You are on page 1of 60

GIVE CREDIT WHERE

CREDIT IS DUE

I N C R E A S I N G AC C E S S TO A F F O R DA B L E
MAINSTREAM CREDIT
U S I N G A LT E R N A T I V E D A T A

Political and Economic Research Council
The Brookings Institution Urban Markets Initiative

CONTENTS

Executive Summary ................................................................................................................................................2

Overview.................................................................................................................................................................5

I. Introduction........................................................................................................................................................5
Excluded from the Miracle .............................................................................................................................6
The Information Cycle ...................................................................................................................................8
Nontraditional Data Can Bridge the Information Gap....................................................................................9
The Critical Role of Credit Files ..................................................................................................................10

II. Methods ..........................................................................................................................................................11
Objectives .....................................................................................................................................................11
The Data for the Simulations........................................................................................................................12
Approach ......................................................................................................................................................13
Limitations....................................................................................................................................................14

III. Impact on Consumers’ Credit Profiles............................................................................................................16

IV. Observed Differences in Access to Credit .......................................................................................................22

V. Impact on Scoring Models ...............................................................................................................................24
Impact on Predictive Power .........................................................................................................................24
Mortgage Screening Model...........................................................................................................................27
Additional Results on the Predictive Power of Alternative Data ....................................................................28
Impact on Delinquency and Acceptance Rates .............................................................................................30

VI. Demographic Impacts.....................................................................................................................................32

VII. Summary and Policy Implications .................................................................................................................36
Encouraging Alternative Data Reporting.......................................................................................................36
Preserve Voluntary Reporting........................................................................................................................37
Reporting Enhances Community Development Efforts.................................................................................38

VIII. Future Research Directions .........................................................................................................................40

Appendix A. Sample Characteristics .....................................................................................................................42

Appendix B. Detailed Model Results ....................................................................................................................46

Endnotes ..............................................................................................................................................................53

About the Authors ................................................................................................................................................56

Acknowledgments ........................................................................................................................Inside Back Cover

GIVE CREDIT WHERE
CREDIT IS DUE:

I N C R E A S I N G AC C E S S TO A F F O R DA B L E M A I N S T R E A M
C R E D I T U S I N G A LT E R N A T I V E D A T A

POLITICAL AND ECONOMIC RESEARCH COUNCIL & THE BROOKINGS INSTITUTION URBAN MARKETS INITIATIVE © 2006

credit files—is similar to that of the general population sumers include populations with thin credit files (as measured by credit score distribution). The risk profile of the thin- We examined a sample of approximately 8 million file/unscoreable population—after energy utility and TransUnion credit files with a strong focus on con. given a 3 percent target mation. to evaluate the risk profile of a potential Key findings include: borrower. if widely incorporated into credit reporting. and other recurring obligations. as well as • Nontraditional data make extending credit easier. with effective interest rates as high as 500 percent.1 Our findings indicate that alternative data. these millions outside the credit mainstream are The evidence suggests that most individuals in this poorer. can bridge • Those outside the credit mainstream have similar the information gap on financial risk for millions of risk profiles as those in the mainstream when Americans. or loans for education) and thereby improving their lives. the information segment are not at high risk in terms of lending. mainstream lenders have too little information on them to evaluate risk and thereby extend credit. of these models were then compared with Mainstream lenders can use “alternative” or “nontradi. by more than 20 percent among previously unscore- able populations. and including telecommunications data increases the tional utility and telecommunications payment infor. For a variety of reasons. electric. payment/bankruptcy outcomes observed during the fol- tional” data. considering that many of including nontraditional data in credit assessments. The scores. As a result. gas. which contained alternative or nontradi. insurance. The credit reports increases the acceptance rate by 10 percent. Using can direct markets toward a faster alleviation of nontraditional data lowered the rate of serious default poverty in this country. telecommunications data sets are included in their sumers outside of the credit mainstream. The con. This study offers a feasible market solution to bring make a variety of credit decisions. were applied to models used by lenders to default rate. acceptance rate by 9 percent. report files.EXECUTIVE SUMMARY Despite the vast accomplishments of the American credit system. dictions. rent. More concretely. The lack of reliable credit places them at a great disadvantage in building assets (such as homes. (fewer than three sources of payment information. small businesses. those in most need of credit often turn to check cashing services and payday loan providers. Including energy utility data in all consumer credit mined owing to insufficient information. including payment obligations such as lowing year. less advantaged Americans. approximately 35 million to 54 million Americans remain outside the credit system. or trade lines) on payment timeliness. “unscoreable” segments whose risk cannot be deter. 2 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . or pre- those outside the mainstream credit fold within it.

given a 60 percent target acceptance rate. remarkable is that two-thirds of both the thin-file For example. 14 percent for Asians. we earning between $20. models.6 percent of thin-file borrowers with only traditional data in their credit reports. of the analysis sample from being unscoreable to Including fully reported energy utility and telecommu- scoreable.4 percent in the utility sample and tional consumer credit reports measurably improves 11 percent in the telecommunications sample. For instance. and a nearly 10 percent rise for the general increase in their acceptance rate. increase access. Sizable segments would see their credit nications trade lines (i. harming more in the mainstream that helping those now excluded. and those who prefer sample. data. In addition. The rate of increase was pret little or no credit information as the highest risk.3 percent) and the thin-file telecom. adding telecommunications data reduces the default sion of alternative trade lines in conventional credit rate by 27 percent.. 14 percent for cations trade lines are added to credit reports. These reductions allow lenders to reports improves access to mainstream sources of con. Specifically. we still find a modest 2 percent improvement in model performance with the • Nontraditional data decrease credit risk and addition of alternative data. different accounts) in tradi- scores improve—22. and provisioning requirements. The results of simu- lations reported here suggest that little will change for the mainstream population. cent of thin-file borrowers whose credit report Such improvements could have further positive econo- included nontraditional data opened a new credit mywide effects.000 and $29. Preliminary evidence strongly suggests that the inclu.7 percent) become scoreable loans 90 days or more past due) declines 29 percent. In a one-year observation period. excluded from the calculations. gins. Most the performance of loans for a target acceptance rate. in which the Spanish as their primary language saw a 27 percent general sample is used but unscoreable credit files are increase in their acceptance rate. As a result. Similarly. Hispanics saw a 22 percent one set of calculations we assume that creditors inter- increase in acceptance rates. in our study. a lender’s default rate (percentage of outstanding munications sample (67. For example. 21 percent for Blacks. 14 per. by integrating fully reported energy utility utility sample (60. 16 per. account compared with only 4. see a 300 percent rise for a sample of thin-file con- renters (as opposed to homeowners) saw a 13 percent sumers. make more capital available and improves their mar- sumer credit. we see a those aged 66 older.000 or less annually. capital adequacy. One worry is that including nontraditional data will be counterproductive. The addition of the alternative data moves 10 percent • More data can reduce bad loans.999.e. Including alternative data was especially beneficial for This migration greatly affects the performance of members of ethnic communities and other borrower examined scoring models.2 GIVE CREDIT WHERE CREDIT IS DUE 3 . when alternative data are included in their credit files. in subgroups. and 15 percent for those to measure the model performance. when fully reported utility or telecommuni- cent for those aged 25 or younger. 21 percent for those who earn significant rise in the KS statistic—an industry gauge $20. In the most conservative case.• Minorities and the poor benefit more than • More comprehensive data can improve scoring expected from nontraditional data. • Nontraditional data have little effect on the credit mainstream.

Some models appear in a host of credit scoring models. Although using alternative implemented. Whenever few. Information must first be gathered and chances of building assets. without it.In summary. new models optimized for such data data in consumer credit reports affects how the data must be built and old models modified. The steps. What has cations accounts as a financial trade. Simply bringing the information an information gap exists. while changed is the availability of information. there is use of alternative data in consumer (and commercial) no incentive to take them. ■ unscoreable Americans who reside in urban areas and elsewhere. The online will spur many of the steps. are important. The benefits of using nontraditional data will not be lions into the credit mainstream and improve their instantaneous. Public officials can play a credit reports can close an information gap that has positive role by removing barriers to reporting where negatively affected the lives of millions of thin-file and they exist. 4 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . nothing must be altered to not treat utilities and telecommuni- about the data subjects has changed. markets fail to thrive. nontraditional data promise to bring mil.

and the role of information in bringing those outside the credit mainstream into accessible. afford- able credit channels. • Section V examines the impact of utility and telecommunications trades on the predictive power of several scoring models and the implications for both the cost and availability of credit. ■ GIVE CREDIT WHERE CREDIT IS DUE 5 . • Section II describes the objectives. • Section VII summarizes the empirical results and concludes with a discussion of their implica- tions for public policy.OVERVIEW • Section I provides a brief overview of the impact of the U.S. • Section IV compares the number and size of new accounts that were opened by consumers with an existing utility or telecommunications trade (the “analysis” sample) to the number and size of new accounts that were opened by otherwise similar consumers without such trades (the “validation” sample). • Section VI examines the demographic groups that would most likely be affected by a more systematic reporting of utility and telecommunications data. those left behind. Appendix A describes the analysis sample in more detail and assesses the extent of potential biases. data sources. and methodology of the study. focusing on the number of consumers who can be scored and the resulting distribution of credit scores. Appendix B presents the complete results of our model simulations. • Section VIII offers directions for future research. • Section III shows how the addition of utility and telecommunications trades has affected consumers’ credit profiles. credit system.

high credit risk.” The former Federal credit—an estimated 35 million to 54 million Reserve Board Chairman Alan Greenspan said that Americans remain outside of the mainstream national such a system and technologies using it had “a dra. the national credit reporting system touches the lives of millions of Americans each day. 6 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . credit system. The robust and full-file data maintained by consumer reporting agen- cies have contributed to a significant expansion in con- sumer and small-business lending without increasing risk in the national credit system. lack- This success ranges from those applying for a home ing sufficient information for automated underwriting mortgage loan or refinancing an existing mortgage to tools. Thus. I N T RO D U C T I O N E X C L U D E D F RO M T H E M I R AC L E The American credit system is in many ways the envy of the world.on consumers and households and their credit because there is little or no credit information in access to credit in this country at reasonable rates.” their credit files. equate a lack of information with unacceptably those applying for a credit card or a retail store card. This group is excluded from instant matic impact.. The steady development of information-sharing. and easy entry by new com- petitors have extended credit to tens of millions of Americans. As a result.. The national credit reporting system has become the Despite the impressive track record of the national basis for “automated underwriting.” a practice that has credit system under the Fair Credit Reporting Act— become so successful that former Federal Trade record homeownership. mainstream lenders.I . In the years since the financial services industry began using standardized payment information for scoring. fairer lending across all seg- Commission Chairman Tim Muris referred to it as ments of society. a democratization of access to “the miracle of instant credit. automated credit scoring. homeownership rates have grown and credit has become available to those for whom credit was reserved for the elite.

GIVE CREDIT WHERE CREDIT IS DUE 7 .

they must ana- system are trapped in a catch-22 by their lack of a lyze it. The lack of tools credit bureaus have become powerful information stems from a gap in adequate information on which to sources and “translators” of the potential of consumer make credit decisions about these individuals.T H E I N F O R M AT I O N Identifying information gaps. lenders gained information by assessing when denied access to credit? Lenders currently lack the capacity. and credit-worthiness of tens of mil. to make it useful information. and character of bor- the right tools to adequately assess the credit risk. consultants—consumer credit bureaus. those with little credit his. ing. those outside the mainstream credit decision makers begin with raw data. or add value to it. Consumer tory) and “unscoreable” Americans. data are turned into information by external lions of “thin-file” (that is. rowers. credit. or the information cycle. developing solutions to bridge them. collateral. In today’s world of automated credit underwrit- credit capacity. and educating decision makers in new CYCLE ways to better understand underserved credit markets requires a clear understanding of the process of knowl- edge creation. credit markets. credit history: how does one build a credit history Prior to 1970. 8 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .3 Although I n one sense.

water) and telecommunications as the minority communities and younger borrowers. throughout the course of unscoreable populations. exchanges in these two indus- tries involve “credit-like” transactions—that is. ing capabilities provide a customized understanding of remittance payments. making data collection more feasible. Other alternative data sets—such as auto insurance. and.5 In this study. tional data” refer exclusively to utility and telecommu- doing” gap: the gap between a lender’s perception of a nications data. (4) Increased fairness in lending. and telecommunications data in consumer credit are often forced to treat these borrowers as excessively reports can achieve the following results: risky simply for want of better information. and being credit-like. unless otherwise specified. but their near-term prom- automation has enabled a deeper penetration of some ise for the thin-file/unscoreable population is not as markets. credit capacity. and credit-worthiness of the thin-file/unscoreable N O N T R A D I T I O N A L DATA population. concentration. lender’s analytic teams and model. particular individual’s potential and the reality of his or her credit risk. and rental data—did not score the market context to use in turning the information as highly as utility and telecommunications data. CAN BRIDGE THE (2) Increased access to affordable mainstream credit for thin-file/unscoreable population. especially for heating oil. they receive into knowledge on which to act. O ne potential solution to the credit Catch-22 is while the credit effects on “thicker-file” individu- pervasive reporting of nontraditional or alter. The lack of data and infor. as the penetration rates for these services are frequently 90 percent or more. PERC singled out energy utilities (gas. credit capacity. I N F O R M AT I O N G A P (3) Thin-file/unscoreable individuals will derive the greatest benefit from including alternative data. who are aware that this is not the case. most promising data sets to help bring consumer out- liers into the fold. GIVE CREDIT WHERE CREDIT IS DUE 9 . (1) Increased ability of mainstream lenders to ade- quately assess credit risk. and the payments are made in regular installments. it customarily overlooks the thin-file and evident.In credit decisions. They were also likely to yield results for a large segment of the 35 million to 54 million thin-file/unscoreable individuals. Finally. electric. For simplicity’s sake. als will be less evident. and credit-worthiness.4 This study tests the hypothesis that including utility Many lenders. native data in consumer credit reports. this study. Although These sets may have value. The utility and telecommunications industries are relatively concentrated. the terms “alternative data” and “nontradi- mation on these populations can lead to “knowing. These two data sets ranked highest along three metrics—coverage. a good or service is provided in advance of a payment.

