Report to the Congress on Credit Scoring and Its Effects on the Availability and Affordability of Credit
Submitted to the Congress pursuant to section 215 of the Fair and Accurate Credit Transactions Act of 2003
August 2007
Print (1.84 MB PDF)
The Effects of Credit Scoring on the Availability and Affordability of Credit
Assessing the effects of credit scoring on the availability and affordability of credit is difficult. As noted, the Federal Register notice seeking public comment on this topic and the various meetings jointly sponsored by the FTC and the Federal Reserve revealed relatively little specific evidence. Such a response was not surprising. Creditors long ago incorporated credit scoring into their systems for underwriting, account maintenance, and marketing. So the question of the effects of scoring, to a large extent, involves gathering information about experiences that may have been decades in the past, a task all the more difficult because credit scores were often implemented in conjunction with the use of automated credit-underwriting systems. Adding to the complexity are changes in the availability and affordability of credit that were contemporaneous with the advent of credit scoring but unrelated to it. Three of the most prominent of these broader changes were technological advances, interest rate deregulation, and a relaxation of rules limiting the geographic reach of banking institutions.
First, the second half of the twentieth century was marked by tremendous technological advances that sharply reduced the costs of data processing and telecommunications and provided opportunities for creditors to expand access to credit and to reduce prices. These advances affected all aspects of the lending business and, even in the absence of credit scoring, likely would have increased the availability of credit.
Second, financial deregulation has also affected credit availability.51 For example, until the late 1970s, state usury laws established limits on the interest rates credit card issuers could charge on outstanding balances, which limited issuers' ability to price for credit risk. Beginning in the late 1970s, court decisions and legislation by some states relaxed restrictions on credit card rates, which in turn allowed national banks to charge market-determined rates throughout the country. The ability to more accurately price for credit risk encouraged lenders to offer credit to higher-risk individuals, who previously went without credit or obtained it from sources outside of the mainstream financial markets. In competitive markets, the ability to price customers according to the risks they pose also works to reduce cross-subsidization; that is, risk pricing reduces the need to charge lower-risk customers higher rates than necessary to help pay for losses to higher-risk customers who weren't paying an appropriate price. Reducing prices for the lowest-risk borrowers may encourage further use of credit.
Third, the easing of certain federal restrictions on the geographic scope of banking institutions, primarily during the 1980s, encouraged competition in credit markets and thus likely further broadened access to credit. Relaxation of limits on the ability of banks to purchase other institutions and to establish branch offices both within and across state boundaries may have further promoted competition.
Concurrent with these changes in the lending environment were changes in the structure of the credit-reporting industry. In the 1970s and earlier, a creditor wanting to assemble an electronic file of the credit histories of a nationally representative sample of individuals, to use either in model development or for marketing purposes, would have had to obtain credit records from many local credit-reporting agencies and integrate the information from each to obtain a relatively comprehensive credit history on these individuals. If a creditor wanted to develop a credit history scoring model, it would have had to assemble an initial set of data on the credit histories of a group of individuals and then repeat the process later to gather information on how these individuals had performed on their accounts.
By the late 1980s, such tasks were both much simpler and much less expensive. A creditor could approach any one of the national credit-reporting agencies to gather the needed information, including historical files that eliminated the need for data requests at two distinct points in time. If a creditor was willing to rely on a generic credit history score, it simply purchased such a score from the credit-reporting agencies. The availability of inexpensive generic credit history scores for most individuals encouraged competition by allowing creditors to solicit the business of individuals for whom they had no previous lending experience.
The confluence of technological advances and the easing of regulatory restrictions obscure the effects of credit scoring on the availability and affordability of consumer credit in general as well as on specific credit products. The past three or four decades have seen substantial changes in how consumers use credit, including an expansion in the practice of substituting one form of credit for another. For example, revolving credit, particularly credit card debt, has substituted for small installment loans because of its ease of use and availability. Similarly, home mortgage debt has substituted for all types of consumer credit through equity extraction done most often through cash-out refinancings or home equity loans.52 These substitutions are attributable to relative price changes among credit instruments, appreciation in home values (allowing more equity extraction), and economies in offering different credit services. Credit scoring likely has contributed to changing uses among credit instruments, but differentiating its effects is likely impossible.
