders, we do not find evidence in support of the beneficial roles of

regulations. Regulation, instead of being useful, tends to be rather harmful as the higher

concentration of minority borrowers and high loan-to-income loans significantly increases the

probability of failure. The high concentration of minority loans may have been the result of

banks’ attempt to support under-represented borrowers and areas according to the 1977

Community Reinvestment Act (CRA) and the Clinton and the G.W. Bush Administrations’

National Homeownership Strategy. Bank regulatory requirements, such as capital and loan

reserves, are generally uninformative in explaining the failures. Interestingly, the probability of

bank failures increased with the loan reserve ratios suggesting that banks knowingly accumulated

risk but the regulatory cushion was insufficient.

       Overall, we show that at the institution level, the failures of bank and non-bank lenders in

the recent financial crisis can be explained by similar factors, the low concentration of good

credit quality borrowers, and the high concentration of minority borrowers and refinance loans.

Neither the open market competition for non-bank lenders, nor bank regulations for bank lenders

were found to facilitate survival. Moreover, the risk management tools at the bank level rather

encouraged risk taking because banks were more likely to issue large quantity of risky loans

while accumulating loan reserves on the side, perceiving the risk to be insured. Our finding that

the liquid asset ratio is the most efficient tool in lowering risk of insolvency, suggests that

regulators may need to review existing regulatory risk management tools and focus on new

measures instead of strengthening the older ones.

2.        Background

2. 1      Recent Development of the US Subprime Mortgage Market

          Since the mid 1990s, subprime lending became a rapidly growing segment of the

mortgage market because it provided opportunities for homeownership to those who previously

could not qualify due to discrimination or due to a record of poor credit history. However, this

expanded access to mortgage loans came with a higher price (Fortowsky and LaCour-Little,

2002; Brent, 2007), thus making subprime lending also known as high-cost lending (Pennington-

Cross 2006). 4 Today, many argue that regulatory changes resulting in loosening of regulations

played an important role in the development and the collapse of the market (Mian and Sufi,

2008). For example, the 1982 Alternative Mortgage Transaction Parity Act (AMTPA) allowed

the introduction of adjustable-rate mortgages, balloon payments, and interest-only mortgages.

The 1986 Tax Reform Act (TRA) allowed interest deductions on mortgages and thus made

mortgage debt cheaper to consumers and also encouraged homeowners to do cash-out

refinancing by renewing their loans (Pennington-Cross 2006). Lastly, the 1998 Homeowners

Protection Act (HPA) required the homebuyers with loans higher than 80% loan-to-value ratio to

pay for extra insurance to reduce the risk of the loan.

    For detailed review on the evaluation of the US subprime mortgage market read Courchane, Surette, and Zorn,
    (2004) and Chomsisengphet and Pennington-Cross (2006).

           In the new millennium, the booming economy and the increasing real estate prices

created increasing demand that the lenders were more than ready to meet, especially as the

relaxed regulations allowed them to charge higher rates and fees at lower cost with

disintermediation, with reliance on mortgage brokerages. In addition, the lenders seemingly

passed off most of the risk by selling off loans in the secondary market and continued to leverage

their positions and “pump” credit into the market. Although in hindsight it is hard to argue in

support of the subprime mortgage market, it did provide more than 5 million home purchase

subprime loans, including more than 1 million loans first time home purchases (mostly to

previously under represented borrowers). But total of about 10 million new subprime loans

included loans for refinancing, investments and speculation that are difficult to view in positive

lights (Jaffe, 2008).

2.2        Mortgage Market Participants and Regulations

           Recently, the retail mortgage market became increasingly heterogeneous. New financial

institutions, such as mortgage companies and finance companies entered the market, besides the

traditional bank lenders such as commercial banks, saving and loan associations, and credit

unions. Banks are required to fulfill Basel II requirements, and annually evaluated based on

CAMELS criteria, as well as CRA guidelines. 5 The non-bank lenders, finance companies,

mortgage companies, and merchant banks, subsidiaries of industrial firms and/or foreign banks,

are less regulated. In addition, these lenders are also relatively opaque because their financial

    According to CRA, commercial banks and savings associations are expected to meet the needs of borrowers in all
    segments of their communities, including low- and moderate-income neighborhoods. The objective of CRA is to
    reduce discriminatory lending practices against low-income neighborhoods, known as redlining.

information is not readily available in the market, making it difficult for lenders and investors to

understand the risk these financial institutions were taking. 6

        Despite the economic importance of these non-bank lenders in the US market, there is

limited research on the lending practices of these financial institutions. One notable exception is

Ambrose’s (2003) study, which compares bank and non-bank lenders in the commercial

mortgage market and shows theoretically and empirically that bank and non-bank lenders use

different lending strategies. While banks use interest rates in allocating credit, non-banks adjust

credit allocation with the use of the loan-to-value ratio. Therefore, the borrower clientele is likely

to be different for those of banks and non-banks, because clients with a need for large loans may

incur the higher rates with a bank loan. These findings are consistent with that of the US, UK and

Australian residential mortgage market prior to 2007, where borrowers in search of lower rates

(at least lower initial rates) chose non-bank lenders.

        Bank lenders are regulated at the state or the federal level and had to follow international

rules while the non-bank lenders were mostly self-regulated. In a perfectly competitive and

transparent market environment, competition would eliminate the bad lenders that aggressively

target protected borrowers and engage in unethical business practices. 7 Despite high degree of

transparency ensured by the equal lending opportunities act and HMDA, the information on

lending practices was neither available timely nor well organized. Thus, market discipline via

investors monitoring and competition was ineffective. Until today, the lending practices of the

different subgroups of lenders, such as investment banks, mortgage companies, finance

companies, thrifts and banks are relatively unexplored. The likely reason is the limited financial

    In the aftermath of the crisis, the obtaining reliable financial information on the non-bank, mostly subprime
  lenders, is a challenge.
   Protected Borrowers are minorities and/or low-income borrowers who are classified as protected borrowers under
   the Fair Housing Act (Federal Reserve, 2006) and were found to be especially targeted by aggressive lenders,
   because of the generally lower level of financial education (Hill and Kozup, 2007)

information at the institution level for the non-public lenders, at least in the US market, hinders

comprehensive empirical studies.

2.3     Bank failures

        Despite the relative infrequency of bank crises around a world, a significant research has

focused on the topic because of the macro economical importance of this sector. Especially, the

spread of bank failures, so called contagion effect, concerns regulators. Initially two competing

hypotheses emerged: pure panic and information-based contagion. Aharony and Swary (1996) is

study provides additional evidence consistent with the latter hypothesis. More recently, Iyer and

Peydró (2010) further explore the informational hypothesis and reveal that interbank linkages

with failed bank are important source of contagion especially involving weak banks. They

suggest that regulators pay more attention to interbank links in the future to reduce the contagion

effect, and to evade similar situation as in the recent crisis.

        Interestingly, the role of market discipline, as monitory tool has attracted much attention

at the end of 1980s, when the FDIC proposed the use of subordinated notes and debentures

(SND) as a way of increase market discipline and augment regulatory discipline on banks and

banking organizations. Avery, Belton and Goldberg (1988) find that the potential for market

discipline is weak at best, as the pricing signals that bankers receive from the public subordinated

debt market appear to be at odds with the directions desired by regulators. While Hannan and

Hanweck (1988) find that the market prices the risk as the CD rates are higher for banks with

more variable income.

        Another stand of studies focuses on establishing early warning system for regulators.

Whalen (1991) propose a proportional hazard model of bank failures using bank data including

170 failures from 1985 to 1990. Using typical financial ratios and local economic condition

measures his model identifies both failed and non-failed banks with high accuracy. Cole and

Gunther (1995) use a split-population survival-time model to distinguish factors that result in the

ultimate failure from those that influence the survival time of failing banks. A few measures of

bank’s financial condition, such as capital reserve, troubled assets, and net income, are important

in explaining the timing of bank failure, while other variables in bank failure models, such as

measures of bank liquidity, are not associated with the time to failure. More recently, Cole and

Wu (2010) introduce a dynamic hazard model with time-varying covariates as an early warning

model for bank failures, which significantly outperform simple probit models (often used by

regulators) with and without the macroeconomic variables.

