|
Article Excerpt
In this article we make two contributions toward a better understanding of the causes and consequences of discrimination in credit markets. First, we develop an explicit theoretical model of loan underwriting in which lenders use a simple Bayesian updating process to evaluate applicant creditworthiness. Using a signal correlated with an applicant's true creditworthiness and their prior beliefs about the distribution of credit risk in the applicant pool, lenders are able to evaluate an applicant's expected or "inferred" creditworthiness to determine which loans to approve and which to deny. Second, we explicitly model the self-selection behavior of individuals. Because these decisions shape lenders' prior beliefs about the distribution of credit risk, they also affect the Bayesian posterior from which lenders compute an applicant's inferred creditworthiness, implying that statistical discrimination can arise endogenously. As an example, we show that in a market in which only some lenders have Beckerian tastes for discrimination there are conditions under which lenders without racial animus will also discriminate. Our model's flexibility makes it ideal for analyzing a wide variety of empirical and policy questions.
**********
Ever since the initial release of the so-called "Boston Fed Study" on mortgage discrimination (Munnell et al. 1992), academics, bankers, activists and policymakers have struggled to agree on how best to rectify discriminatory practices in consumer credit markets. At the same time, no clear consensus has been reached on whether or not lenders actually do discriminate. At the heart of this conundrum is the difficulty in establishing what discrimination looks like and how it might be detected.
In the intervening years, virtually all research on lending discrimination has been empirical in nature, much of it focusing on the validity of the results of Munnell et al. (1992) and on finding ways to analyze and detect discrimination in mortgage lending data. (1) In contrast, there has been little theoretical work to help explain discriminatory behavior in credit markets or provide a framework for studying the loan underwriting process. (2) This lack of economic theory has not only forestalled policy debates, but has also hindered the design of appropriate empirical tests for lending discrimination.
This article makes two contributions toward a better understanding of the causes and consequences of discrimination in credit markets, yielding interesting insights into the behavior of both lenders and applicants. First, we develop an explicit theoretical model of the loan underwriting process that accounts for lenders' efforts to ascertain applicant creditworthiness in the presence of imperfect information. In our model, lenders use a simple Bayesian updating process to underwrite loans, using both the information provided on the loan application and the lender's own prior underwriting experiences to determine which loans to approve and which to deny. This structural model allows us to define discrimination with respect to observable variables, making it more useful to both empiricists and policy makers. It can also be used to design empirical tests to uncover any discrimination that may exist, as well as to reveal the underlying motivation that gave rise to this discrimination.
The second contribution lies in our focus on the self-selection behavior of individual applicants. We illustrate how the choices that individuals make regarding whether or not to apply for loans--and to which lenders they apply--can lead to a correlation between the creditworthiness of a lender's applicant pool and race. Profit- and utility-maximizing lenders have an incentive to use the information this correlation reveals in order to more accurately assess credit risk. In other words, applicant self-selection behavior can lead to endogenous differences in the average creditworthiness of different racial groups at any given lender. As a result, statistical discrimination can arise even if the underwriting variables considered by the lender are a sufficient statistic of creditworthiness and are unbiased with respect to race (i.e., there is no omitted variable problem).
The predictions that arise from our model are consistent with many existing empirical studies on discrimination in credit markets. In addition, our model clarifies the importance of the source of discrimination for testing and policy. It is important to keep in mind, however, that the primary contribution of this article lies not in specific empirical predictions, but more generally in the structure it provides for thinking about how market frictions may affect the underwriting process. Our framework for analysis provides a foundation for researchers and policy makers to better understand the true causes of observed racial differences in credit markets and to analyze how various policy corrections may affect the market.
Becker (1971) pioneered the economic analysis of discrimination, developing a theory based on the preference a bigot has for one group over another. (3) As Becker argues, a "taste for discrimination" makes a bigot willing to expend a cost (forego a benefit) to associate with a preferred group. As applied to credit markets, a bigoted lender will hold applicants from its preferred group to a lower credit standard than applicants from another group, causing the lender to make loans to high-risk applicants from the preferred group, while denying equally risky applicants from other groups. (4) Alternatively, bigoted lenders may charge lower interest rates or fees to members of the preferred group.
Becker's theory is one of preference-based discrimination. In contrast, statistical or belief-based discrimination arises when the characteristics of an individual's group are used to evaluate his or her personal characteristics. (5) Thus, if a lender believes that minorities are less creditworthy on average than whites, it may wish to apply a more stringent underwriting standard to minorities. This occurs not because the lender dislikes minorities, but rather because it believes them to be less profitable on average.