Considering that many of these millions are in their ability to access credit and improve their lives.T H E C R I T I C A L RO L E O F A recent report by the Information Policy Institute examined the feasibility of collecting these and other CREDIT FILES types of nontraditional credit data on a widescale basis.” The use of informa.6 Although such payments are not credit obligations in the traditional sense. For example. it charge as effective rates as high as 500 percent. the majority do and thus alter their life chances. many are forced to turn to providers who can help alleviate poverty in this country. some in the mortgage industry now accept a “nontraditional credit report” based on the consumer’s demonstrated performance in meeting such ongoing obligations as rent. business formation. Of the different sources considered. if any. Although importance. What follows is an attempt to measure that Alternative or nontraditional data offer one possible promise. From a those with spotty payment records or with little. utility and telecommunica- established credit. for among other reasons. the information Moreover. those with no credit histories and those with poor credit are often treated similarly. they are generally believed to reflect a con- sumer’s willingness and ability to repay credit-like obligations. engaged. access to education. This payment information there- fore plays a significant role in shaping the social for. solution to the problems posed by no credit histories. especially via automated mod. utilities. promises a market solution to problems of credit access. and telephone bills. just that utility and on several different accounts. is prohibited by law or regulation. in some states. 10 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . less advantaged Americans. are more telecommunications data may be one effective way of likely to be granted credit at more favorable terms than folding in those outside the credit mainstream. the reporting of such data tion in credit decisions. standpoint of practicality. els. rents. not. The net The promise is that new data sources can help tens of effect is that millions of Americans remain outside the millions of Americans take a step toward asset forma- credit mainstream and are consequently handicapped tion. and small utilities from reporting. In general. That is. The financial services industry has long recognized the need to find alternative ways of evaluating the credit- worthiness of thin-file consumers. tions payment data may be the fastest way to extend credit to underserved communities. for the claim can be extended to “and some utility and telecommunications companies cur- thereby shape the ability of individuals to build assets rently report data to credit bureaus. increasing uncertainty about the reaction of regulators inhibits homeownership rates. In fact. poorer. the concentration plays a critical role in determining both the of the data. consumers tive payment information—auto insurance. or trade lines. and who have demonstrated a history of timely payments so forth—are of less value. has extended credit to millions. None of this is to suggest that other types of alterna- tunes of individual Americans. utility and telecommunications trades again appeared to be most I nformation contained in consumers’ credit files promising. unlike with rental information. which is Behind this simple fact is an issue of considerable widely dispersed among diverse landlords. and in many others. Unfortunately. Relatively few data furnishers must be amount and the terms of credit that they receive.

or a combination of the two. • First. the number of consumers who would likely be affected by the reporting of these trades is undoubtedly very large.e. This study aims potential borrowers with thin credit files should be to fill that gap and to provide clear estimates of the reduced. telecommunications trades could affect the distri- ferent ways: bution of credit scores. Although the impact on consumers with credit. Depending on the con- sumer’s payment record and overall credit profile.II. Although precise statistics are difficult to assemble.. we might also expect reductions in the average sample of credit files. using a representative ing. to date. and who thereby can access or negative. Increasing the reporting of utility and telecommunica. the predictive makers can assess what is at stake and chart viable power of scoring models should be improved. we found. It should be noted that there have been previous systematic reporting of utility and telecommunica- attempts to encourage the utilities to report to the tions data should add one or more trade lines to the credit bureaus. as the consumption of these services is nearly universal. “utility” trades include payments for electricity. many traditional scor. industry and policy that can be used to score consumers. 13 percent of credit price of credit. the scale of the impact credit profile of the typical consumer. METHODS OBJECTIVES This report examines the impact that the broader reporting of telecommunications and utility trades could have on consumers’ access to different types of credit. the number of remained one without measurement. the systematic reporting of utility and tions trades could affect consumers in at least two dif. lower mainstream credit.4 percent had measure of this expected reduction. undertake a direct files had no payment histories. Because the mate changes in the performance of portfolios. All of the models used in this study require just one trade to To the extent that this information leads to better lend- produce a score. In doing so. ing models require at least one valid trade. lished credit could be large. gas. the impact on consumers with little or no estab- thiness of thin-file borrowers. and 19. We do not. but rather esti- only one or two payment trade lines. • Second. costs. Nonetheless. which GIVE CREDIT WHERE CREDIT IS DUE 11 . By increasing the number of trade lines impact of reporting. and heating oil. Although the industry has developed several well-established credit histories would likely be mini- alternative approaches for evaluating the credit-wor. however. mal. which courses to assist those who have poor or no access to in turn should lead to higher acceptance rates. In our analysis. Yet. land lines) and mobile phones. while “telecommunications” trades refer to tradi- tional telephone service (i. it would increase the number of consumers the impact on an individual’s score could be positive who can be scored.

127 6.5 12 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . “Fully reported” trade lines include information on the It is therefore reasonable to expect that lenders would timely payment of bills as well as any derogatories extend credit more deeplyBR1 than the estimates gener. all else equal.519.501 18.206 5.020 100. As shown in the chart.025 0. a lender deciding • An analysis sample of approximately 8.) Although most utility and telecommunications those with thin credit files.826 92.5 3 608. given risk level if they know that risk level with greater certainty.6 1. or fuel) or telecommunications trade (wire- data in addition to traditional credit data—will be less or land line) as of March 31. another benefit TransUnion from the detailed credit reports of of including alternative data in consumer credit reports two mutually exclusive samples of consumers: is that the uncertainty associated with a given credit score should decline. A lender may be more companies routinely report collections.1 shows the number and distribution of con- sumers in the analysis file by the number of utility and telecommunications trades. most of the records in the analysis file have a utility as opposed to a telecommunications trade.5 545. This may be particularly true for cies.3 Total 7. adding predictive information to a credit scoring model reduces the uncertainty of credit scores. because the additional data reduces uncertainty about • A validation sample of approximately 4 million ran- the credit score.1 million con- whether to extend credit to two individuals with identi. As is evidenced in this and March 31.g.1 5. 2005.817 0. Table 2.795 100. The lender may even prefer to extend domly selected individuals designed to represent the credit to an individual with a more accurate but lower broader population of consumers with no fully credit score than to an individual with a less accurate reported utility or telecommunications trades on but higher credit score.4 2 1. cal credit scores—the first of which uses alternative electric. (e. For example. more likely to lend to the first applicant.0 590.811 67. O ur analysis uses a data set constructed by Although not quantified in this study. sumers with at least one “fully reported” utility (gas.1. The research T H E DATA F O R T H E presented in this report has been designed to estimate the probable magnitude of these different effects and S I M U L AT I O N S to identify the types of consumers who are most likely to be affected. % 1 5. delinquent accounts referred to collection agen- ated in this study.502 8. Table 2.0 are a major component of loan pricing. other studies. Distribution of Consumers by Number of Telecommunications and Utility Trades: 2005 Analysis Sample Number of Trades Consumers with Utility Trades Consumers with Telecom Trades No.. the reporting of likely to lend (and at better rates) to an individual of a timely payments is far less common. % No. 2005.8 38. Just over 7.9 4+ 419.076.414.

Specifically. and the validation files over a 12-month time period: ual consumers (that is.10 and telecommunications trades on a variety of credit scores. We also compare the confidentiality of individual consumers.000 consumers in the analysis file alternative data in consumer credit reports have a fully reported telecommunications trade. 2006. addresses. social March 31.million consumers in the analysis file have at least one A P P RO AC H fully reported utility trade. 2006 (the end of the perform- demographic data were merged with the credit reports. no names. comparative analysis of new accounts opened by those ance period. they are treated separately is paid to the credit profiles and score distributions of throughout this report. and about one-third have more than one (for example. 2005 (the date that the samples were security numbers. and TransUnion). we compared the number we purged all identifying information from the file. as measured by the Kolmogorov-Smirnov (K-S) credit reports for both the analysis and the validation statistic. This analysis used a or her utility and telecommunications data. consumers who use a T combination of gas and electricity in their homes. Several commercial grade scor- samples at two points in time: March 31. • We sent the data to an independent service provider. ance period. “VantageScore.500 records have both a utility and a telecom. and on credit scoring models and on credit access only 8 percent have more than one. bureau data in two ways: The next step in the analysis examined the impact of • We used a variety of credit scoring models to score removing the telecommunications and utility trades on each consumer in the sample with and without his the consumer’s credit score. their demographic file (that is. or account numbers).” a generic scoring model recently intro- duced by the three national credit bureaus (Experian. Finally. 2005 (the ing models were analyzed to determine model predic- date that was used to generate the samples) and tiveness. Specifically.) In particular. only 591. including the telecommunications and characteristics. this group as well as access to credit with and without alternative data. derive a credit score for each consumer. and the effect of any reported utility utility trades). Once the drawn) and March 31.” during which the predictions of the with and without alternative data and an examination model were evaluated.) GIVE CREDIT WHERE CREDIT IS DUE 13 . Equifax. actual experiences of the consumers in the analysis the data contain no identifying information of individ. the analysis relatively little overlap between the two groups (only focuses predominantly on the 35–54 million about 1.9 out the utility or telecommunications trades. The intervening year is the “perform. with and with- ethnicity. age. We augmented the credit of acceptance rates for various communities. credit access is probed through a March 31. We used the model to who appended information on the individual’s race. and size of new accounts that were opened by con- sumers with an existing utility or telecommunications trades (the analysis sample) with the number and size of new accounts opened by otherwise similar con- sumers without such trades (the validation sample. and household income. The third step in the analysis focused on the impact that utility and telecommunications data would have We took deliberate steps to ensure the privacy and on consumers’ access to credit. Americans outside the credit mainstream. is examined. We then compared the distribution of these hypothetical scores The resulting data set contains a wealth of information with the score based on the consumer’s existing credit on the credit profiles of consumers. Because there is by various communities.) In his study examines the impact of including contrast. Attention munications trade). Then credit scoring model perform- We collected detailed information from the consumers’ ance.

or a combination of the two. and test the extent to which the resulting models considered in this report include: scores accurately predict consumer performance over a 12-month period: April 1. 2006. which predicts the probability that a In general. this should ultimately lead to higher acceptance rates. 14 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . The scoring trade line(s). The impact on declare bankruptcy in a two-year period.13 models. lower delinquency rates. • TransRisk New Account.That is. does this information impact credit behavior? • A mortgage screening model developed by a major lender that exclusively relies on credit bureau data We then examined how the reporting of utility and and predicts the probability that a consumer will telecommunications trades would affect the predictive have at least one 60-day delinquency on a mortgage power of several generic and industry-specific scoring account over a two-year period. if the presence of utility and telecommuni- consumer will have at least one 90-day delinquency cations trades helps to improve the models’ accuracy. and estimated the impact that this would have on both the availability and cost of credit. on a new or existing account over a two-year period. which predicts the proba- bility that a consumer will have at least one 90-day The final step in the analysis explored how different delinquency on a new account over a two-year demographic groups are likely to be affected. 2005 to March 31. We first period. We next estimated the probable impact of such trades which predict the probability that a consumer will on acceptance rates within each group. estimated the relative importance of energy utility and telecommunications trades for different demographic • Two separate bankruptcy scores (one from a large groups by examining each group’s share of total trades. financial institution and one from TransUnion12). with and without their utility and telecommunications cated in the United States to consumers.11 Credit We used these different models to score consumers scores are the principal means by which credit is allo.14 • VantageScore. and acceptance rates again reflects the extent to which the predictive power of scoring models improves with the addition of utility and telecommunications trades.