2.4    The Fall of Retail Mortgage Lenders: Regulation, Transparency, and Competition

       While there is significant heterogeneity across national bank crises, the main causes are

high credit growth, a negative GDP growth and a high real interest rate. In the recent financial crisis,

the triggers were the high credit growth and the increasing interest rates in conjunction with the

cooling of the real estate market. In 2006-2007 the “surprising” cooling of the residential house

market and the increase in interest rates made refinancing more difficult, while the negative

equity loans encouraged even non-financially constrained borrows to strategically default on

their loans. For the over-leveraged mortgage companies that heavily relied on external funding

(e.g., Asset backed commercial paper market), the soaring mortgage delinquencies quickly

resulted in liquidity problems.

       As the quality of mortgage backed securities declined, the asset backed commercial paper

market dried up and making the issuance of new securitized products difficult thereby cutting

funding for non-bank lenders. The financing difficulty and the short term liquidity problems

were crucial for many lenders that could not cope with these difficulties and failed. While the

Fed and the Treasury with the support of the Administration made attempts to stabilize the

financial market after the critical 2008 failures (e.g., Lehman Brother and Bearn Stern), there

was too much focus on the surface liquidity. Regulators have not realized that banks will be less

willing to lend to each other and the 2008 and 2009 stimulus bills were unsuccessful in

tempering the general economic downturn.

3.     Hypotheses

       In the recent financial crisis, a system wide shock affected countless financial institutions,

commercial banks, non-bank lenders. Insurance companies, including investment banks were

also affected because of their indirect involvement in the subprime mortgage market via

securitized investment products. For banks, capital reserve ratios and CAMELS ratings should

have provided warning signs about the overexposure to real estate market risk, and the potential

risk for bank failure, while for non-banks, the open market competition was expected to resolve

any inefficiency with the help of transparency.

       This study aims to contribute to existing bank failure literature by examining the factors

that contributed to the failure of lenders (bank and non-banks) during and after the subprime

crisis. More specifically, we categorize our lenders into two groups: (1) Bank and (2) Non-bank

lenders, where banks are primarily defined as depository institutions. The comprehensive

analysis of the two different types of lenders in the recent mortgage crisis allows us to answer

various important research questions.

       First, we focus on non-bank lenders, where in the absence of financial information we

can examine only the role of aggressive lending, competition and transparency in failures with

lending information. We examine the relationship between aggressive lending, bank

concentration and probability of failure, to shed light on the efficacy of self regulation, with the

following two hypotheses

        H1A: The probability of non-bank failure is higher in competitive environment

        H1B: The probability of non-bank failure is higher with higher concentration of minority

        and refinance loans.

        If self-regulation is effective, banks in more competitive environment are less likely to

fail. Earlier studies find that subprime lending concentrates in low income and minority

neighborhoods (Calem, Gillen, Wachter, 2004) where the potentially high cost loans are not

necessarily aiding the revitalization of the neighborhoods. A more recent study by Williams and

Bond (2007) shows that new subprime loans from traditional lenders had some positive effect in

decreasing segregation, but the loans from new subprime lenders, often specializing in minority

and low income borrowers rather increased segregation. Thus, the subprime lenders did not seem

to achieve the social benefit that regulators hoped they would. Since the information is public on

minority and low income borrower concentration, in an efficient market, the past public

information should be reflected in the value of the institution (especially for public firms) which

is unlikely to predict failures.

        Second, we examine the role of regulations by comparing the failures of banks and non-

banks. If regulations to some extent were efficient then the role of aggressive lending in bank

failures should be less pronounced.

        H2: The relationship between failure and aggressive lending proxies, such as the

        concentration in minority and low income borrowers is economically less important for

        banks because of banking regulations (besides the market discipline).

But, even if we find that the banks are also more likely to collapse when they have high

concentration of minority and low income borrowers, we still cannot conclude that these

institutions are actively engaged in aggressive lending. Unlike in the case of non-banks, banks

may have high concentration of the underrepresented risky borrower to encourage first time

home ownership in line with the Administration’s initiative.

          Lastly, focusing on banks, the authorities, through the regulations on bank activities,

capital adequacy, annual supervision (Barth 2004), should have noted that banks excessively

increased their real estate risk. Thus, we examine the effectiveness by testing whether low capital

reserve or loan reserves were signaling future insolvency. Lastly, we examine whether liquidity

ratios were effective warning signals as some countries, in addition to Basel II requirements also

enforce higher liquid asset ratios to alleviate the risk of insolvency as a result of liquidity

problems which was critical in the recent crisis. 8

          H3A: Capital reserve ratios and loan reserve ratios are insignificant in predicting

          lender’s failure

          H3B: High liquid asset ratios can significantly reduce the probability of bank failures (as

          these ratios are in place in countries that have been affected by the 1997 Asian Financial


4.        Data

4.1       Identification of Failed Mortgage lenders

    For example, the banking regulation in Singapore requires the maintenance of up to 18% liquid asset ratio and a
    minimum of 3% cash balance.

        We collect information on all failed residential lenders from Jan 1, 2007 till 31 Oct 2010,

including banks and non-banks. Failed banks are primarily identified from the Federal Deposit

Insurance Corporation (FDIC) website under the section of ‘Failed Bank List’ for the

corresponding sample period of our research. 9 Information on failed non-banks lenders, such as

mortgage firms, or subsidiaries of investment companies is very restricted as these financial

institutions are not linked to FDIC, or OCC. 10 With web query and the use of ‘The Mortgage

Lender Implode-O-Meter’ website, we identified about 300 non-bank institutions with location. 11

        Our research focuses on studying mortgage lending practices during the subprime crisis,

thus it is crucial that the failed financial institutions in our sample are actively involved in the

mortgage lending business. To identify the respective mortgage lenders, we decided to work with

the Home Mortgage Disclosure Act (HMDA) Lender File. Most lending institutions with offices

in metropolitan areas are required by the Home Mortgage Disclosure Act of 1975 to disclose

information with regards to their lending activities. Annually, the HMDA data contains over

8,000 lenders, accounting for at least 80% of the loans extended in the U.S. (Avery 2005).

        We match the hand collected default sample of 731 failed lenders (banks and non-bank

lenders) with the HMDA dataset by (1) Lender name, (2) City and (3) State, which results in a

match of 517 lenders with unique IDs. We include lenders that are among the 8,886 unique

lenders that reported to HMDA in 2006 before the onset of the crisis. We use manual name

matching and ensure that the failed institution was the same as that in HMDA by checking the

location variables. The allocation of a unique HMDA ID code also allows us to account for

overlaps that might occur, because of the double counting subsidiaries and the parent company

 The Federal Deposit Insurance Corporation, FDIC. Nov 5, 2010).
   Interesting that one of the most organized sources of information about non-bank financial is available via Home
   Mortgage Disclosure Act’s LAR database.
   The Mortgage Lender Implode-O-Meter. (accessed Nov 5, 2010).

(bank). In addition, we remove all foreign banks, or subsidiaries of foreign banks, but keep all

subsidiaries or American banks that did not fail, classifying these failures differently in the

empirical analysis.

           The HMDA agency codes allow us to identify the regulatory oversights for the

institutions. All financial institutions which are regulated by the (1) Office of the Comptroller of

the Currency (OCC), (2) the Federal Reserve Board (FRB), (3) Federal Deposit Insurance

Corporation (FDIC) or the (4) Office of Thrift Supervision (OTS) are identified as banks. All

others which are not regulated by the aforementioned four authorities are identified as non-

banks. Overall we identify 517 failed lenders with valid lending information, including 344 bank

and 253 non-bank lenders. In addition, using the HMDA performance data, we also identify

2,240 bank lenders with valid HMDA and banking information and 3,215 non-banks with valid

HMDAID and lending information. Lastly, for banks (failed and non-failed bank lenders), we

obtain financial statement data from Bankscope to control the bank health and profitability.