The three sections that follow provide more discussion of the ways in which credit scoring has affected the availability and affordability of credit. The first section is a theoretical discussion of how credit scoring as a technological advance would be expected to affect access to credit. The second is a review of previous research or other evidence on the actual effects of credit scoring on access to credit. The third is an analysis of data from surveys of consumer use of credit that provides indirect evidence on the question of how credit scoring may have affected access to credit.
Expected Effects of Credit Scoring
In considering how credit scoring may have affected access to credit, it is useful to view credit scoring as a technological innovation in credit underwriting and ask, What effect would one expect such a technological innovation to have had on access to credit?
Effects of Credit Scoring as a Technological Innovation
Viewed as a technological innovation, credit scoring raises the efficiency of the credit underwriting system. The efficiency can be expressed in two dimensions--cost and accuracy; that is, greater efficiency can lower the cost of underwriting, or increase its accuracy, or to some extent both, depending on the way lenders respond to the gain in efficiency. If lenders use all the efficiency gain to reduce costs, then the underwriting system may not be more accurate and could be less so. If lenders use all the efficiency gain to improve accuracy, then the costs of the underwriting system may not go down and could even rise.
Changes in costs or accuracy have distinct effects on consumer access to credit, and these effects can be opposite in direction. Regarding a change in cost, the effects on access to credit will almost always be in a predictable direction. Regardless of competitive conditions, if costs are reduced, one would expect that at least some of the reduction in costs would be passed through to consumers in lower rates or fees. Lower interest rates and fees would be expected to increase access to credit, both by attracting more borrowers and by encouraging borrowers to use more credit. If costs rise (perhaps as lenders go beyond the efficiency gain to improve accuracy even more), then credit becomes more expensive, and the effect on access would be negative.
In contrast, regarding a change in accuracy, the effects on access to credit are ambiguous--knowing the direction of change in accuracy is not sufficient to determine whether access to credit will expand or contract. For example, given an increase in accuracy, access will increase (or decrease) if the number of borrowers who previously would have been denied credit but now qualify is larger (or smaller) than the number who previously would have been granted credit but now do not qualify. A similar logic applies given a decrease in accuracy (such a decrease could arise if lenders go beyond the efficiency gain to reduce costs to the point at which accuracy declines).
Effects of Facilitating Product or Service Acquisition and Credit Shopping
An advantage of credit scoring is that it allows a quicker decision than manual, or judgmental, underwriting. Increased speed benefits consumers. First, faster credit decisions allow consumers to purchase, and thus benefit from, products or services more quickly. Second, faster decisions more quickly give consumers the feedback they need for credit shopping. Receiving such feedback informs consumers about their circumstances; the more quickly they get it, the more efficient will be their credit shopping and decisionmaking. Increasing the efficiency of credit shopping may increase the competitiveness of loan markets.
Another way that credit scoring may increase the efficiency of credit shopping is by reducing lenders' costs of prescreening potential borrowers (through targeted solicitations). The lower costs encourage creditors to conduct more prescreening, which benefits consumers by giving them more information about alternatives.
Promoting Consistency and Discouraging Discrimination
One feature of credit scoring generally not shared by judgmental underwriting is its objectivity and consistency; judgmental systems are by their nature subjective and may not produce consistent decisions between applicants with substantially similar credit histories. Credit scoring applies an algorithm to standardized credit information, so a given set of such information produces a given credit score no matter when it is prepared or for which borrower it is prepared. In judgmental underwriting, on the other hand, multiple analysts evaluate credit history in different ways, often emphasizing different factors; thus, the same inputs do not always lead to the same interpretation. For a given level of accuracy, improved consistency can lower costs by reducing costly management oversight that is necessary to ensure that different loan underwriters are applying a firm's lending rules in a manner consistent with company policy and applicable legal requirements. In competitive markets, such cost savings would be expected to be passed on to consumers in the form of reduced loan interest rates or fees.