L I M I TAT I O N S The analysis sample is also limited in two other
respects. The analysis file is necessarily restricted to
consumers with either a utility or telecommunications

T
he analysis has a few limitations that should be trade. As a result, the findings they cannot be used to
noted. Most relate to the underlying character- make inferences to the broader population, which
istics of the analysis sample and the scoring includes an unknown number of consumers with nei-
models. ther a utility nor a telecommunications account. In
addition, consumers in our analysis file are unlikely to
have all of their utility and telecommunications trades
reported. Despite the fact that many consumers pay
SAMPLING ISSUES both a utility and telephone bill, there is relatively little
overlap between the two trade accounts in our sample.
Because of the local nature of both utility and Furthermore, the telecommunications data are domi-
telecommunications providers, we knew from the start nated by wireless accounts and may therefore underes-
that the analysis sample would not be representative. timate the full effects of reporting both land lines and
In fact, 84 percent of our data on consumers with util- cell phone accounts. As a result, our analysis will likely
ity trades is concentrated in the three states—Illinois, underestimate the potential impact of full reporting.
Wisconsin, and Pennsylvania—where several large
local utilities have begun to report their data. Likewise,
81 percent of the records with telecommunications
data were from Pennsylvania and Texas. MODELING ISSUES

The validation sample was designed to test the extent It is important to recognize that many of our findings
to which the analysis file is representative in other are based on the current versions of existing scoring
ways, for example, the number of trades in the con- models. In the event that utility and telecommunica-
sumer’s files excluding telecommunications and utili- tions data were more broadly reported, many scoring
ties. The results of this analysis are presented in models would undoubtedly be optimized to reflect this
Appendix A. As discussed there, the analysis file important change. However, on the basis of an earlier
appears to be broadly representative of all consumers analysis of a similar issue,15 we believe that any biases
in terms of their overall credit profiles and demo- introduced by this simplification will not affect our
graphic mix. In general, however, consumers with util- overall conclusions regarding the probable impact of
ity or telecommunications trades seem to have stronger full reporting. This limitation likely means our findings
credit profiles than the general population, although will tend to err on the side of caution, attenuating the
this is less true for consumers with telecommunica- actual impact we would expect with increased report-
tions trades. ing of alternative trades.16 ■

GIVE CREDIT WHERE CREDIT IS DUE 15

I I I . I M PAC T O N
CONSUMERS’ CREDIT
P RO F I L E S
The full reporting of utility and telecommunications data would
clearly affect the credit profiles of most consumers by adding one or more trade lines to their

files. Logically, consumers with little, if any, “traditional” forms of credit would have the most to gain.

(Simulations below suggest that this is in fact the case.) This section details the results of our estima-

tion of the potential magnitude of these effects by examining the impact of the utility and telecommuni-

cations trades on the consumer’s total number of trade lines as well as their credit score.

Table 3.1 compares the distribution of consumers by As shown in Table 3.1, the reporting of both utility
their total number of trade lines, with and without any and telecommunications trades has a sizable impact on
utility or telecommunications trades.17 The first two the credit profiles of the consumers in our sample. For
columns refer to the sample of 7.5 million consumers example, when utilities are included in consumers’
with an existing utility trade. Column 1 shows the dis- credit reports (column 1), about 12 percent of the
tribution of these consumers on the basis of the total sample can be classified as having a thin credit file
number of trades that currently appear in their credit (fewer than three established trades). However, when
files (that is, including any utilities.) Column 2 pres- the utility trades are removed from their credit records
ents the counterfactual, the distribution of these same (column 2), the proportion of thin-file borrowers rises
consumers when their utility trades are excluded. The to 17 percent, and about 10 percent of the sample
last two columns present comparable information for have no reported trade lines at all.19
the sample of 590,795 consumers with at least one
fully reported telecommunications trade. Column 3 The impact of adding the telecommunications trades is
shows the distribution of these consumers based on similar, although the impact on the share of con-
the information currently appearing in their files (i.e., sumers with no established trade lines is more pro-
including any telecommunications), while column 4 nounced. For example, when their telecommunications
illustrates what this distribution would have looked trades are reported (column 3), about 18 percent of
like had the telecommunications trades not been the sample would be classified as having a thin credit
reported. file. However, when their telecommunications trades
are removed (column 4), the share rises to 23 percent,
and 14 percent of the sample would have had no
established trades.

16 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E

Table 3.1. Impact of Utilities and Telecom Trades on Total Number of Trades23

Consumers with Utility Trades Consumers with Telecom Trades
Total Number Including Excluding Including Excluding
of Trades Utilities (#1) (%) Utilities (#2) (%) Telecoms (#3) (%) Telecoms (#4) (%)
Thin-File
0 - 9.6 - 14.0
1 7.7 4.0 13.4 4.9
2 4.1 3.4 5.0 4.1
Thick-File
3 3.5 3.2 4.1 3.7
4 3.2 3.1 3.8 3.5
5 3.1 3.1 3.5 3.3
6 3.1 3.1 3.4 3.2
7+ 75.2 70.5 66.8 63.3
All Consumers 100.0 100.0 100.0 100.0

Sample Size 7,519,020 7,519,020 590,795 590,795

Data Source: March 31, 2005 Credit Files for Analysis sample.

Differences in the impact of telecommunications and “unscoreable” to a “scoreable” situation, should be
utility trades most likely reflect underlying differences viewed as a significant change. Where along the score
in the populations with such trades. For example, a range the change occurs is also important. For
comparison of the underlying credit profiles of the two instance, a consumer gaining 50 points and moving
groups of consumers (columns 2 and 4) suggests that from 900 to 950 may gain little in practical terms rela-
consumers with telecommunications trades have a tive to a consumer also gaining 50 points but moving
smaller number of traditional trade lines than con- from 650 to 700.
sumers who are responsible for utility payments. In
this respect, consumers with telecommunications One would expect the reporting of utility and telecom-
trades appear to be more similar to the general popula- munications data to increase the number of consumers
tion than do consumers with utility trades (see who could be scored by increasing the their trade
Appendix A). Because it is easier to get a cell phone lines. However, there is no a priori reason to expect
than to rent or buy a home, this pattern makes sense. that the reporting of utility or telecommunications
data will change a consumer’s existing credit score in
Figures 3.2c and 3.2d show the impact of adding the one direction as opposed to another. Although a good
utility and telecommunications trades to the con- payment history on a larger number of trades will tend
sumer’s VantageScore. (This score ranges from 501 to to increase a consumer’s score, a poor payment history
990, with higher scores signifying lower credit risks). on additional trades would most likely reduce it.
Figures 3.2a and 3.2b show the distribution of con-
sumers by the change they experience when adding
their utility and telecommunications trade lines to
their scores. In general, a change of more than 25
points in the VantageScore, or a change from an

GIVE CREDIT WHERE CREDIT IS DUE 17

2d Impact of Telecom Trades on VantageScore Change on VantageScore 50% Including Telecoms Excluding Telecoms Percent of Consumers with Telecom Trades 30% 45% Percent Comsumers Affected 40% 25% 35% 30% 20% 25% 15% 20% 15% 10% 10% 5% 5% 0% > 50 25–49 10–24 < 10 No < 10 10–24 25–49 > 50 Can Remain Decline Decline Decline Decline Change Increase Increase Increase Increase Now be a “No 0% Scored Score” 601–620 621–680 681–740 741–800 801–850 851+ Credit Score Change (in points) Credit Scores Source: March 31. 18 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .2b Impact of Telecom Trades Figure 3. 2005 Credit Files for Analysis sample. Figure 3.2.2a Impact of Utility Trades Figure 3. Impact of Utilities and Telecommunications Trades on VantageScore All Consumers Figure 3.2c Impact of Utility Trades on VantageScore Change on VantageScore 50% Excluding Utilities Including Utilities 45% 30% Percent of Consumers with Utilty Trades Percent Comsumers Affected 40% 25% 35% 30% 20% 25% 20% 15% 15% 10% 10% 5% 5% 0% > 50 25–49 10–24 < 10 No < 10 10–24 25–49 > 50 Can Remain Decline Decline Decline Decline Change Increase Increase Increase Increase Now be a “No Scored Score” 0% 601–620 621–680 681–740 741–800 801–850 851+ Credit Score Change (in points) Credit Scores Figure 3.

“The impact of adding utility and telecommunications trades is considerably greater for thin-file consumers than for the population at large. However. In fact. often place unscoreable consumers among the highest among consumers who could be scored without their risk. It should be kept in mind that lenders scored. resulting in no change or changes less decreases the proportion of consumers who cannot be than 10 points.6 percent) was about the same as the share (Some lenders of course will attempt to collect infor- whose score decreased by more than 25 points (5. given that more than 25 points with the addition of the utility they have little on which to base their decisions. That is.” Adding utility data to the consumer’s credit report the sample. trades (4. a share of the 12 percent would be utility trade lines. but percent).) little or no significant effect on about 69 percent of GIVE CREDIT WHERE CREDIT IS DUE 19 . the inclusion of the utility data had this track is far more costly. the share whose score increased by treated as belonging to the lowest-risk tiers.2 mation to get a better sense of the applicant’s risk. from about 12 percent to 2 percent.

25–49 pt.3a Impact of Utility Trades Figure 3. No < 10 pt. No < 10 pt. Can Remain 851+ 801–850 741–800 681–740 621–680 561–620 501–560 No All Increase Increase Increase Increase Change Decrease Decrease Decrease Decrease Now be a “No Score Consumers in Score in Score in Score in Score in Score in Score in Score in Score Scored Score” Credit Scores Credit Score Change (in points) Figure 3. 10–24 pt.3b Impact of Telecom Trades Figure 3. Can Remain 851+ 801–850 741–800 681–740 621–680 561–620 501–560 No Score Increase Increase Increase Increase Change Decrease Decrease Decrease Decrease Now be a “No in Score in Score in Score in Score in Score in Score in Score in Score Scored Score” Credit Scores Credit Score Change (in points) Source: March 31. < 10 pt.3d Impact of Telecom Trades on VantageScore Change (Consumers with Less than on VantageScore 3 Traditional Trades) Including Telecommunications Excluding Telecommunications Percent of Consumers with Telecom Trades 70% 80% 60% 70% Percent Comsumers 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% > 50 pt 25–49 pt 10–24 pt. 20 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .3c Impact of Utility Trades on VantageScore Change on VantageScore (Consumers with Less than 3 Traditional Trades) Including Utilities Excluding Utilities 70% 120% Percent of Consumers with Utilty Trades 60% 100% Percent Comsumers 50% 80% 40% 60% 30% 40% 20% 20% 10% 0% 0% > 50 pt 25–49 pt 10–24 pt.3. 10–24 pt. 25–49 pt. < 10 pt. 2005 Credit Files for Analysis sample. > 50 pt. > 50 pt. Impact of Utilities and Telecommunications Trades on VantageScore Consumers with Less than 3 Traditional Trades Figure 3. Figure 3.

the share who experienced an increase of more than 25 points in their score (3.3 presents comparable statistics for borrowers with less than three traditional trades (or more pre- cisely.) This segment represents the population of most interest. The reporting of telecom- munications data had an even greater effect. As expected. ■ “The primary effect [of using alternative data] is to increase the percentage of consumers who can be scored” GIVE CREDIT WHERE CREDIT IS DUE 21 .1 percent. as many of these borrowers have difficulty accessing mainstream credit. the primary impact of including telecommuni- cations data appears to be on the proportion of con- sumers who cannot be scored.) Although the number significantly affected was higher than it was for the utility data. However. excluding any telecom- munications and utility accounts.2 percent) was only about one-half the proportion of consumers who expe- rienced a decline (7. among the consumers who could be scored without their telecommunications data. declining from 68 percent to less than 1 percent. less than three trades. Figure 3. adding utility data reduced the percentage of thin-file consumers who could not be scored from about 65 percent to just 4 percent. and the primary effect is to increase the percentage of consumers who can be scored. which drops from 17 percent to 1 percent. the impact of adding utility and telecommunications trades is considerably greater for thin-file consumers than for the population at large. telecommunications data had little or no effect on the credit scores of about 63 percent of the population. Again.Roughly comparable patterns can be observed in the sample of consumers with telecommunications data. For example.

1.. 22 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . In addition to comparing the proportion of consumers who opened a new account within this period. Because consumers in the validation sample have no reported utility and telecommunica- tions trades. we compare the actual experiences of con- sumers in the analysis and validation files over a 12- month period beginning April 1. they provide a convenient. The last three columns restrict the analysis to thin-file consumers. credit use) and the average change in the consumer’s aggregate credit limit.I V. To estimate the potential impact of the utility and telecommunications trades on the consumer’s access to credit. although imperfect “control” for assessing the potential effects of full reporting. one would expect that the full reporting of utility and telecommunications data would increase access to credit by reducing the proportion of consumers with thin credit files and increasing the proportion of consumers who can be scored. Although we the impact on consumers with a well-established credit history is relatively modest.e.20 The results of this analysis are presented in Table 4. The first three columns in Table 4. 2005 and ending on March 31. the impact on consumers with less than three traditional trades was quite pronounced. including the average change in the consumer’s total outstanding credit balance (i. 2006. we also looked at other indicators of credit use.1 describe the results for the three populations groups. O B S E R V E D DIFFERENCES IN AC C E S S T O C R E D I T All else equal.