Arrow (1972a, 1972b, 1973) and, separately, Phelps (1972) were among the first to consider models of statistical discrimination, focusing on the problem of an employer with exogenously given beliefs that the average productivity of white labor is higher than that of minority labor. Calem and Stutzer (1995) model a similar problem in the context of credit markets using a Rothschild-Stiglitz (1976) framework. More recent research has suggested that cultural differences between lenders and borrowers (Calomiris, Kahn and Longhofer 1994, Longhofer 1996) or between employers and job applicants (Cornell and Welch 1996) can also lead to a form of statistical discrimination. Ferguson and Peters (1997) extend this idea, arguing that affinities may arise endogenously based on a lender's experience in working with applicants from different groups.
In practice, lender incentives to statistically discriminate could arise for any number of reasons, ranging from real exogenous differences in group creditworthiness, to cultural affinities, to (perhaps mistaken) lender beliefs about the creditworthiness of different groups. In this article, this incentive is driven by applicant self-selection behavior, which creates real differences in the distribution of credit risk in lender applicant pools.
It is important to emphasize that the statistical discriminator's objective is to maximize profit, and its incentive to discriminate arises even if it harbors no racial animus. If, for example, minority applicants are less creditworthy even after controlling for observable factors, it may be profitable for lenders to hold minorities to a more stringent underwriting standard. Nevertheless, this behavior is illegal under current fair-lending legislation. As we will argue, although understanding the source of discriminatory incentives is important in understanding how to detect and eliminate this behavior, it does not change the social prohibition against it.
Recent empirical work has highlighted the importance of the applicant self-selection behavior that we model. For example, Bostic (2003) finds that marginally qualified individuals tend to seek out banks that are owned by members of their same racial group. Bostic and Canner (1998) find that differences in mortgage applicant pools across black-owned, Asian-owned and white-owned peer banks account for most of their denial-rate disparities. Rosenblatt (1997) finds strong evidence that an individual's choice between conventional and FHA mortgage products (as well as whether to apply at all) is based on how well his personal characteristics match the requirements of each. Similarly, Avery, Beeson and Sinderman (1994) show that cross-lender variation in minority and low-income originations primarily reflects differences in home mortgage application rates.
This evidence clearly demonstrates that individuals can and do choose lenders based on many factors, including convenience, past experience, and their expectations of how likely they are to be approved. Whenever these choices differ systematically across racial groups or other protected classes, they can alter the distribution of credit risk in lender applicant pools and give rise to statistical discrimination as well as other adverse market outcomes.
In our article, this self-selection behavior is driven by the presence of a subset of lenders that have "tastes for discrimination" (Becker 1971). In response to this discrimination, minority applicants attempt to avoid "bigoted" lenders, thereby altering the relative distributions of minority and white credit risk in applicant pools for all lenders. This in turn gives non-bigoted lenders an incentive to statistically discriminate against minorities. Nevertheless, it is important to keep in mind that bigotry avoidance is only one among many factors that might result in applicant self-selection behavior, any of which could give rise to statistical discrimination.
In the next section, we introduce our model of loan underwriting and show how lenders use their past underwriting experiences and beliefs to evaluate applicant creditworthiness. In the third section, we formally define discrimination and show how different motives can give rise to this behavior. In the fourth section, we allow minorities to self-select among lenders and demonstrate how this behavior can affect lender underwriting standards and other market outcomes. Because both individual application decisions and lender lending decisions are endogenous in our model, the model provides a rich environment for analyzing a variety of empirical and policy questions; we discuss these in the fifth section. The final section concludes, while proofs of all results and propositions are found in the Appendix.
A Model of Loan Underwriting
Consider a world in which individuals want to purchase a house or another consumer good but lack sufficient funds to do so. As a result, they must obtain loans from a lender. Each individual in the population is assumed to have a true creditworthiness represented by [theta] [member of] [0, 1]. We interpret [theta] as an individual's likelihood of repaying his loan (although this interpretation is not required for our analysis), and we assume that it captures all of the factors that might cause him to default, including disruptions to his income, changes in the value of the asset financed and his personal compunction about defaulting on an obligation.
Creditworthiness is assumed to be distributed throughout the population according to the probability density function f([theta]), with cumulative distribution function F([theta]) = [[integral].sub.0.sup.[theta]] f(t)dt; all lenders and individuals share these prior beliefs about the distribution of true creditworthiness in the population.
Although each individual's creditworthiness, [theta], is given exogenously, an individual's application decision is endogenous. Therefore, we must distinguish between the distribution of creditworthiness...
|