Some observers argue that consistency is not always unambiguously beneficial because it may involve inaccuracy. Credit scoring relies on a database of historical performance to predict future performance. Statistical models will tend to predict well when evaluating individuals whose financial profiles are similar to those included in the historical files used to develop the models. However, statistical models may not work as well in predicting performance for individuals whose profiles are substantially different from those in the estimating database. Judgmental credit evaluation may work better for these individuals. This issue is less likely to be present in credit-scoring models estimated over large populations with diverse experiences with credit that can be used to separately model (for example, by using different scorecards) the behavior of relatively small subpopulations.
Adoption of a mechanical, consistent system for credit evaluation reduces the opportunities for engaging in illegal discriminatory behavior. In contrast, judgmental, subjective decisionmaking offers opportunities for discriminatory behavior, whether such behavior is intentional or not. For example, in a judgmental system, a credit rater may assign different credit ratings to two borrowers who pose identical credit risks if one is, say, a friend or member of the rater's social club, or a credit rater may assign different evaluations to prospective borrowers with identical credit histories on the basis of impermissible extraneous data such as the borrower's ethnicity, religion, national origin, or sex. Such actions are illegal, but in a judgmental underwriting system they are easier to disguise if deliberate, and they slip through more easily if unconscious.53
A rule-based system, if applied consistently, works to deter discrimination unless the rules themselves are discriminatory. Credit-scoring systems explicitly avoid making use of impermissible data, a fact that can be readily verified. Moreover, as noted previously, the records maintained by credit-reporting agencies on the credit experiences of individuals do not include information on personal characteristics such as race, ethnicity, sex, and marital status. However, other factors included in a credit-scoring model may raise discrimination concerns if they are correlated with impermissible data and are assigned an inappropriate weight (a topic addressed in a later section of the report).
Effect of Increased Transparency
Credit scoring can enhance the transparency of lending activities and the credit risks they involve, particularly if the score is estimated independently of the lender and intended for general use. Loans that carry a standardized and accurate metric of risk, such as a credit score, are more "transparent"--that is, because of that score, the risk posed by the loans can be more readily seen by all who would make decisions on the basis of the risk. Such decisionmakers include prospective purchasers of individual loans or loan portfolios, regulators, and credit-rating agencies evaluating the credit risks of a pool of loans or the financial condition of a creditor.
By reducing uncertainty about the credit risks inherent in a portfolio of loans, increased transparency can lower the costs of funding, either by reducing the amount of capital a firm must maintain or by facilitating funding through loan securitization. In a competitive market, cost savings are likely both to broaden opportunities for creditors and to lower prices for consumers.
Credit Scoring with Closed-End and Open-End Credit
As a technological innovation, credit scoring improves the efficiency of underwriting for credit applications (whether for closed-end loans such as home mortgages or automobile loans or for open-end credit such as revolving credit card accounts) and for the ongoing monitoring of existing borrowers using open-end credit.
As noted in the preceding section, the improved efficiency can increase accuracy, or reduce costs, or both. If lenders choose to reduce costs, then borrowers are likely to benefit from the cost savings to the extent they are passed along. If lenders choose to increase accuracy, then credit scoring will have made the system fairer--that is, fewer creditworthy applicants will be rejected, and fewer noncreditworthy applicants will be accepted.
Moreover, the greater accuracy offered by credit scoring can help ameliorate the problem of "adverse selection" that arises when lenders offer a single interest rate to potential borrowers with varying credit risks.54 It can also ameliorate the problem of cross-subsidization of borrowers that arises when lenders use an inaccurate risk-based pricing system. If credit scoring permits the introduction of a more accurate risk-based pricing system, so more borrowers will be charged prices that more closely reflect the credit risks they pose, the result is a system that is more fair and efficient.