14 1.323 504.500 and $1.$402 _ Total available credit + $6973 + $3192 + $12309 + $2466 + $1094 .240 1.93 0. only about 5 percent of thin-file borrowers in the validation sam- ple (column 6) opened a new account between April 1. while thin-file consumers without such trades experienced a small decline ($382).681 1. For example. 2006 Credit Files for Analysis Sample In general. New Credit Accounts Opened February 2005 to January 2006 All Borrowers Thin-File (<3 Traditional Trades) Consumers Consumers Consumers Consumers with Utility with Telecom Validation with Utility with Telecom Validation Trades (#1) Trades (#2) Sample (#3) Trades (#4) Trades (#5) Sample (#6) Pct with new accounts 50. 2006. ■ GIVE CREDIT WHERE CREDIT IS DUE 23 .357 Data Source: March 31. thin-file consumers with utility and telecommunications data increased their credit limits by about $2.785. over the 12-month period. the impact was significantly greater for thin-file borrowers.1.481 3. trades opened 1.030. respectively).$382 Sample size 6.42% 4.100.21% 16. no. However.44% 16. those with a fully reported utility or telecom- munications trades also experienced greater increases in their use of and access to credit.26 0. widespread reporting of utility and telecom- munications data increases consumer access to credit. and telecommunications data Compared with thin-file consumers without such increased their credit limits” trades. with larger increases observed for consumers in the validation sample. Although the proportion of consumers who opened a new account over the observation period was higher for all consumers with a fully reported utility or telecommunications trade. respectively.27 0.396 113. and March 31.036.05 _ Total outstanding balance + $3956 + $1466 + $8489 + $1972 + $891 . Table 4. the pattern for all con- sumers shows the opposite effects. In fact. 2005 and March 31.211.92% 48. compared with 16 percent of thin-file consumers who had either a reported utility “Thin-file consumers with utility or telecommunications trade (columns 4 and 5.07 0.61% Ave. 2005.73% 42.

therefore. I M PAC T O N SCORING MODELS Another way to assess the probable outcome of full reporting is to examine its impact on the reliability or ability to rank risk within the scoring models. primarily those lending for mortgages. and overstate the and utility trades. attempt to validate the credit-worthiness of no-score munications accounts. and a mortgage screen. In reality. including guish between good and poor credit risks. we first assumed that ing model Although none of these models specifically consumers who could not be scored would be treated distinguishes telecommunications or utility trades from as a higher risk than consumers with the minimum other types of accounts. T o examine these potential effects. The accuracy of the various scores 24 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . greater accuracy in estimating credit performance should lead to lower credit costs for lenders. In calculating the KS statistics. applicants by examining nontraditional sources of credit. if better per- formance reflects better capacities of borrowers to pay. in this case. or some combination of the two. Our analysis. it limits over-indebtedness. 2006). reporting utility and telecommunications trades should affect consumers’ access to credit if the additional information provided improves the ability of credit issuers to identify a good credit risk. such as for mortgages. 2005.21 Moreover. In gen- eral. the VantageScore model. would reported trade lines. however. including any utility or telecom. higher acceptance rates. We then used the resulting scores to benefits that arise when consumers move from rank consumers according to their predicted risk. may oversimplify the We began by scoring consumers in the analysis file decision-making process of credit-issuers in some with and without their reported telecommunications instances. compared the different rankings with consumers’ per- formance over a 12-month period (April 1. I M PAC T O N P R E D I C T I V E was summarized by their Kolmogorov-Smirnov (K-S) statistic.V. As shown in prior research. some credit be affected by the consumer’s performance on all issuers. to March 31.22 The K-S statistic ranges from 0 to 100. we relied on with higher values signifying a greater ability to distin- several commercial scoring models. a commonly used metric designed to capture POWER a model’s ability to distinguish between two different groups. two bankruptcy models. a generic new account model. performing and nonperforming accounts. the scores of each model will applicable score. and unscoreable to scoreable.

adding utility and telecommunica. For example.23 For example. As mentioned. The greater lift (that is.000 1.8 percent and 8.5 percent. the K-S statistics for the reordering effect from the addition of the new each of the models has been scaled to equal 100 when data. without the alternative data. This is evident from comparing the results in Table 5. Table 5.1. these con- As shown Table 5. Nonetheless.1 shows the esti.323 6 . rate observed among consumers with scores in the 680 tively. adding able without their utility trades had a delinquency rate the data to the VantageScore model increases its over.262 1.481 Data Source: March 31. sumers do not belong (as a group) in the highest-risk tions data increases the overall accuracy of the scoring category.481 504. With these caveats in mind. Impact of Utilities and Telecommunications Trades on K-S Statistics: General Population Models Consumers with Utility Trades Consumers with Telecom Trades Including Excluding Including Excluding Model Utilities (#1) Utilities (#2) Telecoms (#3) Telecoms (#4) VantageScore 1. ranging from a 5 percent increase for the rates observed among consumers with lower scores second generic model to nearly a 14 percent increase (whose delinquency rates ranged between 33 percent for the bankruptcy scores in the utility sample and and 60 percent).048 1. those who would most benefit from the inclusion of rather than to a better risk-ordering of those who can the alternative data.138 1. of 14 percent. increases of more than 20 percent for the bankruptcy scores in the telecommunications sample.000 1. which is only slightly greater than the all K-S statistic by 9.1 when previously unscoreable con- from the consumers’ credit files. Values above 100 sumers are scored and moved out of the greatest risk when the utility or telecommunications trades are category.211. This reflects the fact that the average rate of included indicate a relative improvement in the serious delinquencies among such consumers is rela- model’s predictive power. also calculated changes in the K-S sta- tistic for samples of consumers who could be scored The improvement in the model’s predictive power with with and without the alternative data.) To protect the pro. These calcula- the addition of the utility and telecommunications tions. consumers who were unscore- models by a significant amount. Hence.1. be scored without the addition of the alternative data.000 Sample size 6.051 1. thus. and well below the are similar.085 1. Results for the other general population models to 740 range of the VantageScore. 2006 Credit Files for analysis sample.000 TransRisk new account 1. but they do exclude ability to score previously unscoreable consumers.000 1. respec. which are based on calculations from samples models (For reasons described below.000 Bankruptcy model II 1. tively low compared with the scoreable consumers at the bottom of the score distribution. 2005 and March 31. the mortgage composed of only those who can be scored with or model has been treated separately.098 1.000 TransRisk bankruptcy 1.135 1. mated impact of adding the utility and telecommunica.211. increase in the KS statis- the utility and telecommunications trades are excluded tic) in Table 5.1 tions trades on the predictive power of the various and 7.000 1.323 504. make no assumptions regarding how those trades appears primarily to be driven by the greater with no score should be classified.214 1. and thus only captures prietary nature of the models. it is useful to explore GIVE CREDIT WHERE CREDIT IS DUE 25 . Table 5.

we should roughly the same order of magnitude.2.2.844 5.844 421. VantageScore model rose by more than a factor of 3 out it. of such consumers regis- tered a score for the VantageScore model. respectively.915 421. 26 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .000 1. we should expect a lift from adding utility and tional trade lines—the thin-file consumers.915 how the models’ performance is affected when includ. only 36 percent and 32 percent. data. the number of no-scores declined to a minimal amount.003 1. the addition of with and without utility and telecommunications another (alternative) trade line for the average con. telecommunications data to the credit files of the thin-file consumers when the scoring models are As before.439.2 makes it clear that for those who can already tions trades. timized models). with the addition of the utility data and by more than a factor of 4 with the addition of the telecommunica- Table 5.005 1.008 1. Impact of Utilities and Telecommunications Trades on K-S Statistics: Excluding Unscoreables24 Consumers with Utility Trades Consumers with Telecom Trades Including Excluding Including Excluding Model Utilities (#1) Utilities (#2) Telecoms (#3) Telecoms (#4) VantageScore 1. and the model’s ability to predict the credit performance of thin-file consumers increased dramatically.010 1. when scoring consumers with one or sumer should have little effect. These results are shown in Table 5. As shown in Table 5. reflecting the greater importance of addi- credit files can benefit those traditionally underserved tional trade lines to consumers (and those trying to by the mainstream financial sector. highest-risk consumers (they were placed at the bot- tom of the score distribution). we first treated those with no score as the optimized for such data.000 Bankruptcy Model II 1. only a modest improvement in underscore the critical nature of such trades in evalu- score model performance (at least with current nonop.000 TransRisk Bankruptcy 1. ating the credit performance of thin-file borrowers.000 Sample Size 5.025 1. for instance. that is.022 1. Again.012 1. Therefore. With the addition of the data. the K-S statistic for ing alternative data for those who can be scored with. In the absence of the utility and telecommunications data. model performance for those with little or no tradi.2.4 shows the change in model performance sumers in the utility subsample used had seven or when scoring thin-file consumers who are scoreable more traditional trade lines. Table 5. more than three-quarters of the con. We see a larger average lift with the addition of the alternative data for the thin- Because the purpose of the study is to determine file consumers than for the general sample results in whether and how the addition of alternative data in Table 5. This should be expected given that.000 TransRisk New Account 1. two traditional trade lines.000 1.000 0.3.000 1. Table 5. These findings expect.439. we now look at estimate their level of risk) with few trade lines. The results for the other models are be scored without the alternative data.987 1. on average.

000 6.000 Bankruptcy Model II 3.000 TransRisk Bankruptcy 3.783 1.294 1.297 1.035 1. the samples we used to esti- mate the K-S statistics were limited to consumers with mortgage trades at the beginning of the performance GIVE CREDIT WHERE CREDIT IS DUE 27 . Table 5.000 5.000 Bankruptcy Model II 1.358 1. but noticeable.000 Sample Size 1.903 36.280. lthough the results are quite robust for the impact.021 1.553 1.281 1.506 These findings are consistent with what one would MORTGAGE expect with the addition of alternative data.506 36.000 4.553 137. applying the same relatively little change.000 0. and (3) consumers with thick files would see generic scoring models.000 4.024 1.595 1. Because mortgage screening mod- els are designed to predict the incidence of 60+ days mortgage delinquencies.000 1.932 1.903 369.280.000 TransRisk Bankruptcy 1.4.000 TransRisk New Account 2.061 1. 2006 Credit Files for Analysis Sample Table 5. 2005 and March 31. (2) thin- file consumers who were scoreable without the new A data would experience a smaller.978 1.256 137.000 0. that (1) the largest impact would be for those who SCREENING MODEL become scoreable after adding the new data.3.000 Sample Size26 369. Impact of Utilities and Telecommunications Trades on K-S Statistics: Thin-File Borrowers Only Consumers with Utility Trades Consumers with Telecom Trades Including Excluding Including Excluding Model Utilities (#1) Utilities (#2) Telecoms (#3) Telecoms (#4) VantageScore 3.993 1.256 Data Source: March 31.000 1. Impact of Utilities and Telecommunications Trades on K-S Statistics: Thin-File Borrowers Only. approach to the mortgage screening models proved problematic. namely.000 TransRisk New Account 1. Excluding Unscoreables Consumers with Utility Trades Consumers with Telecom Trades Including Excluding Including Excluding Model Utilities (#1) Utilities (#2) Telecoms (#3) Telecoms (#4) VantageScore 1.971 1.078 1.050 1.

and the great majority had ON THE PREDICTIVE thick credit files. These results refute any notion that utility performance across all trades—they nevertheless sug. we cal- ing models using an alternative performance measure: culated the correlation between a serious delinquency the incidence of any 90+ day delinquency.4 percent and 3. 28 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . The respective correlations were models.period. they were statistically significant. Although these results should be utility or telecommunications trade is weakly to moder- interpreted with caution—mortgage models are specifi. In particular. tic in the overall samples or in the thin-file samples. 95. and more simply. homebuyers. while the telecommunications data led to points to this. DATA sumers. their obligations compared with someone who had fallen ance of consumers with mortgages. com- pared with 70. We simi- whether or not they had a mortgage trade.9 percent decline.5 percent in an overall sample of con. predictor of future payments.6 percent of mort- gage holders in the utility sample had seven or more P O W E R O F A LT E R N AT I V E traditional trades (i. we Specifically.e. line between March 2004 and March 2005.28 The of the homebuyers model by 13. Likewise. The correlation does not.288 and . That we see a rise in the K-S statis- of the utility data led to a 0. Not surprisingly. explain whether alternative payments are a good designed for mortgage loans. In fact. we could 0. If an data had relatively little impact on a model’s ability to individual has been making his or her utility or telecom- predict the performance of thick-file borrowers. using the sample of consumers with utility recalculated the K-S statistics for the mortgage screen. e would expect that alternative payment data would contain some information useful in Given that including utility and telecommunications predicting future payment outcomes. telecommunications data could enhance a mortgage lender’s ability to identify credit-worthy borrowers. how- credit models are likely to apply to models specifically ever. and telecommunications payments are unrelated to gest that the improvements observed for the generic traditional payments. all of the consumers in these A D D I T I O N A L R E S U LT S subsamples had at least one established traditional trade (their mortgage). it is munications payments on time for a period of time.4 percent decline in the K. we found that utility and . and S statistic of a mortgage screening model designed for including or excluding the unscoreable populations. The results larly calculated serious delinquencies for telecommuni- of this analysis were similar to the generic scoring cations trade lines..27 look at the correlations between a serious delinquency on an alternative trade and a serious delinquency on a To gain a better understanding about how utility and traditional trade. For example. we therefore not surprising that these data had relatively would expect they would be more likely to make timely little impact on the K-S statistics of the mortgage payments (in the present and the future) on a variety of screening models. fewer than 1 percent of mortgage holders in the utility sample had thin credit files com- W pared with about 17 percent in the overall sample. the addition behind on payments. respectively. In addition.2 per. results indicate that a serious delinquency on either a cent. ately correlated with a serious delinquency on a tradi- cally designed to predict mortgage performance not tional trade. trade lines who also had traditional trade lines. not surprisingly given the very large telecommunications data increased the K-S statistics sample sizes.292 and. We based (90+ days) on a utility trade and on a traditional trade this analysis on the entire sample of consumers. excluding utility trades). on the basis of the observed perform.

