CREDIT BOOMS AND LENDING STANDARDS: EVIDENCE FROM THE SUBPRIME MORTGAGE MARKET Giovanni Dell’Ariccia, Deniz Igan, and Luc Laeven∗ First Version: December 2007 This Draft: February 2008 Abstract This paper links the current subprime mortgage crisis to a decline in lending standards associated with the rapid expansion of this market. We show that lending standards declined more in areas that experienced larger credit booms and house price increases. We also find that the underlying market structure mattered, with entry of new, large lenders triggering declines in lending standards. Finally, lending standards declined more in areas with higher mortgage securitization rates. The results are consistent with theoretical predictions from recent financial accelerator models based on asymmetric information, and shed light on the relationship between credit booms and financial instability. JEL classification codes: G21, E51 Keywords: credit boom, lending standards, mortgages, subprime loans, moral hazard, financial accelerators ∗ The authors are all at the IMF Research Department (Financial Studies Division). Dell’Ariccia and Laeven are also at the CEPR. We would like to thank Richard Baldwin, Tam Bayoumi, Stijn Claessens, Enrica Detragiache, Gianni De Nicolo, David Gussmann, Robert Hauswald, Patrick Honohan, Simon Johnson, Pete Kyle, Robert Marquez, Rebecca McCaughrin, Donald Morgan, Marcelo Pinheiro, Calvin Schnure, and seminar participants at the International Monetary Fund, Fannie Mae, George Washington University, American University, the Federal Reserve Bank of New York, and the University of Kansas for helpful discussions and/or comments on an earlier version of this paper. We would also like to thank Chris Crowe for sharing his data. Mattia Landoni provided outstanding research assistance. The views expressed in this paper are those of the authors and do not necessarily represent those of the IMF, its Executive Board, or its Management. Address for correspondence: Giovanni Dell’Ariccia, IMF, 700 19th Street NW, Washington, DC 20431 USA. E-mail: [email protected]. 1 I. INTRODUCTION Recent events in the market for mortgage-backed securities have placed the U.S. subprime mortgage industry in the spotlight. Over the last decade, this market has expanded rapidly, evolving from a small niche segment to a major portion of the U.S. mortgage market. Anecdotal evidence suggests that this trend was accompanied by a decline in credit standards and excessive risk taking by lenders, and possibly by outright fraud.1 Indeed, the rapid expansion of subprime lending, fueled by financial innovation, loose monetary conditions, and increased competition, is seen by many as a credit boom gone bad. Yet, few attempts have been made to link empirically lending standards and delinquency rates in the subprime mortgage market to its rapid expansion. To our knowledge, our paper is the first to do so. How does the recent increase in delinquency rates relate to the boom? How did lending standards change over the expansion? How did changes in local market structure and financial innovation affect lender behavior during the boom? What was the role of monetary policy? To answer these questions, we use data from over 50 million individual loan applications combined with information on local and national economic variables. Reminiscent of the pattern linking credit booms with banking crises, current mortgage delinquencies appear related to past credit growth. In particular, delinquency rates rose more sharply in areas that experienced larger increases in the number and volume of originated loans (Figure 1). We find evidence that this relationship is linked to a decrease in lending standards, as measured by a decline in loan denial rates and a significant increase in loan-to- income ratios, not explained by an improvement in the underlying economic fundamentals. 1 See, for example, FitchRatings (2007). 2 Consistent with recent theories suggesting that banks behave more aggressively during booms than in tranquil times, the size of the boom mattered. Denial rates declined more and loan-to-income ratios rose more where the number of loan applications rose faster. The subprime boom also shared other characteristics often associated with aggregate boom-bust credit cycles, such as financial innovation (in the form of securitization), changes in market structure, fast rising house prices, and ample aggregate liquidity.2 We find evidence that all these factors were associated with the decline in lending standards. Denial rates declined more in areas with a larger number of competitors and were further affected by the entry into local markets of large financial institutions. The increasing recourse to loan sales and asset securitization appears to have affected lender behavior, with lending standards deteriorating more in areas where lenders