The introduction of credit scoring in the ongoing management of open-end accounts could result in benefits far greater than those realized at the underwriting stage. In the absence of the transparency offered by the credit-scoring system, the performance of current borrowers is information that only the lender, and not any of the lender's competitors, is likely to know. With such "asymmetric" information about current borrowers, a competitor may be reluctant to solicit the customers of another lender for fear of what is often termed the "winner's curse": The lender will compete to keep its lower-risk customers and let the soliciting institution--the "winner"--take on the bad risks. If customers are not solicited, the resulting lack of competition would allow lenders to charge higher rates to their current customers than would be appropriate given the risks they pose.
To the extent credit scoring allows creditors to accurately and inexpensively assess the creditworthiness of all open-end credit customers, it can increase competition and produce customer pricing that is better aligned with credit risk. The result is access to credit at a more appropriate price and a fairer and more efficient credit system.
Evidence on the Effects of Credit Scoring
The previous section described the potential ways that credit scoring could have affected access to credit as it became fully integrated into the credit system. Some of the expectations drawn from theory are clear-cut, and others are ambiguous. For example, theory suggests that credit scoring should cause creditors to reduce costs for a given level of accuracy or improve accuracy for a given level of costs. However, theory does not tell us at what point creditors will strike a balance between these two approaches. For example, with new technology a lender could take all of the gains in cost savings and tolerate a decrease in accuracy. Theory is also ambiguous on whether credit scoring would increase or decrease the number or size of loans. On all of these points, the actual outcomes could differ from product to product and lender to lender.
Theory tells us what the potential benefit would be if the ability to use credit scoring enables risk-based pricing. However, theory does not tell us if all the conditions necessary to adopt risk-based pricing will be met. The ability to accurately rank-order credit risk may be only one component of a lender's decision to offer loans with prices that are tied to risk. Thus, the answer to the question of what the adoption of credit-scoring has done to the availability of credit, and to the more basic question of the degree to which credit scoring is more accurate or less costly than judgmental underwriting, remains largely empirical. However, firms that have analyzed these questions have generally considered their results proprietary; thus, the public domain contains relatively little specific evidence to help answer the questions, perhaps because academics and others interested in the topic may not have been able to gain access to needed data. Nevertheless, some limited evidence was provided in the public comments received for this study, and other evidence is available in the literature.
The Accuracy of Credit Scoring
A number of academic studies have compared the accuracy of credit scoring to that of judgmental credit-evaluation systems. These studies consistently find that credit-scoring systems outperform judgmental systems in predicting loan performance. Chandler and Coffman (1979), for example, review evidence indicating that "empirical models are able to outperform their judgmental counterparts on the average" (emphasis in original). Rosenberg and Gleit (1994) review several studies comparing credit scoring with judgmental credit evaluation and report that "a good scoring system outperforms human experts." Thomas (2000) reports on studies finding that credit scoring in the credit card arena reduced default rates 50 percent relative to the rates under judgmental underwriting. Hand and Henley (1997) find that credit-scoring methods "produce more-accurate classifications than subjective judgmental assessments by human experts" (emphasis in original).55 In a comment submitted for this study, Chandler reported on the experience of a large credit card issuer that performed a controlled experiment designed to compare the effectiveness of judgmental and credit-scoring methods. Relative to judgmental methods, the credit-scoring system approved 15 percent more applicants using the established creditworthiness cutoff used by the card issuer, and, after a two-year performance period, the lender experienced an 11 percent lower default rate.56
Additional evidence on the effectiveness of credit scoring comes from Fair Isaac, which reports that in its experience in working with lenders, a change from judgmental credit evaluations to credit scoring substantially improves decisionmaking. Fair Isaac cites findings from a case study in the credit card arena: By switching from judgmental evaluations to credit scoring, "the issuer would have been able to either double its approval rate without increasing its credit risk, or reduce its credit risk by half without decreasing its approval rate."57 More generally, Fair Isaac estimates that "when a creditor switches from judgmental decisions to credit scoring, it is common to see a 20 percent to 30 percent reduction in credit losses, or a 20 percent to 30 percent increase in the number of applicants accepted with no increase in the loss rate."58
In the home mortgage lending arena, Straka (2000) reports that an internal analysis by Freddie Mac found that credit-scoring evaluation outperformed judgmental evaluations on a pool of loans purchased by Freddie Mac under their Affordable Gold Loan program.59 Straka also reports that an analysis conducted by Freddie Mac found that generic credit history scores "worked as a statistically significant and strong predictor in a home mortgage default equation."60
The studies cited in this section generally compare the performance of a "pure" judgmental credit-evaluation system with a "pure" credit-scoring system in controlled tests involving actual extensions of credit. They do not address how a system combining both judgmental assessment and credit scoring might perform. Nor do they quantify the results of credit scoring in actual operation rather than in controlled tests.