4.2        Description of Variables

           Using HMDA annual reports for each lender, we calculate a series of variables that

capture their exposure to the lending market to specific states and some variables that reflect

potentially aggressive lending behavior. We create a number of variables to capture lending

practices, such as total loan application, total loan origination, annual approval rates, and

concentration of refinance and repurchase loans. We attempt to measure the degree of

responsible lending practices with geographical loan concentration, minority loan concentration,

and the concentration of high-cost, high-rate, and HOEPA loans. 12 We also calculate the change

      HOEPA loans are the loans that are required to be disclosed as high cost loans according to the Home Ownership
     and Equity Protection Act Amendments (HOEPA). In general, first (and second) lien loans with rates 8 (10)
     percentage point higher than that of comparable treasures are required to be disclosed.

in these variables over the period of 2003 to 2007, to capture whether lenders change their

lending practices in terms of becoming more (less) diversified or more (less) exposed to

minorities. Full list of lending practices variables, applicable to all bank and non-bank lenders, is

included in Panel A of Table 1.

           In addition, we obtain financial information such as balance sheet and cash flow

statement for all banks through Bankscope. 13 Specifically, we collect information about the

bank’s asset, equity, liquid asset, off-balance-sheet items (OBS), net income, total deposits and

total loans. To assess banks financial stability, we obtain information on total capital and tier one

capital reserve ratio, total loan reserve, and loan to equity ratio, loan to earning asset ratio, and

total problem loans to liquid asset ratio. In addition to the annual averages, we also calculate

trends, that is, the change in deposit base, liquid assets among others. Change in the bank

financial status could provide important warning signal for future funding difficulty, or liquidity

problem, and for the ultimate failure. The complete list of Bankscope variables are described in

Panel B of Table 1.

                                            [Table 1 about here]

5.        Empirical Analysis

5.1       Summary Statistics

          Panels A and B of Table 2 display the summary statistics for failed 249 bank and 329

non-bank lenders, respectively. The summary statistics suggest that failed bank and non-bank

residential mortgage lenders are comparable in their lending activities. Since both types of

lenders issue more refinance than purchase loans, the majority of loans was not to promote first

time homeownership as many regulators argued.

     Bankscope is accessed via Wharton Research Data Service (WRDS).

       About 22-24% of the borrowers from areas with high FICO scores while the average loan

amounts is about 2.2-2.8 times of the annual incomes. The somewhat higher loan-to-income

ratios for bank lenders suggest that banks are more inclined to issue loans to lower income areas,

or to lower income borrowers. One notable exception is the bank lenders’ relatively heavy

exposure to minority areas (51 versus 40%), potentially in response to governmental initiatives to

promote minorities housing. More interestingly, while the average number of applications is

significantly more (35 thousand versus 27 thousands) for banks than non-banks, the issued

amounts are relatively comparable (63 versus 53 million). Because non-banks lenders not only

have higher origination rates, 65% versus 50%, but also issue larger loans.

                                        [Table 2 about here]

       The time trends are also similar and consistent with the media reports following the

crisis. Both types of institutions steadily increased loan originations and focused more on

minority neighborhood. More importantly, the percentage borrowers from high credit quality

areas declined. Interestingly, the approval rates have also declined over time, likely as the result

of generally lower credit quality borrowers. For the still active bank and non-bank lenders, the

trends are quite different in Table 3. For these lenders, there is no significant increase in minority

borrowers or decline in the credit quality.

                                        [Table 3 about here]

       The lenders successfully surviving the crisis were generally less exposed to minorities

and lower income borrowers, reflected by the lower Pctminorty and Loan-to-income variables in

Panels A and B of Table 3. More importantly, these institutions are much less exposed to the

residential mortgage market reflected by the significantly lower application numbers and the

total issuance amount (Amount in Mill). For these institutions, the high origination rates are

unlikely to reflect aggressive lending or lower lending standards because of the high

concentration of loans in high credit quality areas (captured by high PctgoodFICO).

                                       [Table 4 about here]

       While for non-bank lenders we are unable to collect reliable financial information, for

bank lenders we are able to compare important bank risk management measures across failed and

non-failed bank lenders. Consistent with Beltratti and Stulz (2010), we show that the banks that

successfully survived were significantly less profitable from 2003 to 2007 in terms of net income

(NI) and net profits (Profit). But these profits were declining with the onset of the mortgage

crises and the declines in real estate prices. The over-exposure of these banks to the mortgage

market was quite clear, as the total problem loans relative to capital reserves (Tprobloans_cap)

and liquid asset (Tprobloans_liqAt) are about .23 and .47 for the failed institutions already before

2007. For the non-failed banks, the Tprobloans_cap and Tprobloans_liqAt measures are

negligible at about .03 and .19.

5.2    The Role of Lending Practices in Bank and Non-bank Lender Failures

       At first glance, the logistic regression analyses for predicting the failure of mortgage

lenders does not reveal anything surprising. The probability of failure is negatively related with

higher concentration for borrowers from good FICO score areas and lower loan-to-income ratios.

The higher percentage of subprime highrate loans and refinance loans tend to increase the

probability of failure. For banks, the higher concentration of borrowers with minority areas is

detrimental, as a 1% increase in Pctminority increases the probability of failure by 1.8%. Banks

were strongly encouraged to promote minority homeownership; and thus, regulators and political

forces were likely partly responsible for the higher minority lending and the subsequent negative


                                               [Table 5 about here]

           Since overall the coefficient estimates and the reported odds ratios are similar

(economically and statistically not significantly different), the question arises whether banking

regulations were at all significant. 14 We also examine one aspect of self regulation, competition,

and find that lenders in competitive environment are more likely to fail. The 1.214 odds ratio on

the HighComp variable for non-bank lenders suggest that lenders that on average operate in areas

where the completion was stronger than the median level of completion are about 20% more

likely to fail. Intuitively, in the presence of inelastic supply of good credit quality borrowers, the

stronger competition restricts some lenders to attract economically profitable number of new

borrowers with good credit quality. Thus, as lenders are more likely to relax lending standards

and increase risk taking in high competitive markets, the foundation of open economy and the

benefits of competitions are at question.

           In Table 6, in addition to the past average lending practices, we also consider the role of

time trends. Specifically, we examine whether increased loan amount, loan origination and

approval rates that reflect relaxation in lending standards predict lenders failure. Although odds

ratios on the change in loan amount (Chngloanamount) and loan origination (Chngloanorig) are

positive and statistically significant, the economic significance is negligible in comparison with

the effect of the loan-to-income ratio or the concentration of good credit quality borrowers

(PctgoodFICO). But, the “abnormally” large positive odds ratio of the change in approval rates

variable (Chngpctapproval) clearly shows that most failures are concentrated among institutions

that significantly increased the approval rates from 2003 to 2007.

                                               [Table 6 about here]

     Beltratti and Stulz (2010) raise a similar question in examining the role of regulations across countries in
     conjunction with bank performance. They find that on average bank regulatory indices are uncorrelated with bank
     risk measures thereby casting serious doubt on the efficiency of banking regulations.

5.3       The Role of Financial Performance and Bank Management in Bank Failures

          The comparable magnitude of bank and non-bank residential mortgage lenders’ failures

(about 300 in absolute and 10% in relative terms) suggest that regulations were ineffective. If

bank regulations were effective in reducing excessive risk taking or ensuring suitable capital

reserve, then bank failures should be less prevalent than non-bank failures. In addition, the

results in Tables 5 and 6 also suggest that similar lending practices were the driving forces of

bank and non-bank failures. But, based on these findings we cannot assertively conclude that

regulations were ineffective because we may have overlooked a number of non-bank failures.