sold a larger proportion of originated loans. Lending standards also declined more in areas with more pronounced housing booms. Finally, easy monetary conditions also played an amplifying role. These effects were more pronounced in the subprime mortgage market than in the prime mortgage market, where loan denial decisions were more closely related to economic fundamentals. We obtain this evidence in an empirical model where, in addition to taking into account changes in macroeconomic fundamentals, we control for changes in the distribution of applicant borrowers and for the potential endogeneity of some of the explanatory variables. Specifically, we develop a two-stage regression framework, explained in detail later on, that exploits individual loan application data to control for changes in the quality of the pool of loan applicants. We focus on loan applications rather than originations to reduce 2 Indeed, some have compared the current situation to major financial crises in developed countries and emerging market economies (Reinhart and Rogoff, 2008). 3 further the concern for simultaneity biases. For further robustness, we run an instrumental variable specification of our model, where we instrument the subprime applications variable with the number of applications in the prime market. The contribution of this paper is twofold. First, the paper sheds some light on the origins of the current crisis by establishing a link between credit expansion and lending standards in the subprime mortgage market, and by identifying increased loan sales and changes in the structure of local credit markets as factors amplifying the decline in denial rates and the increase in loan-to-income ratios. Second, the boom-bust cycle of the subprime mortgage industry, beyond being of interest in itself, provides an excellent “lab case” to gain insights into the black box of credit booms in less developed credit markets. Subprime borrowers are generally riskier, more heterogeneous, can post less collateral, and have shorter or worse credit histories (if any) than their prime counterparts. These are all features often prevalent in developing countries. At the same time, the wealth of information available and the geographical variation (Figure 2) in this market allow us to control for several factors, such as changes in the pool of loan applicants, that are difficult to take into account when studying episodes of aggregate credit growth. Thus, the subprime mortgage market provides an almost ideal testing ground for theories of intermediation based on asymmetric information (and adverse selection in particular). From a policy point of view, the evidence in this paper alerts against the dangers arising from lax standards during credit booms, and it is relevant for the debate on cyclical management of prudential regulation and on the potential effects of monetary policy on banks’ risk-taking (Jimenez et al., 2007). To the extent that during booms standards decline 4 more than justified by economic fundamentals, our findings are consistent with the view that bankers have “an unfortunate tendency” to lend too aggressively at the peak of a cycle and that most bad loans results from this aggressive type of lending.3 That said, credit booms may still be beneficial. While, in light of the recent financial turmoil, it is easy to argue that standards were excessively lax, it is much harder to compute the benefits associated with greater access to credit and, hence, the net welfare effect of the subprime expansion. The rest of the paper is organized as follows. Section II reviews the related literature. Section III provides a description of the data and introduces some stylized facts. Section IV describes our empirical methodology. Section V presents the results. Section VI concludes. II. RELATED LITERATURE Several studies examine the interaction between economic fluctuations and changes in bank credit (Bernanke and Lown, 1991, Peek and Rosengren, 2000, Black and Strahan, 2002, and Calomiris and Mason, 2003). However, little evidence has been collected on how lending standards are related to credit booms. Asea and Blomberg (1998) find that loan collateralization increases during contractions and decreases during expansions, while Lown and Morgan (2003) show that lending standards are associated with innovations in credit. Jimenez, Salas, and Saurina (2006) find that during booms riskier borrowers obtain credit and collateral requirements decrease. A few papers have examined the recent boom from a house-price perspective, while not strictly focusing on the subprime market (Himmelberg et al., 2005, and Case and Shiller, 3 Former Federal Reserve Chairman Alan Greenspan in a speech delivered before the Independent Community Bankers of America on March 7, 2001. See also Bernanke (2007). 