Cost and Time Savings
The public realm provides relatively little quantitative information on the savings in time and cost that accrue because of credit scoring. The available evidence for home mortgage lending indicates that credit scoring has helped reduce the time needed to make credit decisions from several weeks to a matter of a few minutes.61 Regarding cost savings, lenders that integrated automated underwriting systems into their home mortgage loan origination process are estimated to have reduced origination costs by as much as 50 percent, or roughly $1,500.62 Other research found that underwriting expenses fell 27 percent and "back office" costs dropped 15 percent when larger proportions of loans in pools of home mortgages were evaluated with credit-scoring processes.63 Regarding credit card activities, it is estimated that most credit card issuers can make a decision on a credit card application in less than sixty seconds when a real-time credit-scoring system is used, compared with five minutes in the quickest manual underwriting systems.64 To the extent that the savings in cost and time resulting from credit-scoring systems are passed through to consumers, the savings twill lead to lower interest rates and greater access to credit.
Access to Credit
As noted earlier, relatively few studies have directly examined the effects of credit scoring on access to credit. Using evidence from U.S. banks, Jeong (2003), for example, finds that more-accurate credit screening leads to increased lending.65 In home mortgage lending, Gates, Perry, and Zorn (2002) report that home mortgage approval rates were higher when applications were evaluated with Freddie Mac's automated underwriting system than when the same loans were evaluated by manual underwriting techniques.66 Some of the studies bearing principally on accuracy also found a higher number of approved applicants.67
Information on the volume of credit solicitations also suggests that credit scoring has affected access to credit. The number of solicitations for credit cards has increased substantially over the past fifteen years, a period in which generic credit scores became available, and both the proportion of consumers with credit cards and the average number of cards per person have increased. For example, the number of mailed credit card solicitations increased from 1.1 billion in 1990 to 5.2 billion in 2004. Because credit scoring is the primary technology used for prescreened solicitations, these figures provide indirect evidence that credit scoring has expanded access to credit.68
Evidence from the Survey of Consumer Finances on the Effects of Credit Scoring on Access to Credit and on the Use of Credit
The availability of credit scores and their use in lending have grown over the past twenty-five years. Increased use of credit scoring could affect the availability of credit in at least three ways. First, credit scoring provides lenders with information on the creditworthiness of a large number of individuals whose credit risk was previously unknown or was difficult or costly to ascertain because the borrower and prospective lender had no previous credit relationship. As a result, credit scores could allow lenders to identify borrowers who are reasonable credit risks but who were previously underserved, thereby expanding credit access for these borrowers. Second, a shift from lender-specific evaluations of existing customers to those based on a credit score may affect which applicants are approved by offering a different--potentially more accurate-- assessment of individuals' relative creditworthiness. If so, credit availability may increase for some borrowers while declining for others. Finally, to the extent credit scoring reduces the cost of lending or facilitates more effective risk-based pricing of loans, increased use of credit scoring may expand the range of applicants to whom lenders are able to make loans profitably.