To address the selection problem, we examine the role of regulations in a subsample of bank

lenders by focusing on specific regulatory requirements such as loan reserves and capital reserve


                                         [Table 7 about here]

          In general, consistent with the results from Tables 5 and 6, Table 7 shows that banks are

less likely to fail with lower minority lending concentration, lower concentration of refinance

loans and lower loan-to-income ratios. More importantly, as we examine the role of financial

information, we find that traditional banking with higher deposit ratios (deposit relative to total

bank assets) significantly reduces the probability of failure (1% increase in deposits relative to

total assets reduces the probability of failure by 2.5%.). Naturally the higher percentage of

problem loans and off balance sheet items are associated with significantly higher default


          Overall, considering the role of regulations, we find that although the higher capital

reserve ratio slightly reduces the probability of failure, the loan reserves and liquid asset reserves

are ineffective in evading failures. Especially the high odds ratios on the loan reserve ratios are

alarming because instead of providing insurance (security cushion) for loan losses, the loan

reserves seem to encourage even more risk taking. Putting our findings in the context of

proposed and new regulatory changes, we are suggesting more reliance on new measures, such

as increased liquidity. Liquidity ratios could be implemented in conjunction with different asset

and liability types. The traditional liquidity ratio, as percentage of deposits needs to be revised as

banks are less and less reliant on deposits.

       In response to the crisis, President Barack Obama proposed drastic financial reforms that

manifested in 2010 as the Dodd–Frank Wall Street Reform and Consumer Protection Act.

Naturally, the Act emphasizes the role of regulations, especially aiming to out regulatory

oversight―consolidate bank regulatory agencies and create new oversight council to evaluate

systemic risk. In addition, the Act promotes increased transparency to identify risks at

institutional level which potentially may cause system wide shocks. However, the planned

increased oversight and transparency may not be beneficial without clear objectives; as already,

prior to the crisis, staggering amount of lending information was available in the system which

was hard to process.

5.4    Robustness Analysis

       In robustness analysis, we have used alternative samples. First, we excluded Washington

Mutual and its subsidiaries and replicated the analysis from section 5. In addition, we also

considered alternative classification where a financial institution is considered depository

institution if the deposits account for non-negligible amount of the assets (using 5 and 8%

cutoffs). This latter robustness test is crucial to address misclassification issues, such as the case

of Indymac, where the lender is clearly a non-traditional subprime lender despite that it owns a

commercial bank and can take deposits.

6.     Conclusion

       Since 2008 hundreds of banks filed for bankruptcy, have been acquired, or merged with

FDIC assistance. In addition, hundreds of non-bank lenders (e.g., now infamous subprime

lenders such as New Century) have also failed resulting in a near collapse of the subprime

lending market segment. With the failures of these non-bank lending institutions, large

commercial banks continued to increase their market size and now a world-wide phenomenon is

that a handful of large institutions dominate the lending market. Examining the failures of non-

bank lenders in conjunction with bank failures provide a unique opportunity to shed light on the

effectiveness of market and regulatory disciplines in the mortgage market.

       In this study, we aim to contribute to the literature on bank and non-bank lenders failures

by focusing on institutional details (e.g., bank management and lending practices) in explaining

the cross-sectional variation in lenders’ failures in the aftermath of the recent financial crisis.

Like other studies on the recent mortgage crisis, we are unable to compare the risk management

practices and financial information of bank and non-bank lenders in the absence of reliable

financial data for the non-bank lenders. To address this shortcoming, we use information on

mortgage lending practices from HMDA in our empirical analysis. Specifically, we focus on the

role of market discipline and aggressive lending practices in examining the non-bank failures. By

contrasting the role of potentially aggressive lending practices in bank versus non-bank lender

failures, we provide new insight about the role of regulatory discipline.

       Our findings suggest that lending practices, such as sound lending standards, play a key

role in reducing the probability of lender failures. In general, the higher concentration of high

credit quality borrowers and low average loan amount are associated with lower probability of

failure. For non-banks we find that self-regulation through transparency and competition was not

only ineffective but rather harmful. First, we find that lenders overly exposed to refinance and

minority loans are more likely to fail. These strong relationships between the potentially

aggressive lending practices and failure are especially alarming since the public has been

informed about these lending practices for years prior to the crisis. Second, the competitive

environment did not promote better lending practices but rather more risk taking and increased

the probability of failures. The presence of high concentration of lenders, when the number of

qualified borrowers is limited, is likely to encourage lenders to relax lending standards in an

attempt to issue more loans for greater revenues.

       Interestingly, despite the significant differences in institutional details and regulatory

requirements, the failures of bank and non-bank lenders are driven by similar factors. At the first

glance, the comparable failure rates across the bank and non-bank lenders already suggests that

bank regulations were inefficient. Focusing on the bank subsample in examining the role of

regulatory credit risk measures, the capital reserve requirement provides only weak protection,

while the loan reserve ratio is entirely ineffectively. The strong positive relationship between

failure and loan reserve suggests that banks knowingly take risk increase reserves but the higher

level of reserves are still insufficient. We find that the higher liquid asset ratios provided the best

tool to reduce the probability of failures. Thus, we recommend that regulators consider new

alternative risk measures instead of strengthening the relatively inefficient existing regulatory


7.      References

Aharony, J. and Swary, I., (1996). Additional Evidence on the Information-Based Contagion

     Effects of Bank Failures, Journal of Banking & Finance 20 (1), 57-69.

Ambrose, B. W., Benjamin, J. D., and Chinloy, P., (2003). Bank and Non-bank Lenders and the

     Commercial Mortgage Market. Journal of Real Estate Finance and Economics 26 (1), 81-94.

Avery, R. B., Brevoort, K. P., and Canner, G. B., (2006). The 2006 HMDA Data. Federal

     Reserve Bulletin 3, A73-A109.

Avery, R. B., Belton, T. M., and Goldberg, M. A., 1988. Market Discipline in Regulating Bank

     Risk: New Evidence from the Capital Markets, Journal of Money, Credit and Banking 20 (4),


Belratti, A., and Stulz, R., (2010). The Credit Crisis Around the Globe: Why did Some Banks

     Perform Better? Working paper.

Boyd, J.H., De Nicoló G, and Jalal, A. M. (2009). Bank Competition, Risk, and Asset

     Allocations, IMF Working Paper

Boyd, J. H., De Nicoló G., and Jalal, A. M., (2010). Bank Competition, Asset Allocation and

     Risk of Failures: An Empirical Investigation. Working paper.

Calem, P., Gillen, S. K., and Wachter, S. (2004). The Neighborhood Distribution of Subprime

     Mortgage Lending. Journal of Real Estate Finance and Economics 29 (4), 393-410.

Chinloy, P., and Macdonald, N., (2005). Subprime Lenders and Mortgage Market Completion.

     Journal of Real Estate Finance and Economics 30 (2), 53-165. Special Issue: Subprime

     Mortgage Market, Part II: Theoretical and Empirical Studies.

Chomsisengphet, S., and Pennington-Cross, A., (2006). The Evolution of the Subprime

     Mortgage Market. Federal Reserve Bank of St. Louis Review 88 (1), 31-56.

Cole, R. A., and W. Q., (2010). Predicting bank failures using a simple dynamic hazard model.

   Working paper.

Cole, R. A., and Gunther, J. W., (1995). Separating the Likelihood and Timing of Bank Failure.

   Journal of Banking & Finance 19 (6), 1073-1089.

Courchane, M., J., Surette, B. J., and Zorn, P. M., (2004). Subprime Borrowers: Mortgage

   Transitions and Outcomes. Journal of Real Estate Finance and Economics 29 (4), 365-392.

Fortowsky, Elaine B. and Michael LaCour_Little., (2002). An Analytical Approach to

   Explaining the Subprime-Prime Mortgage Spread. Presented at the Georgetown University

   Credit Research Center Symposium Subprime Lending.

Hannan, T.H. and Hanweck, G. A. (1988). Bank Insolvency Risk and the Market for Large

   Certificates of Deposit. Journal of Money, Credit and Banking 20, 203-11.