5 2003). The literature on subprime mortgages has instead largely focused on issues of credit access and discrimination and on what determines access to subprime versus prime lenders. Our loan level analysis builds on a model from Munnell et al. (1996) who show that race has played an important, although diminishing, role in the decision to grant a mortgage. A few papers examine how local risk factors affect the fraction of the market that uses subprime lending (Pennington-Cross, 2002). Other studies focus on how borrowers choose a mortgage and on their decision to prepay or default on a loan (Deng et al., 2000, Campbell and Cocco, 2003, and Cutts and Van Order, 2005). A few recent papers focus on how securitization affects the supply of loans (Loutskina and Strahan, 2007) and mortgage delinquencies. Demyanyk and Van Hemert (2007) find that delinquency and foreclosure rates of subprime borrowers are to a large extent determined by high loan-to-value ratios. Mian and Sufi (2007) link the increase in delinquency rates to a disintermediation-driven increase in loan originations, while Keys et al. (2007) find that loans that are easier to securitize default more frequently. Most theoretical explanations for variations in credit standards rely on financial accelerators based on the interaction of asymmetric information and business cycle factors (Bernanke and Gertler,1989, Kiyotaki and Moore, 1997, and Matsuyama, 2007; see Ruckes, 2004, for a review of this literature). Others focus on the potential for herding behavior by bank managers (Rajan, 1994), on banks’ limited capacity in screening applications (Berger and Udell, 2004), or on how strategic interaction among asymmetrically informed banks may lead to changes in lending standards during booms (Gorton and He, 2003, and Dell’Ariccia and Marquez, 2006). 6 III. DATA AND DESCRIPTIVE STATISTICS We combine data from several sources. Our main set of data consists of economic and demographic information on applications for mortgage loans. We use additional information on local and national economic environment and on home equity loan market conditions to construct our final data set. The individual loan application data come from the Home Mortgage Disclosure Act (HMDA) Loan Application Registry. Relative to other sources, including LoanPerformance and the Federal Reserve Bank’s Senior Loan Officer Opinion Survey, this dataset has the important advantage of covering extensive time-series data on both the prime and subprime mortgage markets. The availability of data on the prime mortgage market provides us with a control group generally unavailable to studies focusing on aggregate credit. By comparing prime and subprime mortgage lenders we are also able to identify differences between the two lending markets. Given the different risk profiles of the prime and subprime markets, we include variables that proxy for the risk characteristics of a loan application to enhance comparability of the results across the two markets. Enacted by Congress in 1975, HMDA requires most mortgage lenders located in metropolitan areas to collect data about their housing-related lending activity and make the data publicly available.4 The HMDA data covers a broad set of depository and nondepository financial institutions. Whether an institution is covered depends on its size, the extent of its activity in a metropolitan statistical area (MSA), and the weight of residential mortgage 4 The purpose of the Act was two-fold: enhance enforcement of anti-discriminatory lending laws and disseminate information to guide investments in housing. 7 lending in its portfolio.5 Comparisons of the total amount of loan originations in the HMDA and industry sources indicate that around 90 percent of the mortgage lending activity is covered by the loan application registry (Table 1). Our coverage of HMDA data starts from 2000 and ends in 2006. This roughly corresponds to the picking up of both the housing boom and the rapid subprime mortgage market expansion (Figure 3). HMDA data does not include a field that identifies whether an individual loan application is a subprime loan application. In order to distinguish between the subprime and prime loans, we use the subprime lenders list as compiled by the U.S. Department of Housing and Urban Development (HUD) each year. HUD has annually identified a list of lenders who specialize in either subprime or manufactured home lending since 1993. HUD uses a number of HMDA indicators, such as origination rates, share of refinance loans, and proportion of loans sold to government-sponsored housing enterprises, to identify potential subprime lenders. Since 2004, lenders are required to identify loans for manufactured housing and loans in which the annual percentage rate (APR) on the loan exceeds the rate on the Treasury security of comparable maturity by at least three (five, for second-lien loans) percentage