It is commonly believed that widespread adoption of credit scoring has, on the whole, contributed to an increase in the availability of credit. Nevertheless, access to credit may not have improved uniformly for all populations. For example, racial or ethnic differences in credit access could narrow if non-Hispanic whites historically have experienced greater access to credit than blacks or Hispanics and if adoption of credit scoring increased access to credit for all individuals but disproportionately benefited minorities. Conversely, if the adoption of credit scoring increased access to credit for all individuals but disproportionately benefited non-Hispanic whites, gaps in credit access could widen. Hence, the consequences of increased use of credit scores for differences in the availability of credit across demographic groups are ambiguous.
Data from the Survey of Consumer Finances (SCF) can be used to assess how differences in credit use across demographic groups have changed over time. The SCF provides the most comprehensive information available on the net worth, assets, and liabilities of U.S. families, including detail on the types and amounts of debt held by families.69 However, like any data with information on only outstanding loans, the SCF data do not directly measure credit availability, that is, the supply of credit; instead, the data on credit use reflect the confluence of both supply and demand factors. Differences in credit use across subpopulations over time measure the effect of credit scoring on differences in access to credit only if the effect of other factors that influence the availability of credit as well as shifts in the demand for credit were comparable across groups. Though this strong assumption almost surely does not hold perfectly, we nonetheless interpret changes in the differences between groups' use of credit as indirect, suggestive evidence regarding the potential effects of credit scoring on differences in access to credit.
The SCF has been conducted every three years since 1983, and the most recent data available are from the 2004 survey.70 Thus, a time-series of families' credit use between 1983 and 2004 can be constructed to contrast trends in credit use by race or ethnicity, income, and age.71 We also examine whether growth in credit scoring increased the use of some types of credit more than others by comparing trends in families' ownership of several types of debt. The analysis compares the prevalence of credit card debt relative to mortgages and other closed-end installment loans, since credit scoring may have had different effects on the use of collateralized and unsecured credit. A further differentiation is made between credit cards that can be used only at a specific retailer ("store or gas cards") and those--such as MasterCard or Visa cards--that may be used more broadly ("bank-type and travel or entertainment cards").
Changes in Credit Use across Populations
Taken as a whole, the estimates from the Survey of Consumer Finances are generally consistent with the conjecture that adoption of generic credit scores contributed to an expansion in credit availability and, in particular, to greater ownership of bank-type or travel and entertainment cards (tables 4-6). The largest change in credit usage over this period was the increase in the prevalence of bank-type or travel and entertainment cards, which rose 25 percentage points or more for each of the racial or ethnic groups. In turn, the fraction of families with credit cards that had only store or gas cards declined, though not as steeply. The prevalence of installment debt also declined for all groups.
Trends in unadjusted differences in credit usage for blacks, Hispanics, and other families relative to non-Hispanic white families differ across types of debt and do not suggest a clear effect of expansions in credit scoring on differences in access to credit for these minority groups (table 4). On the one hand, with the exception of non-education installment debt, the estimates imply that the differences between blacks and non-Hispanic whites for each type of debt narrowed, on net, between 1983 and 2004. On the other hand, the trends in the gap between Hispanic and non-Hispanic whites are mixed: Differences tended to increase for mortgages, installment loans, and bank-type or travel and entertainment card ownership and declined for other measures. Further, the implied changes in the gaps are often modest relative to the fluctuations across surveys and the magnitude of the gaps.
Moreover, interpretation of the unadjusted differences is not straightforward since they potentially reflect not only racial or ethnic differences in debt ownership rates for otherwise similar families but also differences in the distribution of economic and demographic characteristics across the subgroups. The distributions, for instance, of age and income--which are correlated with debt ownership--differ by race or ethnicity and thus contribute to observed differences in credit use. Similarly, trends in the unadjusted differences may be driven in part by differential rates of change in other demographic characteristics. For example, credit use generally rises with income, so faster income growth over time for blacks than for non-Hispanic whites would narrow differences in debt ownership even if the racial difference in ownership rates for families with similar incomes was unchanged.