Hill, R. P., & Kozup, J. C., (2007). Consumer Experiences with Predatory Lending Practices.

   Journal of Consumer Affairs, 41(1), 29-46,

Huszár, Z. R., Lentz, G. H., and Yu, W., (2010). Does Mandatory Disclosure Affect Subprime

   Lending to Minority Neighborhoods? Journal of Economics and Finance, Forthcoming.

Jaffee, D. M., (2008). The U.S. Subprime Mortgage Crisis: Issues Raised and Lessons Learned,

   Working paper.

Klomp, J., (2010). Causes of Banking Crises Revisited. The North American Journal of

   Economics and Finance 21 (1), 72-87.

Iyer, R., and Peydró, J-L., (2010). Contagion at Work: Evidence from a Natural Experiment.

   Review of Financial Studies, Forthcoming.

LaCour-Little, M., (2001). A Note on Identification of Discrimination in Mortgage Lending.

   Real Estate Economics 29 (2), 329-335.

Mian, A. R., and Sufi, A., (2009). The Consequences of Mortgage Credit Expansion: Evidence

   from the 2007 Mortgage Default Crisis. Quarterly Journal of Economics 124 (4), 1449-1496.

Mian, A. R., Trebbi. F., and Sufi, A., (2010). The Political Economy of the U.S. Mortgage

   Default Crisis. American Economic Review 100 (5), 1967-98.

Poole, W., (2010). Causes and Consequences of the Financial Crisis of 2007-2009, Harvard

   Journal of Law & Public Policy 33 (2), 421-441.

Smith, B. C., (2007). The Subprime Mortgage Market, a Review and Compilation of Research

   and Commentary. Hoyt Institute Working paper.

Williams, R., and Bond, C., (2007). Residential Segregation and the Transformation of Home

   Mortgage Lending. Social Forces 86 (2), 671-698.

Whalen, G., (1991). A Proportional Hazards Model of Bank Failure: An Examination of Its

   Usefulness as an Early Warning Tool. Economic Review, Federal Reserve Bank of

   Cleveland, 21-31.

Table 1. Variable descriptions
In general, all variables are calculated as the time series average of the annual number from 2003 to 2007. For
example, the application variable is average of the annual total applications from 2003 to 2007. The change variables
are calculated as the time series average of the annual changes to capture time trend. For example, the Chngamount is
the average in the five year annual changes in the total issued loan amount from 2003 to 2007.

Panel A. Lending variables based on HMDA dataset
 Variable name          Description of variables
 Failure                Dummy variable that takes on the value one for failed lenders (banks and non-banks) , the
                       lender is unable to operate any longer in the residential market.
 Application           Total number of loan applications
 Originate             Total number of originated loans
 Pctoriginate           Approval rate in percentage, approved loans relative to total loan application
 Loan-to-income       Average loan-to-income ratio, excluding high cost loans, such as HOEPA and spread-
                      reportable loans
 Pctminority           Percentage of total loan application by minorities
 PctgoodFICO           Percentage of total loan application with high credit score, as the percentage of loans from
                       census tracts where the median credit score is in the top quartile relative to total loan
 Avgamount             Annual average of the approved loan amount
 Amount                Annual total loan amount issued by the lender
 Pctpurchase           Percentage of loan application for purchase purposes, as the ratio of purchase applications
                      relative to total loan applications
 Pctrefinance          Percentage of loans application for refinance purposes, as the ratio of refinance
                       applications relative to total loan applications
 HHI                   Banks’ Herfindahl-Hirschman Index (HHI), which is the average of all census tracts,
                       where the lender has issued loans
 High_HPI               Dummy variable which takes on the value one if more than half of the originations are
                       in areas with high property prices. Areas with high property prices are identified as
                       those with HPIs from the top quartile.
 Concentration        Geographic lending concentration measure, calculated as sum of squares of the total
                       lending (number of loans) per state
 Highrate              Annual average of total number of spread reportable loans
 HOEPA                Annual average of total number of HOEPA loans
 Chngamount           Annual average of the increase (or decrease) in the total amount of loan issued from 2003 to
 Chngloanorig          Annual average of the increase (or decrease) in the total number of loan originated from
                       2003 to 2007
 Chngpctapproval      Annual average of the increase (or decrease) in approval rates from 2003 to 2007
 Chngpctminority      Annual average percentage of loan application by minorities from 2003 to 2007
 ChnggoodFICO         Annual average change in the percentage of loans form census tract with high FICO
                       scores, FICO scores from top quartiles

Panel B. Financial variables from Bankscope for bank lenders

Variable name            Bankscope data                        Description of variables (in Mill)
Tasset (in $bill)        data5670                              Total Assets (in Bill)
Totalcap                 data2135                              Total capital reserve
Tier1cap                 data2130                              Total tier one capital
Equity                   data2055                              Total common equity (in Bill)
Liability (in $bill)     data2060 - data2055                   Total liability(in Bill)
                                                               Total earning assets relative to total
Earningasset/liability   data2010 / liability
                                                               Total off balance sheet items relative to total
OBS_cap                  data2065
Liqat                     data2075                             Total liquid assets
NI                        data2115                             Net income
Profit                    data2105                             Profit
NI_liab                   data2115/data5670                    Net income relative to total liabilities
Totalloans                data5330                             Total Loans
TotalfixAt                data5660                             Total fixed assets
Tdeposit_loan             data6080/data5330                    Total deposits relative to total loans
Tprobloan_cap            data2150/data2135                     Total problem loans relative to capital
Tprobloans_liqAt         data5240/ data2075                    Total problem loans relative to liquid assets
Loanres_loan             data2070/ data5330                    Total loan reserves relative to total loans
ChngT1ratio                                                    Percentage change in tier one capital reserve
ChngTotalcap                                                   Percentage change in total capital reserve
ChngLiqAT                (data2075-
                                                               Percentage change in total liquid assets
                         (data2055- lag(data2055))
ChngEq                                                         Percentage change in total common equity
                         (data2065-                            Percentage change in off balance sheet
                         lag(data2065))/lag(data2135)          assets relative to total capital
ChngProfit_cap           (data2105-                            Average change profit relative to common
                         lag(data2105))/lag(data2135)          equity
                         (data2115-                            Average change net income relative to total
                         lag(data2115))/lag(data2135)          loans
                         (data2070- lag(data2070))/lag(        Average change in total loan reserves
                         data5330)                             relative to total capital
ChngTasset               (data5670-                            Average change in total assets scaled by
                         lag(data5670))/lag(data5670)          lagged total assets
                         (data5330-                            Average change in total loans relative to
                         lag(data5330))/lag(data2135)          total capital
                         (data2150-                            Average change in total problem loans
ChngTprobloans_cap       lag(data2150))/lag(data2135)          relative to total capital