a utility trade.146 436.0001) (. Regression Results. as measured by the R-squared.424 0.0002) (. we ran predicting. GIVE CREDIT WHERE CREDIT IS DUE 29 .410 (.110 .130 (. the predictive power or good- and having such a delinquency on any trade the follow.631. The correlation for of the telecommunications data.3009 0.106 . and do a much better job of not add any predictive power. ness of fit of this admittedly crude model rises by 40% ing year is 0. of the model rises by 17% also as measured by the lation between a consumer’s serious delinquency and R-squared. The corre.2136 0. With the addition tions delinquencies during is 0. or a telecommunications trade are quite similar. Dependent Variable: Whether a Consumer Had a Serious Delinquency on Any Trade During March 2005 and March 2006 (Standard Errors in Parentheses) Consumers with Utility Consumers with Telecom and Traditional Trades and Traditional Trades Variables (#1) (#2) (#3) (#4) Constant .32. tion that could be useful in predicting future pay- ment outcomes.0006) Whether a Traditional (90+ DPD) Delinquency.5 indicate that with the addi- on a utility trade during March 2004 and March 2005 tion of the utility data.485 (. To test this.0004) (. Nonetheless.5.082 .0006) (.46.511 . A commercial model would information from traditional trades.140 The correlation between having a serious delinquency The results in Table 5.247 (.0014) (. Table 5.29 serious delinquencies on a traditional trade.0014) Whether a Utility (90+ DPD) Delinquency. this is only suggestive of how the addition of utility and telecommunications payment informa- It could be the case that the predictive information tion would affect model fit in a reoptimized commer- alternative trades embody is already captured in the cial-grade scoring model. the goodness of fit delinquencies on a traditional trade is 0.140 436.0017) R-Squared 0.0004) (.42.2143 Sample Size 5. March 04-March 05 . March 04-March 05 .631. take into account much more adding such alternative trades to traditional trades may detailed information.146 5. March 04-March 05 . Of course.0005) Whether a Telecom (90+ DPD) Delinquency. Such a correlation for telecommunica.412 0. it appears that utility and regressions to determine whether adding alternative telecommunications payment data contain informa- trade information would improve predictability. and therefore be more sophisticated.2506 0.

40% 16.90% 90% 13.80% 15.3 to 1. the trade-off between the size rates among credit card issuers.20% 1. tion of utility and telecommunications data.90% 20.80% 2. that is more or less in line with the current acceptance dictive power.60% 30 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . On the one hand.70% 2. serious delinquencies would fall by portfolios becomes less steep.I M PAC T O N Although it is difficult to predict the market outcome.7. With this target of the lender’s market and the performance of their acceptance rate. Although the data in AC C E P TA N C E R AT E S the tables are based on the VantageScore model.6 and 5.8 percent) with the Table 5.20% 21. Serious Delinquencies by Target Acceptance Rates: VantageScore Model Consumers with Utility Consumers with Telecom- Trades Telecommunications Trades Including Excluding Including Excluding Utilities Utilities Telecoms. As before.70% 18.50% 13.00% 4. we have I n a competitive market.10% 1. given acceptance rate will necessarily decline with an improvement in the model’s predictive power. a rate rate will decline with improvements in the model’s pre. suppose that a credit issuer wished to main- default rate that is associated with a given acceptance tain an acceptance rate of about 50 percent.60% 60% 3. about 22 percent (from 2.6.30 This in two different ways. In short.90% 1. consumers could benefit assumed that lenders would put consumers who can- from an increase in the accuracy of scoring models not be scored in the highest-risk category.30% 40% 1. lenders could maintain their existing a given acceptance rate could improve with the addi- acceptance rates but lower their rates and fees. the types of trade-offs that credit issuers face with full D E L I N QU E N C Y A N D reporting of utility and telecommunications trades are illustrated in Tables 5.10% 12.20% 7.31 For a price reduction would be possible because the example. credit assumes that the unscoreable population is essentially issuers could increase their acceptance rates and keep excluded from consideration (given that they are put at the rates that they charge the same.6 shows how the performance associated with Alternatively. Again. results for the other models are generally similar and are presented in Appendix B. Table 5.80% 17.10% 1. Increasing their the bottom of the risk distribution) but nonetheless acceptance rate without increasing rates and fees is count as potential borrowers/consumers (their pres- possible because the default rate associated with a ence is felt in the numerator of the acceptance rate). Telecoms- Acceptance Rate (#1) (#2) Telecommunications Telecommunications 30% 0.40% 10.30% 3.20% 50% 1.50% 1.10% 70% 5.40% 8.20% 80% 9.30% 4.

3 percent) with the full reporting of telecom- munications accounts.addition of utility data.9 49.0 44. For example. Nevertheless.7 takes the opposite perspective.6 to 3.4 47. As noted earlier.3 51.4 5 69.5 67.1 63.4 53.7.4 54.8 3 60. In a highly competitive market.9 to 60.4 38. Table 5.9 to 49. reporting.4 59. and by about 28 percent (from 4. the savings associated with these declines would ulti- mately be passed through to consumers in the form of lower rates. and shows what would happen to acceptance rates if issuers wished to maintain their current level of risk (as meas- ured by the incidence of serious delinquencies) and expand their business base. many credit issuers attempt to create but they may do so only slightly.3 and 5.4 percent with the addition of utility data using a targeted delinquency rate of about 3 percent—the approximate average for credit cards.4 may tend markets.2 43.0 6 72. acceptance rates could rise from 54. Table 5. acceptance rates could rise from about 44.0 GIVE CREDIT WHERE CREDIT IS DUE 31 .9 4 65.6 48.6 52. With the addition of the telecommunica- tions data.9 59.3 7 74. As a result.7 57.1 55.0 percent without increasing projected losses. ■ to overestimate the actual impact on acceptance rates. and the value it can bring to underserved the findings presented in Tables 5. our alternative credit histories for thin-file borrowers by analysis clearly illustrates the potential impact of such turning to non-traditional credit sources. Acceptance Rates by Targeted Delinquency Rates: VantageScore Model Consumers with Utility Consumers with Telecom Trades Trades Including Excluding Including Excluding Delinquency Rate % Utilities (#1) Utilities (#2) Telecoms (#1) Telecoms (#2) 2 52.4 55.0 65.

To simplify the presentation. The addition of the utility trades would increase acceptance rates for both black and Hispanic borrowers by about 21 per- cent. as before. However. Again. by about 13 percent for consumers under the age of 25.” where a widow is left with little credit history because bills had been in her husband’s name.1c).000 per year (see Figure 6. find any difference by gender. the results are much the same when other models or risk cut-offs are used. we again present our results for just one model—the VantageScore model— and use a “targeted” delinquency rate (3 percent) that approximates the average for credit cards.33 In general. Likewise. acceptance rates would rise by about 25 percent for consumers earning less than $20. lower-income groups.VI. the largest impact is associated with the addition of the utility data. more than twice the increase observed for whites (see Figure 6. DEMOGRAPHIC I M PAC T S Figure 6.1 shows how changes in acceptance rates would vary across different demographic groups32 assuming that the risk tolerance of lenders remains the same. We did not. however. and younger (18 to 25 years old) and older (66+ years) consumers are most affected by the addition of utility and telecommunications data.1a). minorities. although the results are roughly similar for the utility and telecom- munications trades. 32 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . We were curious whether the 65+ group was evidence of “widow effect.1b). and by 14 percent for those over age 65 (see Figure 6.

000+ Figure 6.999 $49.1e Consumers by Language Preference $29.1c Consumers by Age with Utility and Telecom Trades (Assumes 3 percent Serious Delinquency Rate) (Assumes 3 percent Serious Delinquency Rate) Including Utilities Including Telecoms Including Utilities Including Telecoms 25% 16% 14% 20% 12% Acceptance Rate Acceptance Rate 10% 15% 8% 10% 6% 4% 5% 2% 0% 0% All Consumers Asian Black Hispanic White Other 18–25yr 26–35yr 36–45yr 46–55yr 56–65yr 66yr+ Figure 6.000– $50.000– $30. Impact on Acceptance Rates by Demographic Group: (assumes 3 percent serious delinquency rate) Figure 6.1b Consumers by Income Figure 6.999 with Utility and Telecom Trades (Assumes 3 percent Serious Delinquency Rate) Consumers with Utility Trades Consumers with Telecom Trades —Including Utilities —Including Telecoms 30% 25% Acceptance Rate 20% 15% 10% 5% Source: March 31. Figure 6. 0% English Spanish GIVE CREDIT WHERE CREDIT IS DUE 33 .1d Consumers by Homeowner Status with Utility Trades with Utility and Telecom Trades (Assumes 3 percent Serious Delinquency Rate) (Assumes 3 percent Serious Delinquency Rate) Consumers with Utility Trades Consumers with Telecom Trades Including Utilities Including Telecoms —Including Utilities —Including Telecoms 30% 14% 12% 25% Acceptance Rate 10% Acceptance Rate 20% 8% 6% 15% 4% 10% 2% 0% 5% Own Rent 0% <$20.1.000– $100.000 $20.1a Consumers by Race with Utility Trades Figure 6.999 $99. 2005 Credit Files for Analysis sample.

25 8% Income <$20.22 7% 4% 24.54 5% $100.91 13% Hispanic 32% 13.92 11% 24% 12.57 8% 13% 18.21 7% Gender F 14% 18.999 20% 13.999 13% 16.16 11% Other 16% 18.13 12% $20.24 11% 40% 10.01 14% 38% 9.” 34 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .000 31% 11. lower-income consumers.19 8% 10% 19.88 12% 26–35 10% 19.12 9% 20% 16.000–$99.21 6% 56–65 8% 20.000+ 4% 24.54 6% Black 28% 12.32 9% 23% 15.43 11% 18% 13. and the young and the old are more likely to be thin-file borrowers.49 9% $30.89 8% 8% 20.21 7% White 14% 18.11 13% 36% 8.000–$49.46 11% 48% 7.02 8% 18% 17.85 6% 46–55 9% 20.51 7% $50.000–29.999 7% 20.33 7% M 12% 18.19 9% 18% 16.04 7% Race Asian 20% 17.44 9% 16% 16.29 7% 36–45 9% 21.81 8% 12% 19. Reported Trades by Borrower Characteristics Consumers with Utility Trades Consumers with Telecom Trades <3 Mean Utilities as <3 Mean Telecoms as traditional Number Percent of traditional Number Percent of trades (%) of Trades Total Trades trades (%) of Trades Total Trades All 17% 17.19 9% 22% 15.74 7% Age 18–25 24% 11.88 9% 16% 15.2. Table 6.39 6% 66+ 18% 13.24 5% “Minorities.35 9% 19% 16.

lower-income consumers. saw their acceptance increase at nearly twice rate as homeowners with the addition of the utility data. minorities. the estimated impact were not as large. Finally. As a result. This is probably a better measure than ethnicity of the underserved immigrant population from Latin America.Renters. language preference reveals that those who prefer Spanish as their primary language experience a 27 percent increase in their acceptance with the addi- tion of the alternative data. who presumably are less in the financial mainstream than homeowners. the addition of utility and telecommunications trades to their credit records will have a larger effect on their overall credit profiles. As shown in Table 6. Differences in the estimated impact on different demo- graphic groups reflect differences in their underlying credit profiles. Although similar pat- terns for all conditions are observed for the telecom- munications data. ■ GIVE CREDIT WHERE CREDIT IS DUE 35 . Renters may also find improving their credit files particularly important if they hope to become eventual homeowners.2. and the young and the old are more likely to be thin-file borrowers.

adding telecommunications data reduces the number I of unscoreable consumers from about 17 percent to n our view. The net result of full fected by additional information on utility and reporting should be positive for consumers and busi- telecommunications trades for consumers who can be ness alike. it seems safe to assert that relatively few consumers would be harmed by the full reporting PERC surveyed the members of the National of such data. Thin-file consumers would stand to gain by scored without them. Likewise. for consumers having a more accurate assessment of their credit-wor- with a relatively thick credit files. SUMMARY AND P O L I C Y I M P L I C AT I O N S The results of our analysis lend strong support to the suggestion that the sys- tematic reporting of telecommunications and utility trades would benefit consumers and increase their access to low-cost credit. consumer reporting agencies. if any. The primary effect of fully reporting energy utility ENCOURAGING and telecommunications data appears to be on the number of consumers who could be scored. For example. Assuming that our sample is reasonably representative of all consumers with such trades. Any privacy concerns should be carefully young (18 to 25 years) and the relatively old (over 65). credit. the transfer of customer data to third parties was sumers would be significant. and credit issuers would stand to gain by trade lines has little.VII. low-income groups. and consumers at the fully review those laws in light of the findings reported two extremes of the age continuum—the relatively here. Based A LT E R N AT I V E DATA on the tri-bureau VantageScore model. these findings provide a strong public 1 percent. In other words. tomer data with consumer reporting agencies (CRAs). Significantly larger gains would go to We believe that lawmakers in those states should care- minorities. As a result. tions trades. we estimate that overall ten with other concerns in mind—in most cases they acceptance rates could rise by as much as 10 percent are privacy rules—they clearly preclude sharing cus- with the full reporting of utility and telecommunica. and identified four states where In contrast. the addition of these thiness. Although these laws were writ- the VantageScore model. the impact on otherwise unscoreable con. based on statutorily prohibited. effect—either positive or enhancing their ability to expand their markets without negative—on their credit scores or their access to a concurrent increase in risk. the percentage of unscoreable consumers would decline from 13 per. policy rationale for encouraging the full reporting of utility and telecommunications payment data to Scoring models and credit scores are relatively unaf. REPORTING cent to 2 percent when adding utility data. weighed against the demonstrated social and economic 36 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . the impact is likely to be large. Association of Regulated Utility Commissions (NARUC) in 2005.