points and report this information to the HMDA. The rate spread can be used as an alternative indicator (to the HUD list) to classify subprime loans. For the years with available 5 Any depository institution with a home office or branch in an MSA must report HMDA data if it has made a home purchase loan on a one-to-four unit dwelling or has refinanced a home purchase loan and if it has assets above an annually adjusted threshold. Any nondepository institution with at least ten percent of its loan portfolio composed of home purchase loans must also report HMDA data if it has assets exceeding $10 million. Under these criteria, small lenders and lenders with offices only in non-metropolitan areas are exempt from HMDA data reporting requirements. Therefore, information for rural areas tend to be incomplete. Yet, U.S. Census figures show that about 83 percent of the population lived in metropolitan areas over our sample period, and hence, the bulk of residential mortgage lending activity is likely to be reported under the HMDA. 8 data, the ranking of subprime lenders using the rate spread variable alone coincides closely with the ranking in the HUD list.6 The HUD list of subprime lenders is also preferable to the rate spread information for a number of reasons. First, rate spreads are not available prior to 2004. Second, subprime loans do not necessarily have APRs that are three (or five) percentage points above a comparable Treasury rate but may reflect fees and yield spread premiums or other borrower characteristics determined by the lender. Third, and most importantly, the rate spread in HMDA is available only for originated loans, making it impossible to calculate denial rates for prime and subprime applications separately. We remove some observations with missing HMDA data from the sample and also focus on the subset of loans that are either approved or denied. First, we drop applications with loan amounts smaller than $1,000 because loan values are expressed in units of thousands of dollars and rounded up to the nearest number. Second, applicant income is left- censored at a value of $10,000. We therefore eliminate applicants with missing applicant income or applicant income of exactly $10,000. Third, we drop loans for multi-family purpose from the sample, as this is a distinct market from the overall mortgage market for single family homes. Fourth, we drop federally insured loans and refinancing loans as their risk profile is likely to differ from that of other loans. Finally, and importantly, we eliminate all application records that did not end in one of the following three actions: (i) loan originated, (ii) application approved but loan not originated, or (iii) application denied. Other actions represent dubious statuses (e.g. application withdrawn by applicant) or loans purchased by other financial institutions. Including purchased loans would amount to double- 6 The correlation is around 0.8. 9 counting as these loans are reported both by the originating institution and the purchasing institution. We supplement the HMDA information with MSA-level data on economic and social indicators published by federal agencies, including annual data on macroeconomic variables, such as personal income, labor and capital remuneration, self-employment, and population from the Bureau of Economic Analysis (BEA), data on unemployment and inflation (consumer price index) from the Bureau of Labor Statistics (BLS), data on total population from the Census Bureau, and data on house price appreciation in a given MSA (based on a quarterly housing price index) from the Office of Federal Housing Enterprise Oversight (OFHEO). We also obtain data on “seriously delinquent” subprime loans, defined as subprime loans with 60 or more days delay in payment, from LoanPerformance, a private data company. Data on these delinquency rates are available only for 2004 onwards. Over the last decade, subprime mortgage lending has expanded rapidly both in terms of the number of loans originated and the average loan amount. Subprime mortgage originations almost tripled since 2000, reaching $600 billion in 2006. Against an also fast growing market for prime mortgages, this boom brought the share of subprime lending from 9 percent in 2000 to 20 percent of all mortgage originations in 2006. Average loan amount also grew reaching $132,784 in 2006 or 90 percent of the prime mortgage average amount. In absolute terms, the subprime market reached a size of about $1.3 trillion in 2006. A first look at our data suggests that rapid growth in subprime loan volume was associated with a decrease in denial rates on subprime loan applications and an increase in the loan-to-income ratio on the loans originated by subprime lenders (Figure 4). These casual observations lend some support to the view that lending decisions are influenced by market