Table 2. Summary statistics of mortgage lending activities for failed bank and non-banks.
The detailed description of the variables are shown in Panel A of Table 1.
Panel A. Failed banks
Variables                    N            MIN                MAX           MEAN              STD
Application                      249         4.500          2431891.500    35135.071        162639.856
Pctoriginate                     249         0.000                1.000        0.494             0.232
Loan-to-income                   248         0.638               86.021        2.789             5.385
Pctminority                      249         0.000                1.000        0.519             0.204
Pctgoodfico                      249         0.000                0.800        0.222             0.127
Avgamount (‘000)                 249        33.312             9573.753      220.134           601.006
Amount (Bill)                    249         0.004             4732.701       63.792           313.734
Pctpurchase                      249         0.000                1.000        0.404             0.254
Pctrefinance                     249         0.000                1.000        0.554             0.232
HHI                              249       235.556             1533.397      350.721           108.666
High_HPI                         249         0.000                1.000        0.525             0.476
Concentration                    247       414.114            10000.000     4882.047          3523.814
Highrate                         249         0.000                9.000        3.180             2.033
HOEPA                            249         0.000                8.667        1.167             1.597
Chngamount (in mill)             223        -8.807              171.265        2.710            12.485
Chngloanorig                     223    -35538.000           276605.333     5096.777         22026.970
Chngpctapproval                  223        -0.972                0.747       -0.070             0.178
Chngpctminority                  223        -0.556                0.914        0.057             0.171
ChnggoodFICO                     223        -0.667                0.433       -0.017             0.116
Panel B. Failed non-banks
Variables                        N         MIN                 MAX          MEAN              STD
Application                      329           1.500         2431891.500    27142.123       168356.432
Pctoriginate                     329           0.045               1.000        0.656            0.252
Loan-to-income                   327           0.331              24.961        2.239            1.709
Pctminority                      329           0.000               1.000        0.393            0.226
PctgoodFICO                      329           0.000               0.922        0.242            0.186
Avgamount (‘000)                 329          33.312           35004.445      381.216         1986.979
Amount (Bill)                    329           0.003            4732.701       53.354          340.129
Pctpurchase                      329           0.000               1.000        0.412            0.231
Pctrefinance                     329           0.000               1.000        0.467            0.221
HHI                              329         222.196            1967.938      389.288          167.534
High_HPI                         329           0.000               1.000        0.330            0.460
Concentration                    329         397.048           10000.000     7992.232         2860.919
Highrate                         329           0.000               9.000        2.779            2.057
HOEPA                            329           0.000               7.000        0.784            1.370
Chngamount (in mill)             308          -2.402             171.265        2.239           13.142
Chngloanorig                     308       -5623.333          276605.333     4392.027        24855.874
Chngpctapproval                  308          -0.939               0.747       -0.028            0.149
Chngpctminority                  308          -0.750               0.889        0.036            0.179
ChnggoodFICO                     308          -0.667               0.500       -0.013            0.141

Table 3. Summary statistics of HMDA variables for match (non-failed) banks and non-banks
The detailed variable description is in Panel A of Table 1.

                                              Non-failed (active) banks                                  Non-failed (still active) non-banks
Variables                 N        MIN           MAX           MEAN         STD             N        MIN         MAX           MEAN          STD
Application              2240         1.000    146091.75        816.835   4685.682         3215        1.000   1287758. 2749.048 31748.302
Pctoriginate             2240         0.006         1.000         0.791      0.147         3215        0.000       1.000          0.751        0.224
Loan-to-income           2238         0.141       31.374          1.757      1.068         3209        0.112      29.151          1.951        1.040
Pctminority              2240         0.000         1.000         0.242      0.206         3215        0.000       1.000          0.360        0.250
PctgoodFICO              2240         0.000         1.000         0.303      0.285         3215        0.000       1.000          0.259        0.218
Avgamount (‘000)         2240        11.406     4694.989        164.942    205.832         3215        5.317   15982.78         148.327      365.562
Amount (bill)            2240         0.000       13.005          0.109      0.566         3215        0.000     207.055          0.446        5.098
Pctpurchase              2240         0.000         1.000         0.424      0.155         3215        0.000       1.000          0.358        0.269
Pctrefinance             2240         0.000         0.953         0.403      0.146         3215        0.000       1.000          0.467        0.231
HHI                      2240         0.000     2099.517        416.636    187.489         3215        0.000   2505.969         359.056      136.657
High_HPI                 2240         0.000         1.000         0.204      0.398         3215        0.000       1.000          0.381        0.476
Concentration            2240       580.954    10000.000      9008.233    1543.376         3154      358.374   10000.00 8311.468           2362.765
Highrate                 2240         0.000         9.000         2.449      1.865         3215        0.000       9.000          1.605        1.828
HOEPA                    2240         0.000         7.667         0.468      1.041         3215        0.000       8.000          0.253        0.764
Chngamount (in mill)     2159       -75.027         5.602        -0.128      2.385         2914      -82.235      31.920         -0.263        3.943
Chngloanorig             2159   -142237.333    14902.333       -224.284   4005.847         2914   -27501.000   79202.33 -593.505           8888.673
Chngpctapproval          2159        -0.859         0.786         0.004      0.104         2914       -0.929       1.000          0.001        0.163
Chngpctminority          2159        -0.925         0.938        -0.012      0.134         2914       -1.000       1.000          0.007        0.164
ChnggoodFICO             2159        -0.979         0.862         0.009      0.147         2914       -0.881       0.731         -0.001        0.133

Table 4. Summary statistics of balance sheet information for failed and non-failed banks
The detailed discretion of the variables is shown in Panel B of Table1.

                                                  Failed (inactive) banks                                         Non-failed (active) banks
 Variables                 N         MIN        MAX            MEAN       STD              N          MIN         MAX           MEAN           STD
 Tasset (in bill)              267      0.000         0.485        0.009        0.053          2240       0.000        0.088           0.001      0.004
 Totalcap (in mill)            264      1.645    48683.388      890.537      5281.968          2238       1.930    10474.928          85.771    462.418
 Tier1cap (in mill)            266      4.975      477.071       14.588        31.600          2238       7.153       86.675          14.755      7.010
 Equity (in mill)              267      1.916    56778.500      998.658      6027.316          2240       2.270     8425.375          93.490    499.440
 Liability (in bill)           267      0.000         0.431        0.008        0.047          2240       0.000        0.080           0.001      0.004
 Earningasset_liab             267      0.788       50.110         1.215        3.017          2240       0.690        1.932           1.021      0.054
 OBS_cap                       263    -27.182       41.486         2.077        3.905          2238       0.000      262.189           1.555      6.056
 liqAT                         267      1.781    60515.500      855.240      5277.740          2240       0.911    21979.375          55.796    523.882
 NI (in mill)                  267   -150.924     6896.042       91.790       653.583          2240    -101.196      781.654           6.972     40.075
 Profit (in mill)              267   -116.029     9177.000      131.147       879.833          2240    -126.889     1166.354          10.225     59.493
 NI_liab                       267     -0.052       36.541         0.138        2.236          2240      -0.081        0.381           0.009      0.011
 Totaloans                     267      0.000   319395.917 5774.676         33285.149          2240       6.866    64732.829        574.229    2855.614
 TotalfixAt                    267      0.000     6081.938       98.650       592.719          2240       0.039      947.683          11.421     43.855
 Tdeposit_loanratio            266     -0.384         3.893        1.144        0.367          2239       0.485       93.037           1.373      2.023
 Tprobloan_cap                 264     -8.033       42.639         0.229        2.674          2238      -0.010        1.776           0.026      0.058
 Tprobloans_liqAt              267      0.000         4.639        0.475        0.554          2240       0.000       10.198           0.191      0.331
 Loanres_loan                  266     -0.140         0.087        0.016        0.012          2240       0.001        0.111           0.013      0.006
 ChngT1ratio                   262     -0.947         6.491       -0.033        0.524          2238      -7.671        5.172           0.023      0.275
 ChngTotalcap                  262     -0.938       68.465         0.725        4.643          2238      -4.672       91.399           0.426      2.485
 ChngLiqAT                     265     -0.659       69.314         1.727        5.646          2240      -0.973      449.640           1.274     14.296
 ChngEq                        265     -1.872       68.407         0.762        4.775          2240      -5.934       99.540           0.526      3.253
 ChngOBS                       259     -1.000      411.263         7.582       37.780          2237      -1.000     2099.727           4.789     54.596
 ChngProfit_cap                264   -335.640       17.786        -1.156       20.700          2239      -3.342        6.478           0.074      0.369
 ChngNI_cap                    264   -220.990       10.959        -0.779       13.626          2239      -3.295        3.820           0.052      0.247
 ChngLoanRes_loan              264     -0.062         0.275        0.009        0.025          2240      -0.043       48.866           0.053      1.292
 ChngTasset                    265     -0.856      146.488         1.059        9.070          2240      -0.974      234.780           0.705      5.877
 Chngtotaloans_cap             264     -6.866      238.246         5.564       22.086          2239     -13.993      733.508           3.080     18.880
 ChngTprobloans_cap            263     -0.601         2.478        0.047        0.190          2239      -1.526        3.845           0.006      0.123