and they given an environment of heightened sensitivity to data will likely be charged a relatively high deposit because privacy and data security concerns. but not contribut- as the primary policy barrier to sharing energy utility ing to it. Given that have no traditional payment histories. benefits. Most of these firms. we encourage state lawmakers in those few states to carve out an exemption in existing Interestingly. Such an imbalance in The NARUC survey identified regulatory uncertainty using payment history information. telecommunications. either report only nega- ing agencies. energy utility and telecommu- law for reporting payment data—not detailed account nications firms have been major consumers of credit information such as customer proprietary network reports from the big three national credit bureaus. despite the absence of a contributed to it. For some uses of con- unwilling to provide energy utility and telecommunica. statutory prohibition. given that they the majority of states have no law on the books either will be building no positive payment histories by using precluding or permitting data sharing with CRAs. P R E S E R V E VO L U N TA R Y REPORTING W hen considering data-sharing legislation. such as for marketing and pre- tions firms with explicit permission (especially written screening lists. is particularly costly to those consumers who and telecommunications data with CRAs. where permission) to share customer payment data with companies wishing to use the information must have CRAs. and the utility or telecommunications services. defaults. sumer credit files. Specifically. In fact. The decision of any energy or telecommunications providers to become a “full file reporter” must ulti- mately be driven by a combination of each firm’s self- interest (in reducing account delinquencies)34 and by the understanding that doing so helps to promote access to mainstream credit markets for previously underserved groups. for years. in some cases. tive information (delinquencies. and collec- tions). In these states. we advocate the pas- sage of a law clearly permitting the sharing of cus- tomer data with CRAs. there is a principle of reciprocity. it is important to preserve the voluntary nature of the national credit reporting sys- tem. or do not report at all. however. information or CPNI—to accredited consumer report. some regulators have told inquir- ing energy utility and telecommunications firms that they were not permitted to share customer payment data with CRAs. or other alternative data will result in a radical and disruptive paradigm change to the world’s most successful credit reporting regime. But these benefits may hold little GIVE CREDIT WHERE CREDIT IS DUE 37 . Mandating the reporting of energy utility. regulators are they have no payment history.

efficient market responses may emerge by data aggre- T gators and credit bureaus to bring the nontraditional he sociodemographic analysis of the thin-file data online. Given the concentration of LMI households in most urban areas. For those living in uncertainties surrounding the transfer of such data such areas.value to entities. it is impossible to purchase a car for traveling to work. vant for this study was the finding that the portion of borrowers with extremely weak credit scores (scores lower than 75 percent of the total population) was considerably higher in urban markets than the national average. to secure a home mortgage loan or a small business loan to begin the process of asset building and wealth creation. this translates to a substantial barrier to accessing affordable capital to build assets in these urban markets. that provide services REPORTING ENHANCES typically considered necessities and often face little or no competition. However. It is their bottom line. and credit management. For low. and the preva- lence of automated underwriting among mainstream lenders.to moderate-income persons in urban areas. nearly 41 percent have credit scores in the bot- tom quarter for the nation. score distribution. when regulatory and legislative poorer. Most rele- building and wealth creation.to moderate-income individuals (LMIs) for 50 metropolitan areas. 38 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . the ability to improve one’s life often are cleared up. Many of these individuals reside in when statutory prohibitions are removed or amended. depends on access to credit. these market responses can happen recent immigrants. “domestic emerging markets”—urban markets and and more important. such as utilities. Without sufficient credit history. credit use. but also their customers. DEVELOPMENT OF Nonetheless. it is impossible to findings from this analysis of more than 14 million partial credit files during a one-year period indicate begin the process of asset high variance across metropolitan areas in credit use.35 Key history. they will composed largely of members of ethnic minorities. industrial and rural areas. as the value of consumer payment data COMMUNITIES from nontraditional sources becomes more evident. likely become more interested in supplying payment many of whom are economically disadvantaged and are data. and Without sufficient credit delinquency patterns for low. to secure a student loan for the college of choice. As potential furnishers of nontraditional and unscoreable population confirmed beliefs data realize how providing payment data not only helps about the characteristics of this group. A recent study analyzed credit scores.

Thin-file borrowers with one or more alter. lenders can profitably expand into previously overlooked markets—markets that may even become competitive. in an environ- ment of pervasive alternative data reporting everything changes. but landscape of consumer banking in urban areas should credit access for LMI borrowers is dramatically fundamentally change to the benefit of those who live improved. mainstream credit—albeit just one part of the solution—can greatly assist with the economic development of urban markets. at four times the rate of thin-file borrowers without resulting in improved opportunity and quality of life. millions of credit-worthy bor- rowers in urban areas who previously had to rely on check-cashing. and lenders will have the data necessary to build new alternative scoring models or optimize existing scoring models. Empowered with new tools and information. two avenues for asset-building. can have deep and systematic native trade lines in their credit files accessed capital affects on community development and asset-building. This. payday lenders. In short. then credit bureaus. Given the size of this population. Not only are the credit prise. GIVE CREDIT WHERE CREDIT IS DUE 39 . lenders will have the tools to process the newly avail- able information to make credit decisions. a secure a small business loan to launch a new enter- kets. if—and this is a big if—alternative data are reported in sufficient quantity in the near term (currently. in turn. the Americans improved by using alternative data. making possible The results from this study offer great promise for the dream of homeownership and the ability to secure community development in domestic emerging mar. The miracle of instant credit can palpably affect the lives and life chances of millions. Enhanced access to affordable. In an environ- scores of a majority of thin-file and unscoreable ment of pervasive alternative data reporting. In short. First. preliminary evi- dence strongly suggests that using alternative data in consumer credit reports makes a difference in credit access and fairness in lending. Perhaps most important. there. a small but growing minority of energy util- ity and telecommunications firms fully report customer payment data to one or more credit bureaus). or other predatory lenders can gain access to affordable mainstream credit. and its risk profile when alternative data are considered. especially in urban areas. analytics firms. ■ any alternative trade lines.

These two models rely on data from 40 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . consumer reporting agencies are not actively exhorting energy utility and telecommunica- tions firms to fully report data because their major clients—large financial institutions—are not demand- ing alternative data and alternative scoring models. there is a chicken-and-egg quality to alternative data. There does appear to be interest in using alternative payment data in the market. just under 5 per- cent of all credit files have one or more alternative trade lines. which uses self- reported (but verified) alternative payment informa- tion. FUTURE RESEARCH DIRECTIONS Evidence presented in this study supports the use of alternative data as one means to help bridge the credit information gap for millions of thin-file and unscoreable Americans. thus sidestepping legal and regulatory barriers and accessing payment information not in standard credit reports. the advantage of this model (self-reported data) also likely limits its impact in bringing useful alternative data online. First.VIII.36 Fair Isaac’s Expansion Score and First American’s Anthem model are scoring models specifically designed to use alterna- tive payment data. Lenders are not demanding alternative data and alter- native scoring models because so little alternative data is fully reported. However. it is not an easy solution. and alternative data composes less than 1 percent of all trade lines in a major credit bureau’s database. One example is Payment Reporting Builds Credit (PRBC). By one estimate. Although alternative data can be held out as a promising potential solution to the problem of too little credit information. That is.

fully reporting customer data to TransUnion. For instance. leading to its broad use. however. that much more needs to be done to jump-start a cycle of alternative data use and report- ing. and assum.38 A systematic survey of energy utility and telecommuni- cations firms on their experience reporting data to con- sumer reporting agencies could identify hurdles to reporting. assuming Brookings Urban Markets Initiative roundtable on alternative data and credit scoring. directly or in part. Nicor Gas reported a 20 per- cent reduction in delinquencies one year after it began interest. At a 2005 reporting agencies. and remain somewhat of a black box.niche aggregators. However. making this study possible. is in their best delinquencies. Data furnishers—in this Data furnishers—in this case case utility and telecommunications companies—must be convinced that reporting data to CRAs. that a small number of important lenders are beginning to use them in credit decision suggests that a demand for alternative scoring models exists. is in their best interest. while the reporting of utility and telecommunications payments is far from pervasive. can lead to a dramatic reduction in delinquencies and charge-offs. Demand will likely grow as more alternative data come online. It is clear. Anecdotal evidence companies—must be convinced suggests that fully reporting customer data to credit bureaus. the TransUnion database nonetheless had more than 8 million consumer files with at least one alternative payment reported for at least a year as of March 2005. From a policy perspective. utility and telecommunications ing Fair Credit Reporting Act data furnisher obliga- tions. Such an outcome could go a long way toward helping untold millions of thin-file and unscoreable Americans build assets and create wealth in a sustain- able fashion. led to a substantial reduction in guidelines. ■ GIVE CREDIT WHERE CREDIT IS DUE 41 . WE Energies and Fair Credit Reporting Act Verizon stated that fully reporting customer data. the results of such a survey and analysis could serve as the basis for a national outreach program to expedite an environ- ment in which alternative data are pervasively reported.37 Similarly. and consistently communicating the benefits that reporting data to credit of reporting to customers.

their Again. Likewise. they are essentially separate groups and have been treated as such throughout this report. other characteristics are about the same. age and income of the consumer. In general. a lower proportion of Hispanics munications trades have somewhat stronger credit and a higher proportion of blacks than the population profiles than the general population as measured by at large (as measured by the validation sample). and assess the gender. consumers with either a utility or telecom- proportion of males. Although the differences are relatively modest for CREDIT DIFFERENCES consumers with telecommunications trades. we removed the utility and on their own. Table A1 compares the demographic characteristics of the consumers in each sample. Table A2 compares their credit profiles excluding their utility and telecommunications trades. extent to our analysis file is representative of the the three population groups look remarkably similar. telecommunications) as well as their credit scores. SAMPLE C H A R AC T E R I S T I C S This appendix compares the characteristics of the analysis file with the characteristics of the validation sample. Removing these trade lines enabled us to compare the different Table A1 compares the samples based on the race. consumers with utility trades are distinguished from those with telecommunications trades. but their total number of trades (excluding utilities and again. In making primarily reflect household heads or individuals living these comparisons. to have somewhat lower incomes and a higher propor- tion of males compared to validation sample. they are more pronounced for consumers with a reported util- Table A2 compares the characteristics of the samples ity. In presenting the statistics on the analysis file. Although there is a small overlap between the two groups. This pattern is not surprising given that the latter on the basis of credit profiles of consumers. samples on an “apples to apples” basis. consumers with utility trades tend to have a higher In general. the three population groups look fairly similar. Table A3 presents the distribution of the samples by state. broader population of consumers in terms of their While consumers with telecommunications trades tend underlying credit profiles. these differences are not pronounced. although some different differences can be observed. telecommunications trade lines from the credit reports 42 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . DEMOGRAPHIC DIFFERENCES of consumers contained in the analysis file.APPENDIX A.

0 100.3 Hispanic 8.0 $50.000–$29.9 10. Pennsylvania (16 percent) and Wisconsin (24 percent.3 2.3 17.6% 1.2 53.8 50.0 Total 100. Distribution of Samples by Demographic Characteristics Consumers with Consumers with Validation Utility Trades (%) Telecommunications Trades (%) Sample (%) Race Asian 3.795 3.5 10.6 $20.2 66+ 18.1 Total 100.3 17.1 19. Consumers with utility trades are concentrated in Illinois (44 percent).4 M 59.8 12.2 49.5 White 67.0 100.0 100.020 590.9 11.1 70.5 30. As expected.1 Other 11.7 12.8 6.5 16.525 GEOGRAPHIC DIFFERENCES Table A3 presents the distribution of the three popula- tion groups by state.8 14.6 Total 100.3 20.000+ 17.9 20.0 100.7% 4.0 11.4 24.5 5.7 34.0 100.0 $100.1 $30.7 2.2% Black 8.1 15. GIVE CREDIT WHERE CREDIT IS DUE 43 .3 36–45 23.6 26–35 15.000–$99.3 46–55 24.8 17.999 18.0 100.0 Income <$20.1 25.0 Gender F 40.5 24. the samples are not representative in terms of their geographic location.3 18.000 17.0 Age 18–25 1.985.6 68.3 Total 100.8 25.519.8 46.999 9.0 100.000–$49.0 100.) The telecommunications sam- ple is also primarily in Pennsylvania (69 percent) and Texas (13 percent).4 21. Appendix Table A1.999 36.0 Sample Size 7.2 9.3 56–65 16.