Accounting for Changing Demographics
To account for differences in the levels and trends in demographic characteristics across racial or ethnic groups, adjusted differences were estimated using multivariate regressions. The first group of regression-adjusted differences provides estimates of the differences in credit use within each year that remain after accounting for differences in other family characteristics. The adjustments are based on logit regressions that model debt ownership as a function of age, income, and marital status and that are estimated over non-Hispanic white families.72 The fitted model is used to estimate counterfactual shares of families with debt that would be observed if the relationship between demographic characteristics and credit use for non-Hispanic whites held for all racial and ethnic groups. The adjusted gap for credit cards for blacks and non-Hispanic whites, for instance, is the average difference between the actual share of black families with cards and the counterfactual percentage predicted from the model estimated over non-Hispanic white families.
Accounting for differences in demographic characteristics typically reduces the estimated level of the gaps between blacks and non-Hispanic whites in each year, with the exception of ownership of only store or gas cards.73 The adjusted differences in the shares with mortgage debt, any credit card, and bank-type or travel and entertainment cards between Hispanics and non-Hispanic whites tend to be smaller than the unadjusted differences, but other gaps widen on average. This pattern of changes likely reflects the fact that blacks are particularly concentrated in the lower portion of the income distribution, whereas Hispanics are especially overrepresented among younger families.74 Because debt ownership tends to rise with income, counterfactual ownership rates for blacks are lower than the overall share of non-Hispanic white families with debt, so adjusted gaps are smaller than the unadjusted differences. Similarly, the adjusted Hispanic-white differences are larger for those types of debt, such as non-education installment loans, that are more common among younger borrowers. In most cases, the trends for both blacks and Hispanics point to slight increases in differences in credit usage relative to non-Hispanic whites. The disparity in ownership of only store or gas cards reversed as the share of minorities with a credit or charge card that owned only a store or gas card became greater than that of non-Hispanic whites, though the proportions are low in recent years for all groups.
The first set of adjusted differences focused on how the typical difference in credit usage attributable to race or ethnicity alone has changed over time as a result of changes in credit markets as well as shifts in the economic and demographic characteristics of families in each racial and ethnic group. An alternative technique controls for demographic shifts by holding the age, income, and marital status of families constant at their 2004 levels. Here, logit models, like those described above, are estimated for each racial or ethnic group and used to predict two counterfactual debt ownership probabilities. To calculate, say, the adjusted difference in mortgage ownership rate for Hispanics in 1983, we contrast the rates predicted by applying the fitted Hispanic and white models in 1983 to Hispanic families in 2004. As with the other adjustment technique, racial or ethnic gaps implied by varying only the relationship between debt ownership and demographic and economic factors are evaluated. In this instance, however, the counterfactuals are estimated using the characteristics of 2004 Hispanic families rather than those of Hispanic families in 1983. By using the 2004 characteristics to predict counterfactual rates in each year, we attempt to control for differences in demographic shifts across groups when examining the evolution of racial and ethnic gaps in credit usage over time.
The estimated rate of increase in the gaps in debt ownership between blacks and non-Hispanic whites is generally slightly higher after holding the distributions of other characteristics fixed, suggesting that differential rates of demographic changes for blacks and non-Hispanic whites over the period served, on net, to narrow such differences in debt ownership. The trends in the counterfactual gaps between Hispanics and non-Hispanic whites tend to be smaller after fixing age, income, and marital status at their 2004 values. The differences both between blacks and non-Hispanic whites and between Hispanics and non-Hispanic whites increased for ownership of non-education installment debt and bank-type or travel and entertainment cards. The predicted fraction of Hispanic families with a credit card rose faster than the share of non-Hispanic whites, decreasing the disparity in this measure. Other changes in the gaps between Hispanics and non-Hispanic whites were smaller or more sensitive to the model used to predict counterfactual ownership rates.
The next portion of the analysis considers how debt ownership rates changed across income groups in the 1983 through 2004 SCF surveys (table 5). The table compares the credit use of families in the top and bottom thirds of the income distribution with the proportion of middle-income families with each type of debt.75 Credit use rises with income, except in the case of the proportion of families that own only store or gas cards. The unadjusted gaps between high- and middle-income families declined for all types of debt but especially the shares with bank-type or travel and entertainment cards and outstanding credit card balances. The differences in credit usage between lower- and middle-income borrowers declined for overall debt, credit card balance, and credit card ownership. The fraction of middle-income families with a mortgage rose notably in 2004, a shift that contributed substantially to a decline in the gap relative to higher-income families and a widening in the gap relative to lower-income families.