Table 5. Probability of failure analysis of bank and non-bank lenders
The dependent variable failure takes on the value 1 for lenders that failed by August 2010, zero otherwise. The
explanatory variables capture the time series average lending practices based on HMDA data for 2003 to 2007.
Pctoriginate is the ratio of origination relative to loan application in percentage. Logamount is the natural
logarithm of the average loan amount. Pctorg*Logamount is an interaction variable of Pctoriginate and
Logamount variables. PctgoodFICO is the percentage of total loan application from census tracts with Average
FICO scores above the median. Pctminority is the percentage of loan applications by minorities. Pctrefinance is
the percentage of loan application for refinance purposes. Loan-to-income is the average loan-to-income ratio
for all loans by lenders. HighComp is a dummy variable that takes on the value 1 for lenders operating in highly
competitive environment, where the Concentration measure is below the median. Highrate is the annual average
of total number of spread reportable loans. High_HPI is the dummy variable which takes on the value one if
more than half of the originations are in areas with high property prices. The 1 percent, 5 percent and 10 percent
significance levels are denoted with ***, **, and *, respectively. To save space the intercepts are not shown.

                                         Non-banks                       Banks
                                 odds ratio    odds ratio      odds ratio    odds ratio
                                     (1)            (2)            (1)            (2)
 Pctoriginate                      0.918*        0.918**          1.067          1.068
                                  (-1.859)       (-1.987)       (1.401)        (1.412)
 Logamount                          1.158          1.143        1.827**        1.800**
                                   (0.937)        (0.941)        (2.365)        (2.293)
 Pctorg*Logamount                   1.005          1.005         0.993*         0.993*
                                   (1.388)        (1.452)       (-1.664)       (-1.673)
 PctgoodFICO                        0.985          0.986         0.987*         0.988*
                                  (-1.466)       (-1.262)       (-1.924)       (-1.928)
 Pctminority                        0.998          1.000       1.018***       1.018***
                                  (-0.477)       (0.026)        (5.186)        (4.223)
 Pctrefinance                    1.010***       1.010***          1.013          1.014
                                   (4.012)        (4.435)        (1.413)        (1.514)
 Loan-to-income                     1.001          0.999       1.175***       1.179***
                                   (0.162)       (-0.265)       (3.607)        (3.666)
 Focus                           0.812***       0.794***          0.807          0.800
                                  (-2.897)       (-2.819)       (-1.228)       (-1.295)
 HighComp                           1.160         1.214*          1.356          1.383
                                   (1.355)        (1.656)        (1.569)        (1.552)
 Highrate                                       1.235***                      1.075***
                                                  (4.577)                       (2.583)
 High_HPI                                          1.026                         1.190
                                                  (0.671)                       (1.166)

 Observations                      3394           3394           2562           2562
 pseudo R-squared                  0.390          0.404          0.182          0.184

Table 6. Probability of failure analysis of bank and non-bank lenders with time trends
The dependent variable failure takes on the value 1 for lenders that failed by August 2010, zero otherwise. The
explanatory variables capture the time series average lending practices based on HMDA data for 2003 to 2007.
Pctoriginate is the ratio of origination relative to loan application in percentage. Logamount is the natural
logarithm of the average loan amount. Pctorg*Logamount is an interaction variable of Pctoriginate and
Logamount variables. PctgoodFICO is the percentage of total loan application from census tracts with Average
FICO scores above the median. Pctminority is the percentage of loan applications by minorities. Pctrefinance is
the percentage of loan application for refinance purposes. Loan-to-income is the average loan-to-income ratio
for all loans by lenders. HighComp is a dummy variable that takes on the value 1 for lenders operating in highly
competitive environment, where the Concentration measure is below the median. Highrate is the annual average
of total number of spread reportable loans. High_HPI is the dummy variable which takes on the value one if
more than half of the originations are in areas with high property prices. Chngloanamount is the average
annual change in the total loan amount by the lender from 2003 to 2007. Chngpctapproval is the average annual
percentage change in approval rates. Robust z statistics are reported in parentheses. The 1 percent, 5 percent and
10 percent significance levels are denoted with ***, **, and *, respectively. To save space the intercepts are not

                                         Non-banks                         Banks
                                odds ratio      odds ratio       odds ratio        odds ratio
                                    (1)             (2)              (1)               (2)
 Pctoriginate                    0.891**         0.878**           1.069              1.056
                                 (-2.434)        (-2.453)          -1.483            (1.144)
 Logamount                        0.973            0.942          1.796**           1.732**
                                 (-0.157)        (-0.302)           -2.37            (2.233)
 Pctorg*Logamount                 1.007*          1.008*           0.993*             0.994
                                  -1.952          (1.900)         (-1.728)          (-1.571)
 PctgoodFICO                    0.977***        0.977***          0.986**           0.986**
                                 (-3.841)        (-3.907)         (-2.096)          (-2.188)
 Pctminority                      0.998            0.998         1.017***          1.017***
                                 (-0.374)        (-0.518)          -3.396            (3.512)
 Pctrefinance                   1.012***        1.012***           1.013              1.012
                                  -8.062          (8.188)          -1.214            (1.292)
 Loan-to-income                   1.004*           1.003         1.158***          1.154***
                                  -1.892          (1.245)          -3.096            (2.998)
 Focus                          0.770***        0.755***           0.775*            0.739*
                                 (-5.142)        (-6.201)         (-1.803)          (-1.936)
 HighComp                         1.211            1.310             1.35             1.363
                                   -1.29          (1.344)          -1.534            (1.472)
 Highrate                       1.202***        1.183***           1.045              1.044
                                  -4.845          (4.083)          -1.358            (1.338)
 High_HPI                         0.997            1.028           1.094              1.087
                                 (-0.027)         (0.290)          -0.581            (0.548)
 Chngloanamount                   1.038            1.053         1.000***          1.000***
                                  -1.393          (1.531)          -3.465            (3.695)
 Chngpctapproval                                5.963***                            6.382**
                                                  (6.129)                            (2.397)

 Observations                     3086             3086            2461              2461
 pseudo R-squared                 0.437            0.444           0.186             0.193

Table 7. Probability of failure analysis of bank lenders with financial information
The dependent variable is the failure dummy in a logistic regression. The lending characteristics variables are
defined in Tables 5 and 6. Tier1capratio is the total tier1 capital ratio in percentage. NI_asset is the ratio of Net
income to earning assets. Probloanratio is the ratio of total problem loans to total loans. Depratio is the ratio of
deposits to total assets for the bank. LogtotalfixAt is the natural logarithm of total fixed assets. Tprobloan_cap is
the total problem loans relative to the total capital reserve. OBS_cap is the total off balance sheet items relative
to total capital reserve. Loanres_loan is the total loan reserves relative to the total loans. Tprobloans_liqAt is the
total problem loans relative to total liquid assets. Robust z-statistics are reported in parentheses. The 1 percent, 5
percent and 10 percent significance levels are denoted with ***, **, and *, respectively. To save space the
intercepts are not shown.