0 501–560 8.4 741–800 10.4 All Consumers 100 100 100 Sample Size 7.7 11.7 9.4 9.4 4.3 621–680 9. Appendix Table A2.0 13.6 12.2 6 3.985.9 4 3.6 14.9 9.4 5 3.7 3. Distribution of Samples by Credit Profiles of Consumer: Excluding All Utility and Telecommunications Trades Consumers with Consumers with Validation Utility Trades (%) Telecommunications Trades (%) Sample (%) % Distribution by No.522 a The score was obtained by removing the utility and telecommunications trades from the consumer’s credit files 44 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .5 3.1 12.2 681–740 10.3 3.6 8.1 7.6 801–850 10.0 7+ 70.0 9.9 8.3 53.9 All Consumers 100 100 100 % Distribution by VantageScorea 851+ 27.795 3.6 6.1 1 4.4 561–620 10. of Traditional Trades 0 9.1 3.1 5.5 3 3.0 4.519.9 20.9 21.9 13.5 63.3 21.1 3.7 14.2 3.020 590.7 No Score 12.1 3.6 16.2 3.9 2 3.

1 1.6 Alaska 0.0 Michigan 0.1 0.4 0.5 Washington 0.1 0.1 1.0 0.5 1.0 0.0 0.9 68.4 1.0 0.0 0.0 0.985.0 Arkansas 0.1 Iowa 0.3 0.9 Hawaii 0.4 1.7 2.0 1.3 DC 0.0 0.9 New Mexico 0.1 0.2 0.5 1.2 0.2 2.795 3.3 3.0 Kentucky 0.3 Pennsylvania 15.4 0.9 Oklahoma 0.1 1.9 Massachusetts 0.2 0.3 0.0 0.7 Connecticut 3.5 Maryland 0.0 0.3 2.6 0.1 0.0 Missouri 0.0 California 0.0 0.0 Montana 0.6 0.2 Utah 0.2 1.7 0.2 0.2 Florida 1.7 3.7 Colorado 0.9 2.6 12.2 Tennessee 0.4 Idaho 0.6 New York 0.3 Wyoming 0.1 Delaware 0.9 2.4 Louisiana 0.6 Maine 0.3 Indiana 0.2 1.0 North Dakota 0.3 3.0 0.020 590.2 Virginia 0.2 Sample Size 7.5 1.0 0.0 0.1 0.3 3.1 1.0 0.0 0.8 12.1 1.0 0.6 Nevada 0. Appendix Table A3.0 0.2 Ohio 0.7 6.0 0.0 0.7 Mississippi 0.0 0.1 1.3 1.0 Kansas 0.0 0.2 0.1 0.2 0.0 1.4 Illinois 44.0 0.2 0.1 1.0 0.9 New Hampshire 0.8 0.3 North Carolina 0.522 GIVE CREDIT WHERE CREDIT IS DUE 45 .6 8.3 Nebraska 0.1 0.4 1.2 6.519.1 0.8 Vermont 0.0 0.0 0.0 Texas 0.7 3.3 South Carolina 1.1 0.0 0.0 0.4 New Jersey 0.6 Wisconsin 23.6 Georgia 0.6 0.0 0.1 0.3 0.0 0.0 0.1 0.4 South Dakota 0.7 2.2 2. Distribution of Samples by State Consumers with Consumers with Validation State Utility Trades (%) Telecommunications Trades (%) Sample (%) Alabama 0.1 2.1 1.5 Minnesota 0.2 No Data 0.0 0.0 0.1 0.3 Oregon 0.7 Rhode Island 0.1 1.1 2.1 0.2 Arizona 2.4 2.1 1.3 West Virginia 0.

9 1. Excluding Unscoreables Consumers with Utility Consumers with Telecom Trades Trades Including Excluding Including Excluding Utilities Utilities Telecom Telecom Acceptance Rate (#1) (%) (#2) (%) (#1) (%) (#2) (%) 30 0.0 61.5 50 1.0 6 75.6 53.1 58.2 2.8 8.3 1.7 2.2 63.8 57.8 5 72.3 1.3 65.7 7 78.5 68.5 1.1 9.1 40 1.7 4.3 46.0 0. Acceptance Rates by Targeted Delinquency Rates: VantageScore.1 18.8 Source: PERC Appendix Table B2.9 18.1 1. D E TA I L E D M O D E L R E S U LT S Appendix Table B1.0 64.9 13.4 60 2.4 3 64.7 4 69.7 54.8 53.8 Source: PERC 46 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .7 14.3 77.9 48.9 1. Excluding Unscoreables Consumers with Utility Consumers with Telecom Trades Trades Including Excluding Including Excluding Utilities Utilities Telecom Telecom Acceptance Rate (#1) (%) (#2) (%) (#1) (%) (#2) (%) 2 56.4 2.5 90 12.2 4.APPENDIX B.1 80 7.9 72.4 4.5 9.7 70 4.5 66. Serious Delinquencies by Target Acceptance Rates: VantageScore.9 75.8 61.4 62.1 14.

7 Source: PERC GIVE CREDIT WHERE CREDIT IS DUE 47 .5 50 1.7 80 8.4 5 66.5 13.2 14. Acceptance Rates by Targeted Delinquency Rates: TransRisk New Account Model Consumers with Utility Consumers with Telecom Trades Trades Excluding Excluding Delinquency Rate All Trade Utility Trades All Trades Telecom Trades All 2 50.2 1.7 4 62.7 57.5 7 77.5 45.1 52.5 74.9 44.7 90 13.0 62.6 62.1 70 5.9 1.4 2.3 55.2 48.9 10.1 59.3 3.2 57.2 8.0 70 4.1 5 71.0 1.9 10.0 45.8 14.7 47.7 17.5 64.9 21.0 1.5 50.8 Excluding Unscoreables 2 56.8 3 57.1 7.0 61. Appendix Table B3.1 54.0 7 73.2 2.5 59.8 20.8 56.1 50 1.2 1.3 68.9 53.9 1.8 60.9 Excluding Unscoreables 30 0.5 67.8 6 70.9 54.6 70.2 19.1 40 1.9 5.9 7.1 1.5 90 13.9 6 74.8 15.7 9.2 38.9 2.1 1.5 60 2.4 4.5 1.9 80 9.7 4.0 Source: PERC Appendix Table B4.7 9.1 13.8 2.5 1.6 60 3.7 4 68.8 2.8 4.0 53.6 65.0 76.2 41.6 48.7 50. Serious Delinquencies by Target Acceptance Rates: TransRisk New Account Model Consumers with Utility Consumers with Telecom Trades Trades Excluding Excluding Acceptance Rate (%) All Trade (%) Utility Trades (%) All Trades (%) Telecom Trades (%) All 30 0.3 1.3 1.1 58.8 53.4 14.3 40 1.7 3 64.3 19.5 4.6 17.4 62.6 64.

0 84.7 73.5 67.41 0.70 0.5 68.02 1.0 74.9 58.09 0.07 0.5 79.12 0.06 0.28 0.17 1.07 50 0.7 63.07 0.8 69.07 0.0 79.75 90.1 Excluding Unscoreables 0.60 1.25 0.7 83.22 0.1 0.07 0.11 0.8 94.7 1.38 0.7 0.13 60 0. Appendix Table B5.83 80 0.11 50 0. Acceptance Rates by Targeted Bankruptcy Rates: TransRisk Bankruptcy Model Consumers with Utility Consumers with Telecom Trades Trades Excluding Excluding Bankruptcy Rate (%) All Trade (%) Utility Trades (%) All Trades (%) Telecom Trades (%) 0. Bankruptcies by Target Acceptance Rates: TransRisk Bankruptcy Model Consumers with Utility Consumers with Telecom Trades Trades Excluding Excluding Acceptance Rate (%) All Trades (%) Utility Trades (%) All Trades (%) Telecom Trades (%) 30 0.27 70 0.75 90 0.06 0.3 58.21 0.38 0.00 96.25 60 0.3 90.9 86.3 84.06 0.6 50.06 0.06 0.76 Excluding Unscoreables 30 0.50 83.52 70 0.50 85.25 72.9 60.4 88.07 40 0.25 72.9 0.74 0.07 0.08 0.4 59.12 0.11 0.06 0.07 40 0.2 Source: PERC 48 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .41 0.9 80.6 1.28 1.0 0.14 0.44 90 0.06 0.41 0.07 0.50 80 0.19 0.00 94.1 71.06 0.3 74.52 0.75 0.69 1.18 Source: PERC Appendix Table B6.74 1.0 86.75 90.21 0.

64 0.11 0.12 50 0.27 70 0.04 0.07 0.03 0.23 0.22 0.25 72 71 58 59 0.67 0.56 1.03 Source: PERC Appendix Table B8.00 1.69 0.05 40 0.40 0.07 0. Acceptance Rates by Targeted Bankruptcy Rates: Bankruptcy Model II Consumers with Utility Consumers with Telecom Trades Trades Excluding Excluding Bankruptcy Rate (%) All Trades (%) Utility Trades (%) All Trades (%) Telecoms Trades (%) 0.06 0.68 90 0.23 90 0.50 84 74 77 61 0.50 85 84 73 72 0.06 0.13 0.25 60 0.03 0.40 0.15 60 0.13 0.04 0.39 0.70 1.46 80 0.12 0.47 70 0.21 0.14 0.76 80 0.75 93 92 83 82 1. Bankruptcy Rates by Target Acceptance Rates: Bankruptcy Model II Consumers with Utility Consumers with Telecom Trades Trades Excluding Excluding Acceptance Rate (%) All Trades (%) Utility Trades (%) All Trades (%) Telecoms Trades (%) 30 0.06 50 0.06 0.03 0.45 0.08 0.75 91 81 86 70 1.07 0.03 40 0. Appendix Table B7.04 0.23 0.14 0.24 0.76 Excluding Unscoreables 30 0.03 0.03 0.90 1.29 0.39 0.00 97 97 90 89 Source: PERC GIVE CREDIT WHERE CREDIT IS DUE 49 .37 0.28 0.66 1.00 95 86 92 76 Excluding Unscoreables 0.25 71 62 62 50 0.

00 36–45 1.05 1.04 1.06 1.00 1.00 1.00 1.00 1.000 1.00 Age 18–25 1. Appendix Table B9.02 1.00 1.02 1.02 1.999 1.00 26–35 1.999 1.09 1.02 1.02 1.00 $100.03 1.00 $30.00 1.07 1.00 1.000–$49.00 M 1.08 1.00 Gender F 1.00 Income <$20.05 1.00 Source: January 31.00 56–65 1.00 1.00 66+ 1.02 1.00 1.00 Black 1.00 White 1.000–$99. Impact on Acceptance Rates by Demographic Group (TransRisk New Account): (Assumes 3% Serious Delinquency Rate) Consumers with Utility Consumers with Telecom Trades Trades Including Excluding Including Excluding Utilities (#1) Utilities (#2) Telecoms (#1) Telecoms (#2) All Consumers 1.02 1.04 1.00 1.000+ 1.00 1.03 1.00 46–55 1.00 Race Asian 1.00 $20.00 1.00 1.08 1.04 1.02 1.000–$29.05 1.04 1.04 1.03 1.03 1.01 1.03 1.04 1.05 1.00 1.00 1.03 1.03 1.00 1.03 1.00 Other 1.00 1.00 $50.06 1. 2005 Credit Files for Analysis sample 50 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .999 1.03 1.04 1.00 Hispanic 1.03 1.00 1.00 1.03 1.03 1.

Appendix Table B10.00 1.000–$99.00 $30.00 1.00 Gender F 1.06 1.000–$49.00 1.13 1.36 1.03 1.00 1.19 1.32 1. Impact on Acceptance Rates by Demographic Group (TransRisk Bankruptcy): (Assumes 0.00 White 1.00 GIVE CREDIT WHERE CREDIT IS DUE 51 .02 1.00 $50.12 1.00 $100.17 1.00 26–35 1.06 1.10 1.06 1.00 1.17 1.12 1.07 1.000 1.18 1.21 1.999 1.00 Hispanic 1.06 1.00 1.00 Age 18–25 1.00 56–65 1.08 1.51 1.11 1.67 1.14 1.05 1.00 1.00 1.00 1.43 1.00 1.00 Income <$20.00 Black 1.00 1.70 1.25% Bankruptcy Rate) Consumers with Utility Consumers with Telecom Trades Trades Including Excluding Including Excluding Utilities (#1) Utilities (#2) Telecoms (#1) Telecoms (#2) All Consumers 1.24 1.00 66+ 1.00 Other 1.00 1.09 1.08 1.09 1.16 1.999 1.39 1.00 $20.00 36–45 1.13 1.00 46–55 1.00 M 1.00 2.00 1.00 1.00 1.07 1.00 Race Asian 1.12 1.000–$29.999 1.00 1.000+ 1.16 1.00 1.09 1.18 1.00 1.

10 1.09 1.00 1.54 1.00 Source: PERC 52 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E .06 1.00 $100.13 1.00 Gender F 1.999 1.000–$99.10 1.19 1.000+ 1.000–$29.00 46–55 1.00 1. Appendix Table B11.00 1.00 Income <$20.00 $20.00 2.13 1.00 Black 1.25 1.19 1.69 1.32 1.00 White 1.29 1.999 1.00 Age 18–25 1.16 1.03 1.00 56–65 1.00 1.00 66+ 1.00 1.00 Race Asian 1.14 1.10 1.14 1.000 1.21 1.00 $50.00 1.00 36–45 1.00 1.10 1.04 1.00 M 1.000–$49.07 1.25% Bankruptcy Rate) Consumers with Utility Consumers with Telecom Trades Trades Including Excluding Including Excluding Utilities (#1) Utilities (#2) Telecoms (#1) Telecoms (#2) All Consumers 1.00 Hispanic 1.07 1.21 1.36 1.14 1.18 1.00 1.08 1.40 1.19 1.24 1.00 1.06 1.00 1.00 $30.58 1.27 1.00 1.00 1.00 1.07 1.999 1.00 26–35 1.00 1.00 1. Impact on Acceptance Rates by Demographic Group (Bankruptcy Model II): (Assumes 0.00 1.00 1.17 1.09 1.12 1.00 Other 1.00 1.

For more information.to late 2005.” Although FICO does not reveal the underly. June 2003). one should focus on the results that show that conduct the demographic analysis. See “The Fair Credit Reporting Act: Access. 12. Change: New Ways of Moving Markets” (Washington: Brookings Institution. that fail to receive a traditional FICO score due to non-existent or ‘thin’ credit histories.g. The Fair Credit Reporting Act: 7. To gauge the longer-term effects of bringing 9. they have no established credit and have never probability for each score (consumers would be resorted). Fair Isaacs & the consumer’s credit report and contains no information on the Co. The mortgage screening model is based entirely on data found in scoring models for thin-file borrowers. GIVE CREDIT WHERE CREDIT IS DUE 53 . available at July 2005) http://www. Although most scoring models use a 24-month performance period. and Opportunity” (Washington: Information Policy Institute. National Chamber Foundation. July 2005.” Information Policy Institute. ethnic minorities. Dana Nottingham opening keynote speech at the 2006 UMI of deleting certain types of derogatory data from consumers’ Forum credit files. Giving Underserved Better Access to the Credit System: The Promise Access. our sample will exclude many individuals who now have a reported utility or telecommunications trade. consumers decision tool.S. DC: The of Non-Traditional Data. “Using Information to Drive models instead treat these trades as general trades. see www.pdf 8.. See effects of bringing new data online. and do not have this sort of the addition of the new data helps to better sort consumers by risk. sociodemographic data in their credit files. 2003). See Michael Turner. The 3. TransUnion and financial institutions providing the scores did not new data online. nontraditional credit data generally capture we used a 12 month period to capture a larger number of consumers the consumer’s performance on obligations such as rent-to-own with an established telecommunications or utility trade at the begin- agreements. and we wanted to capture as many consumers as possible. for utility or telecommunications data. et al.org/pdf/fcra_report. the score “can used as an initial screen to process loans. something that will undoubtedly occur as the reporting of such data increases. 2005. Efficiency. and the old. Our approach was similar to one employed in an earlier Information Policy Institute study. as opposed to credit effectively predict risk for the growing number of U. Since some consumers are not included in credit some probability of default.. “Giving Underserved Consumers Better Access June 2003). New York City.com/ ning of the performance period (March 31. past two years. None of the models in this study has been optimized specifically young. Even so. the been reported by a collection agency). which examined the impact 4. loan-to-value ratio). Michael A. (FICO) recently introduced a FICO Expansion Score using characteristics of the mortgage itself (e. We find that better sorting leads to increased access to credit. Credit providers and commercial scoring firms have also developed 13. 6. 5. and the new data would change this bureau files (e. Such changes in the actual value of the scores are short-run lion randomly selected consumers in the validation sample. 11. For example. According to Fair Isaacs. par- ticularly among low-income consumers. Pari Sabety and Virginia Carlson. addition of the new data meant the same thing as a score of 700 with the new data. 14. (New York: Information Policy Institute.. ing drivers of its score.ENDNOTES 1.) The number of fairisaac/solutions/FICO+Expansion+Score/Expansion+Score+ providers reporting such trades has increased significantly in the Overview. This statistic is based on the credit records of approximately 4 mil- 2. to the Credit System: The Promise of Non-traditional Data. the 10. (Washington. Efficiency and Opportunity.g. Turner. It is nontraditional credit data. 13 percent is most likely a scores would need to be rescaled so that a score of 700 before the lower bound estimate. Because the scores represent Appendix A. TransRisk Bankruptcy model.fairisaac.infopolicy. because many wireless companies began reporting in mid.

The results 20. the performance measure used to assess the impact of the nontrad. In assessing the impact on the model.pdf be scored with or without their utility and telecommunications trades are presented in Appendix B. any control sample would be restricted to consumers with observed for these models are similar to those described in an established. with the addition of the alternative utilities and 21. it was impossible to determine the extent to which con- sumers in the validation sample have such accounts. 2005. 19. Therefore.pdf utility and the telecommunications trades on a given model was the same as that used to construct the model. we also ran a logit timate the marginal impact of reporting utility and telecommunica. For a given acceptance rate. the marginal 22. and March 31. generally have stronger credit profiles than consumers with quency on a new account between April 1. we take this performance as the floor of what we model’s K-S statistic are considered significant by model develop- should expect from models reoptimized for this data. assumes that consumers who have such accounts have stronger credit profiles than those who do not. (New York: The Information Policy Institute. Results when the calculations are limited to consumers who can www. con- reflected in the consumer’s total number of trades. Ideally. See Michael Turner et al. Although they are treated as a single category als who had scores with and without the alternative data (utility or in Table 1. therefore. their current credit profiles (Columns 1 and 3 in Table 2) will include at least one reported trade line. These calculations are based on subsamples consisting of individu- munications trade. the 31. For example.org/pdf/ differs. telecommunications data. However. 27. Thus. Efficiency & Opportunity Part II. 17. In general.001. 29. 16. we tent with our earlier finding that consumers in the utility sample based our analysis on the occurrence of at least 90-day delin. sisted of individuals with at least one traditional trade. a consumer had to have at 26. the text.infopolicy. increases of more than 10 percentage points in a data. The performance measure used to assess the accuracy of a given impact of the utility and telecommunications trades is considerably model was geared to the specific purpose of that model. These sample sizes correspond to the VantageScore model since trades and traditional trades. (Nagelkerke R-Squared) rose by 40 percent and 17 percent. Thus. scoreablity differs across models. available at 30. multiple utility or telecommunications accounts are telecommunications trades). we limited the performance period to 12 months to capture as many consumers as possible in our analysis file. a better job separating the good risks from the bad risks with the inclusion of the alternative 23. while our performance period Institute. This pattern is consis- over a two-year period. available online www. The nonoptimized models do. July 2005). our comparisons may overes. 24. generally.15. as well as for thin-file consumers and CRA loans. although smaller when this restriction is imposed. Some consumers have more than one reported utility or telecom. Total number of trades includes both the number of alternative 25. telecommunications trades. by definition. September 2002). In general.org/pdf/institute_fcra_ptII. our assessment of two bankruptcy models was 54 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . Likewise. scoreablity differs across models.infopolicy. These subsamples. The Fair Credit Reporting Act: Access. ers. Because the dependent variable is dichotomous. The p-values were less than . the rate of serious delinquencies that new account model is designed to predict the probability that a is observed for consumers with utility trades is lower than it is for consumer will experience a 90-day delinquency on a new account consumers with telecommunications trades. respectively. 2006. The lender also has screening models designed for refinancing. 18. These sample sizes correspond to the VantageScore model since least one fully reported utility or telecommunications trade. Giving Underserved Better Access to the Credit System: The based on the number of consumers who experienced a bankruptcy Promise of Nontraditional Data (New York: Information Policy within the observation period. but unreported telecommunications or utility trade. In order to be selected for our sample. If one 28. regression and found that the goodness-of-fit of the models tions trades.

“TransUnion Case Study: How Reporting Helped Nicor Gas Reduce Bad Debt” (Chicago: TransUnion. 33. GIVE CREDIT WHERE CREDIT IS DUE 55 . 34. The socio-demographic data appended to the indivdual credit files were generated by Acxiom from a combination of data sources including. 37.htm 38. TransUnion. Its impact is limited by the number of people who utilize its service and by the fact that individuals can choose which payment infor- mation they would like to include. “TransUnion Case Study: How reporting helped Nicor Gas reduce bad debt. self-reported sources.” Presented at CFED Asset Building Conference. 2002) 35. “Where is the Asset Building Opportunity?: A Profile of Credit Utilization and Management in 50 Metros. and public record information. 2006. 36. It is likely that consumers will want to only include the payment histories which paint them in the best light. See appendix B for results for additional results based on other score models. Phoneix. TransUnion.brookings. AZ. see www.32. estimates from some of the indi- vidual’s characteristics. extrapolation from census data. Matthew Fellowes. Il: TransUnion. 2002). thus biasing the picture of themselves they present to users of the information. For a complete transcript of the event.edu/metro/umi/events/20051215_paid. Sept.” (Chicago.

A.A. Ph. Dr. M. Ann Schnare. Walker is currently completing his dissertation at Duke University. Ph.A. Lee worked with the Consulting Group of Cushman and Wakefield where she specialized in crafting workforce and location investment solutions for corporate and public organizations through the alignment of workforce. business process.D. Ph. is Director of Research at the Political and Economic Research Council. the GAO. the FTC. Walker’s concentration is econometrics and statistical methods. Patrick D. currently serves as President and Senior Scholar of the Political and Economic Research Council. Ms. While at Freddie Mac. Ms. Lee holds a Master of City and Regional Planning with distinction from the Georgia Institute of Technology. Dr.D.D. the White House (CEA and NEC) on technology and consumer credit issues. Schnare received her Ph. in Political Science from Columbia University in 2004. where he has taught both undergraduate microeconomics and econometrics. Walker serves as Fellow.D. 56 P O L I T I CA L A N D E C O N O M I C R E S E A R C H C O U N C I L • T H E B R O O K I N G S I N S T I T U T I O N U R B A N M A R K E T S I N I T I AT I V E . Walker. After serving as a Graduate Fellow at the Columbia Institute for Tele-Information (CITI) at the Columbia Business School in New York City.D. is an Adjunct Fellow at the Political and Economic Research Council.. Dr. in economics from Washington University. Dr. Before establishing her own consulting practice. and currently serves on the DHS’s Data Privacy and Integrity Advisory Committee. Turner. He received his M. in economics from Duke University. Economic Policy Analysis at the Political and Economic Research Council. Prior to join- ing Brookings. and pre- sented studies to a host of government agencies including the FCC. Housing Economics and Financial Research. in Economics from Harvard University in 1974 and her B. Corporate Relations and as Vice President. Turner has testified before Congress and numerous state legislatures. Schnare worked for Freddie Mac as Senior Vice President. Robin Varghese. Dr. strategy and location to achieve optimal enterprise value. Varghese was a senior researcher and Graduate Fellow at the Columbia Institute of Tele-Information (CITI). he was named Executive Director of the Information Services Executive Council (ISEC). Varghese received his Ph. she was responsible for developing research and the Congressional and public relations as the firm adopted the use of credit scoring for mortgages. She received a double Bachelor of Arts in Sociology and Urban Studies from Northwestern University. Alyssa Stewart Lee is Acting Director of the Urban Markets Initiative of the Brookings Institution.A B O U T T H E AU T H O R S Michael A.

. Living Cities: The National Community of our supporters.AC K N O W L E D G E M E N T S A lthough the authors benefited tremendously The Urban Markets Initiative thanks its founding from the shared insights and expertise of all funder. . In particular. Inc. the Hispanic National Mortgage Association. and JP Morgan Chase—the ideas and opinions expressed herein remain strictly those of the authors from the Information Policy Institute at PERC and the Urban Markets Initiative at the Brookings Institution. as well as expert analytical urban communities. Although the study was enriched by the input from dozens of oth- ers in addition to those already mentioned—most notably executives from Bank of America. our silent partner in this effort. and Barbara Anthony at the SAS Metropolitan Policy Program Institute provided invaluable assistance by verifying The Brookings Institution and validating the statistical rigor and analytical soundness of our research program. GE Money. Dr. and the SAS Institute. in Information Policy Institute at the Political and particular Jennifer Barrett and Sheila Colclasure. and the federal government TransUnion provided the anonymized files used in committed to improving the vitality of cities and the report’s simulations. and Jason Wright worked tirelessly on many aspects of this project. Development Initiative. Bob Ryan. Urban Markets Initiative Sunny Zhang. Equifax. pro. the quality of the study would have suffered greatly. Experian. John Carter at Equifax taught us Alyssa Stewart Lee that credit scores do not always go up when positive Acting Director payment data are added. President and Senior Scholar graphic analysis. Chet Wiermanski. we are particularly grate. HSBC. Without their assistance. Corporation. Acxiom For More Information please contact: Corporation. financial institutions. Clark Abrahams. services that greatly assisted PERC’s overall effort. Living Cities is a partner- ful to the staff at TransUnion Corporation. Economic Research Council many of the key findings in the study would have remained hidden. the Consumer Data Industry Association. Acxiom ship of leading foundations. Michael Turner vided a data append that enabled the sociodemo. Without Acxiom’s assistance. Finally. nonprofit organizations.

1732 www.797.656. 20036-2188 Tel: 202.brookings.797.C.brookings. 919.2798 • Fax 212.6000 • Fax: 202.infopolicy.6004 www. 202.org .797.The Brookings Institution 1775 Massachusetts Avenue.edu Direct: 202.htm Information Policy Institute.4375 • Fax 202.2965 www.6139 • Fax/direct: 202.edu/metro/umi.797.797.338.6516 www. NW • Washington D. Political & Economic Research Council Tel.brookings.741.edu/metro Urban Markets Initiative Tel.