As expected, the adjusted differences across income groups narrow after accounting for differences in other demographic characteristics in each survey year.76 The shifts in the levels were roughly comparable across years so that in most cases conclusions regarding trends in gaps are largely unchanged. The counterfactual gaps and trends in credit use for both sets of adjustments shown are also similar to one another. Use of most types of credit rose more steeply among middle-income families than for other families, on average, over the period. As a result, differences between lower- and middle-income families grew, whereas those between middle- and higher-income families narrowed for many measures. An exception to this pattern are the differences in the shares of families with credit cards and credit card balances, which narrowed across income groups. The relatively large increases in prevalence of revolving credit among lower-income borrowers did not carry over to bank-type or travel and entertainment cards, however, for which credit scoring might be expected to have had the largest effect on credit availability.
The final portion of the analysis considers changes in debt ownership rates across age groups (table 6). Trends in credit use within each of the four age ranges mirror those discussed above, with the exception of the increase in installment borrowing among families with a head aged 62 or older. As illustrated by the first columns, the oldest families are the least likely to have debt. Rather than taking one age group as the basis for comparison, the counterfactual estimates are the predicted level of debt ownership for a family with a 48-year-old head but otherwise identical demographic and economic characteristics.77 The adjusted differences, presented in the second and third columns, indicate that, in most cases, credit use rose more quickly for the oldest group than would have been predicted based on a similar 48-year-old, while increases for other age groups were more similar. As shown in the leftmost columns, the percentage of families with a head younger than age 35 that carried a credit card balance increased at least as steeply as the shares for the next two age groups. Accounting for differences in other characteristics, however, the fraction with a balance did not rise as quickly as would have been predicted for a similar 48-year-old. In contrast, the fractions of both the youngest and oldest families that owned a bank-type or travel and entertainment card rose comparatively quickly. Looking across types of debt, gains for the oldest set of families were generally at least as large as those for the youngest group. To the extent that lower rates of debt among retirement-age families reflect comparatively low demand for credit, the narrowing of differences in credit usage among older families suggests that shifts in demand play an important role in the observed trends over time.
Taken together, the foregoing analyses of differences in credit use by race or ethnicity, income, and age suggest only tentative conclusions. Importantly, the data provide very little evidence that the expansion in credit scoring disproportionately benefited population subgroups that historically had low rates of debt ownership. Instead, trends in gaps relative to other groups with greater credit use appear in many instances to have changed only slightly or to have widened, particularly after attempting to adjust for differences in the level and trends in key demographic variables across groups. Year-to-year fluctuations in estimates and variation across groups likewise prevent conclusive inference.
Limitations of the data and the approach also suggest that the results should be interpreted with caution for several reasons. First, though the SCF data provide a lengthy time series on U.S. families' use of a variety of types of debt, as noted earlier, the data measure credit use rather than access to credit. Many other factors that changed over the 1983-2004 period could have influenced the use of credit by various demographic groups. Differing trends in families' demand for credit, for example, could also have resulted in changes over time in the observed gaps in credit use across groups. Second, since the use of credit scoring began to grow in the late 1970s, the earliest effects of credit scoring precede the 1983 SCF, the first with data on debt use comparable with that gathered in later surveys. Third, regression adjustments like those in this analysis are commonly used to examine differences in outcomes across groups, but other work has often found that estimates of counterfactual gaps may be sensitive to the regression specification, including the set of demographic characteristics incorporated in the model. The need to estimate regressions over each subgroup and the available sample size limits the complexity of the models that can be estimated. The results of this analysis are generally robust to small changes in the model, but estimates based on other reasonable specifications may differ more substantively. Finally, the choice of base and comparison groups can affect the magnitude of estimated counterfactual gaps.