                            Odds ratio    Odds ratio     Odds ratio    Odds ratio    Odds ratio
                                (1)           (2)            (3)           (4)           (5)
Pctoriginate                1.089***      1.131***       1.124***      1.134***      1.123***
                              (2.659)       (5.446)        (3.597)       (4.406)       (4.504)
Logamount                   1.964***      2.782***       2.875***      3.112***      2.590***
                              (5.934)      (15.263)        (6.458)      (10.516)       (9.993)
Pctorg*Logamount            0.992***      0.988***       0.989***      0.988***      0.989***
                             (-2.811)      (-5.203)       (-3.488)      (-4.227)      (-4.312)
PctgoodFICO                 0.989***      0.988***       0.986***      0.988***      0.986***
                             (-2.663)      (-2.966)       (-3.981)      (-4.347)      (-5.382)
Pctminority                 1.018***       1.010**       1.015***      1.014***      1.013***
                              (4.413)       (2.569)        (3.771)       (3.402)       (3.284)
Pctrefinance                   1.008      1.017***         1.012*       1.012**       1.013**
                              (1.568)       (2.685)        (1.821)       (2.145)       (2.300)
Loan-to-income              1.156***       1.068**        1.097**       1.088**      1.098***
                              (2.916)       (2.455)        (2.205)       (2.288)       (2.596)
High_HPI                    1.307***       1.398**         1.246*      1.339***      1.280***
                              (2.595)       (2.484)        (1.696)       (3.350)       (2.816)
Tier1capratio                 0.892*      0.890***       0.857***      0.865***      0.869***
                             (-1.934)      (-2.780)       (-2.921)      (-3.536)      (-3.191)
NI_asset                                  0.539***       0.389***      0.432***      0.428***
                                           (-6.301)       (-7.715)      (-7.915)      (-8.137)
Deposit_ratio               0.985***
Probloanratio                              1.728***
LogtotalfixAt                                1.057
Tprobloan_cap                                            1.266***      1.262***         1.048
                                                          (2.881)       (4.650)        (0.948)
OBS_cap                                                   1.042**
Loanres_loan                                                           2.052***       1.827***
                                                                        (7.505)        (5.954)
Tprobloans_liqAt                                                                      3.139***

Observations                  2487           2487          2451          2452           2452
pseudo R-squared              0.147          0.328         0.261         0.282          0.295

Appendix – Table 1

The extract shows some of the largest bank and non-bank lenders as of 2007, and the amount of high cost loans
that these institutions issued. The assets for Banks (depository institutions) are reported in million USD, while
the assets for non-banks reflect the minimum asset requirement (defined by the states).

* reflects subprime lenders with at least one depository facility therefore considered banks in this classification,
although these institutions are widely viewed as non-banks.

Non-banks               High-cost loans        Asset          Banks                    High-cost Loans        Asset
New Century Mortgage Corp.         193,04      10,000         Countrywide Home Loans *           210,688      175.09
WMC Mortgage Co.                   142,16      31,000         National City Bank                 189,415       69.48

Option One Mortgage Corp           118,43      10,000         Fremont Investment & Loan          140,290       11.32

Argent Mortgage Co.                107,53      10,000         Wells Fargo Bank, NA               110,770      403.26

American Home Mortgage             78,936      10,000         Long Beach Mortgage Co.             95,427      330.71

Accredited Home Lenders,           75,931      10,000         Decision One Mortgage               73,379       3.39

Homecoming Financial               62,730      10,000         Beneficial Company LLC              67,303       4.02

Novastar Mortgage, Inc.            43,733      10,000         Indymac Bank, FSB*                  62,790       20.33

Wilmington Finance, Inc.           41,212      10,000         Equifirst Corp.*                    62,685       81.07

Appendix – Table 2
FailD takes on the value one for lenders that fail (closed down, merged, or acquired) by August 2010, zero otherwise. Pctoriginate is the ratio of origination relative to loan
application in percentage. Logamount is the natural logarithm of the average loan amount. Orig*Amount is an interaction variable of Pctoriginate and Logamount variables.
PctgoodFICO is the percentage of total loan application from census tracts with Average FICO scores above the median. Pctminority is the percentage of loan applications by
minorities. Pctrefinance is the percentage of loan application for refinance purposes. Loan-to-income is the average loan-to-income ratio for all loans by the specific lender.
HighComp is a dummy variable that takes on the value 1 for lenders operating in highly competitive environment, where the Concentration measure is below the median.
Highrate is the annual average of total number of spread reportable loans. High_HPI is the dummy variable which takes on the value one if more than half of the originations
are in areas with high property prices. The pair-wise correlation coefficient estimates are reported with the corresponding p-values in italic.

Panel A. Correlation analysis results for non-bank residential lenders
                                                         Orig*                                                          Loan-to-
                 Faild      Pctoriginate Logamount amount            PctgoodFICO           Pctminority   Pctrefinance   income      Focus       HighComp      Highrate    High_HPI
 Faild               1.000
 Pctoriginate       -0.283          1.000
 Logamount            0.405          -0.359       1.000
                       0.000          0.000
 Orig*amount         -0.088          0.782        0.253       1.000
                       0.000          0.000       0.000
 PctgoodFICO         -0.045          0.194        0.098       0.272          1.000
                       0.008          0.000       0.000       0.000
 Pctminority          0.164          -0.325       0.244      -0.201          -0.241            1.000
                       0.000          0.000       0.000       0.000           0.000
 Pctrefinance         0.097          -0.025       0.054      -0.034          0.068            -0.003         1.000
                       0.000          0.148       0.002       0.045           0.000            0.880
 Loan-to-             0.122          -0.195       0.272      -0.027          -0.001            0.179         0.069       1.000
 income                0.000          0.000       0.000       0.110           0.946            0.000         0.000
 Focus               -0.377          0.323       -0.551       0.014          0.021            -0.199        -0.068      -0.141       1.000
                       0.000          0.000       0.000       0.420           0.230            0.000         0.000       0.000
 HighComp            -0.068          0.049       -0.106       0.012          0.046            -0.044        -0.014      -0.018       0.101       1.000
                       0.000          0.004       0.000       0.479           0.007            0.010         0.429       0.289       0.000
 Highrate             0.216          -0.072       0.360       0.146          -0.024            0.043        -0.027       0.071      -0.177      -0.081       1.000
                       0.000          0.000       0.000       0.000           0.167            0.012         0.113       0.000       0.000        0.000
 High_HPI             0.078          -0.199       0.198      -0.055          0.007             0.386         0.060       0.187      -0.045       0.072       0.028       1.000
                       0.000          0.000       0.000       0.001           0.667            0.000         0.000       0.000       0.008        0.000       0.102

Panel B. Correlation analysis results for bank residential lenders
                                                          Orig*                                                   Loan-to-
                 Faild       Pctoriginate Logamount amount           PctgoodFICO    Pctminority    pctrefinance   income     Focus      Highcomp    Highrate   High_HPI
 Faild               1.000
 Pctoriginate       -0.266          1.000
 Logamount           0.276         -0.371        1.000
                     0.000          0.000
 Orig*amount        -0.116          0.737        0.323       1.000
                     0.000          0.000         0.000
 PctgoodFICO        -0.074          0.131        0.110       0.208         1.000
                     0.000          0.000         0.000      0.000
 Pctminority         0.234         -0.298        0.120      -0.233        -0.153          1.000
                     0.000          0.000         0.000      0.000         0.000
 Pctrefinance        0.135         -0.112        0.335       0.096         0.226         -0.170          1.000
                     0.000          0.000         0.000      0.000         0.000           0.000
 Loan-to-            0.136         -0.075        0.160       0.045         0.141          0.214          0.092       1.000
 income              0.000          0.000         0.000      0.022         0.000           0.000          0.000
 Focus              -0.221          0.249        -0.448     -0.010         0.010         -0.183          -0.073     -0.042      1.000
                     0.000          0.000         0.000      0.619         0.622           0.000          0.000      0.032
 HighComp           -0.029          0.052        -0.079      0.007         0.135         -0.031          0.037       0.030      0.055      1.000
                     0.138          0.008         0.000      0.708         0.000           0.115          0.062      0.134      0.005
 Highrate            0.058         -0.055        0.113       0.018        -0.207         -0.137          0.015      -0.111     -0.069      -0.172      1.000
                     0.003          0.005         0.000      0.360         0.000           0.000          0.442      0.000      0.000       0.000
 High_HPI            0.103         -0.164        0.114      -0.083         0.038          0.334          -0.064      0.163     -0.041      0.054      -0.154      1.000
                     0.000          0.000         0.000      0.000         0.057           0.000          0.001      0.000      0.040       0.007      0.000


Shared By: