Credit Scoring and Loan Scoring - Thomas H. Stanton

Grant ReportJuly 1999Credit Scoring and Loan Scoring:Tools for Improved Managementof Federal Credit ProgramsThomas H. StantonFellowCenter for the Study of American GovernmentJohns Hopkins UniversityThe PricewaterhouseCoopers Endowment forThe Business of Government

The PricewaterhouseCoopers Endowment forThe Business of GovernmentAbout The EndowmentThrough grants for research, conferences, and sabbaticals,the PricewaterhouseCoopers Endowment for the Business ofGovernment stimulates research and facilitates discussion onnew approaches to improving the effectiveness of governmentat the federal, state, local, and international levels. Allgrants are competitive.Founded in 1998 by PricewaterhouseCoopers, theEndowment is one of the ways that PricewaterhouseCoopersseeks to advance knowledge on how to improve public sectoreffectiveness. The PricewaterhouseCoopers Endowmentfocuses on the future of the operation and management ofthe public sector.

Credit Scoring and Loan ScoringTools for Improved Managementof Federal Credit ProgramsThomas H. StantonFellowCenter for the Study of American GovernmentJohns Hopkins UniversityJuly 1999Credit Scoring and Loan Scoring 1

TABLE OF CONTENTSForeword ......................................................................................3Executive Summary ......................................................................4I. Introduction and Overview ................................................6II.III.IV.Credit Scoring and Loan Scoring in the Private SectorA. Credit Scores and Loan Scores........................................8B. Conducting Controlled Experiments..............................10C. Automated Scoring-Based Systems................................10Strategic Implications of the New TechnologiesA. Implications for Borrowers............................................12B. Implications for Financial Services Providers ................13C. Implications for Federal Credit Programs ......................14Public Policy Issues Relating to Adoption of Scoringby Federal Credit Agencies ................................................18V. Current Use of Scoring-Based Systems by FederalCredit Agencies..................................................................20VI.Options for Additional Applications to FederalCredit ProgramsA. Scoring-Based Financial Early Warning Systems ..........22B. Adding Credit Scores or Loan Scores to LenderMonitoring Systems ......................................................23C. Scoring to Improve Servicing of FederalDirect Loans ................................................................24D. Scoring to Improve Targeting of FederalCredit Programs............................................................25VII. Conclusion: The New World of Scoring-Based Systems ..27Appendix: Scoring: Where to Begin ..........................................30About the Author ......................................................................31Key Contact Information ............................................................322 Credit Scoring and Loan Scoring

The PricewaterhouseCoopers Endowment forThe Business of GovernmentForewordOn behalf of The PricewaterhouseCoopers Endowment for the Business of Government, we are pleased topublish our second grant report. Thomas Stanton’s report on credit scoring comes at an opportune time.Credit scoring is an important application of technology to the business of government. It allows creditors,such as federal agencies which make loans, to evaluate millions of applicants consistently and impartiallyon many different characteristics. What once took weeks and a great deal of judgment is now completedin minutes, using data and objectivity. Similarly, more creditworthy individuals can be identified andexpedited through this process.The government is one of the largest originators of loans. Unfortunately, it often does not collect on theseloans. The use of credit scoring would enable the government to better target and collect on these loans.In an era of tightening budgets, credit scoring is a tool that will enable government to increasingly choosewisely and also enable it to become a better financial manager.As the availability of credit increases, the use of credit scoring is also likely to increase. Federal creditprograms must not ignore the need to adapt credit-scoring technology as a means to conduct businessin a more efficient and objective manner. As Mr. Stanton points out in this report, information-based technologiescreate both opportunities and risks for federal credit programs. The recommendations containedin this report will shed light on how federal credit agencies can use this technology to mitigate the riskassociated with lending. We hope you will find this report informative and helpful.Paul LawrencePartner, PricewaterhouseCoopersCo-Chair, Endowment Advisory Boardpaul.lawrence@us.pwcglobal.comIan LittmanPartner, PricewaterhouseCoopersCo-Chair, Endowment Advisory Boardian.littman@us.pwcglobal.comCredit Scoring and Loan Scoring 3

Executive SummaryNew information-based technologies are transformingthe credit markets at a rapid pace. Lenders usecredit scores and loan scores to generate loan-levelinformation about a borrower’s propensity to repaya particular loan. They measure the borrower’screditworthiness against a representative database,and use the resulting information to decidewhether a loan should be made and, increasingly,on what terms.Once the loan is made, lenders and servicers usescoring to help determine the most effective servicingand collection techniques to deal with a borrowerwho is delinquent or in default on a loan.Finally, credit scores and loan scores help lendersto assess risk and decide which loans and loanportfolios to securitize or otherwise sell, and helpto price the sales transaction. When linked to datamanagement systems, scoring-based systems allowlenders to originate and service high volumes ofloans with unprecedented speed and accuracy.Credit scoring and loan scoring create both opportunitiesand risks for federal credit programs. On theone hand, federal direct loan and loan guaranteeprograms can adopt some of the new technologiesto improve their own credit administration. On theother hand, in today’s environment the governmentwill lag the private sector in resources and generalcapacity to adopt new information-based systems.This increases the prospect for adverse selection asprivate lenders use credit scoring and loan scoringto serve an increasing number of creditworthyborrowers who formerly would have been borrowersin a federal program.This report presents four recommendations concerningways that federal credit programs can usescoring to help manage credit risk:1. Major federal credit programs that involve loansof a type for which scoring is suitable [especiallyincluding the single-family mortgage programs ofthe Federal Housing Administration (FHA) andthe Department of Veterans Affairs (VA)] shouldcreate scoring-based systems to provide earlywarning of a deterioration in the credit quality ofloans being originated under the program.2. Major federal guarantee programs, including thesingle-family mortgage programs of FHA, VA,the Rural Housing Service, and Ginnie Mae,and the guaranteed loan programs of the SmallBusiness Administration (SBA) and Departmentof Education, should adopt loan-level scoringsystems to monitor the quality of performanceof institutions that originate and service theirguaranteed loans.3. Federal direct loan programs, including federaldirect student loans, the SBA disaster loan programs,and rural housing direct loans, shoulduse scoring to help conduct controlled experimentsin improved approaches to loan servicing.4. Federal credit programs including the FHA single-familyand SBA business loan programsshould use credit scoring and loan scoring toexperiment with improved targeting of creditworthyborrowers for whom traditional creditscores may be inappropriate.Rapid deployment of scoring-based systems in theprivate sector means that some federal programs4 Credit Scoring and Loan Scoring

may be at risk if they continue to do business inthe old ways. The report presents three recommendationsfor dealing with this change in a strategicenvironment:• Federal policymakers, especially at the Office ofManagement and Budget, should encourageappropriate federal credit agencies to makemultiyear commitments of resources to adoptscoring-based systems in the context of welldesignedstrategic plans.• Federal credit agencies should devote neededresources to assuring that they remain wellinformed about technological developments andthe implications for the markets in which theirprograms operate.• Federal policymakers in the Executive Branchand the Congress should consider structuralchanges to credit programs and organizations toincrease their flexibility and capacity to respondto the many technology-driven changes thataffect their ability to continue to serve theirpublic purposes.Credit scoring and loan scoring offer many possiblebenefits, but also raise public policy concerns thatmust be addressed. Federal credit agencies alreadyhave begun to partner with private lenders to applyscoring-based systems to the origination and underwritingof government-insured or guaranteed loans.The Federal Housing Administration, Department ofVeterans Affairs, and Small Business Administrationare incorporating scoring into the loan originationprocess. The VA also is conducting an experimentwith the application of scoring-based systems toloan servicing. Other federal credit agencies arelikely to follow soon.Other important applications include the use ofscoring-based systems to develop or enhancefinancial early warning systems, to monitor lenderperformance, to improve the targeting of federalcredit to the most appropriate borrowers, and toexperiment with lending to subgroups of underservedborrowers in an effort to increase theiraccess to credit.Not all loans benefit from credit scoring or loanscoring. If the probability of default varies largelybased upon factors other than the borrower’s individualcredit, then credit scores do not add muchvalue to credit administration. If loan data are notstandardized, or if sound historical data are unavailable,then loan scores will lack predictive value.Thus, until someone develops a sound databaseand a model with predictive value based upon thatdatabase, loan scoring will not be useful in makingor supervising federally guaranteed loans for FHAfinancedmultifamily rental properties, for example.By contrast, FHA single-family loans are scorablebecause information relating to defaults is standardizedand available from many years of experience.If they lack the requisite data, federal agenciesmay need to adopt and adapt off-the-shelf scoresand scoring systems to their own program needs.Conversely, government agencies are at specialrisk of making misjudgments if they apply privatesectorscorecards to subgroups of borrowers intheir programs who act differently from the generalpopulation for which the scorecards may beappropriate.Ultimately, the application of new technologiesmay require some federal credit agencies to developmore flexible organizational structures and programsif they are to continue to serve their publicpurposes. Leaders of some federal credit agencies,and especially those that benefit from significantexternal support, may need to begin a process ofdialogue with stakeholders as a way to begin toenlist their participation in a consensus that mightbe built around the need to improve the design ofsome programs and organizations.Policymakers may need to redesign the form ofsome government programs so that they can complementrather than be undercut by a dynamicprivate sector. Government organizations themselvesmay need to become much more nimble ifthey are to continue to serve their public purposeseffectively in today’s rapidly changing environment.In the provision of credit, federal programs haveneither the resources nor the policy freedom tooperate at the leading edge of available technologies.On the other hand, the adoption of new practicesin the private sector ultimately may make ituntenable for some federal programs to continuedoing business in the old ways.Credit Scoring and Loan Scoring 5

I. Introduction and OverviewNew information-based technologies are transformingthe credit markets at a rapid pace. Lenders usecredit scores and loan scores to generate loan-levelinformation about a borrower’s propensityto repay a particular loan. They measure theborrower’s creditworthiness against a representativedatabase, and use the resulting informationto decide whether a loan should be made and,increasingly, on what terms.Once the loan is made, lenders and servicersuse scoring to help determine the most effectiveservicing and collection techniques to deal witha borrower who is delinquent or in default on aloan. Finally, credit scores and loan scores helplenders to assess risk and decide which loans andloan portfolios to securitize or otherwise sell, andhelp to price the sales transaction. When linkedto data management systems, scoring-basedsystems allow lenders to originate and servicehigh volumes of loans with unprecedented speedand accuracy.This report will examine the development andapplication of credit scoring and loan scoring byprivate lenders and the relevance of those developmentsto federal credit programs. The reportconcludes that information-based technologiescreate both opportunities and risks for federalcredit programs. On the one hand, federal directloan and loan guarantee programs can adopt someof the new technologies to improve their own creditadministration. On the other hand, in today’senvironment the government will lag the privatesector in resources and general capacity to adoptnew information-based systems. This increases theprospect for adverse selection as private lendersuse credit scoring and loan scoring to serve anincreasing number of creditworthy borrowers whoformerly would have been borrowers in a federalprogram.In other words, the waves of new information-basedsystems have created a sort of arms race. Federalcredit programs cannot rest upon the status quo;they must adopt new technologies and approachesmerely to protect their current positions. This reportsuggests a number of specific uses of credit scoringand loan scoring and related information-based systemsthat can help federal agencies administer theirloan and guarantee programs.Federal programs differ from private businesses insignificant respects, and the process of adoptingprivate-sector practices must be done selectively.Ultimately, the application of new technologiesmay require some federal credit agencies todevelop more flexible organizational structuresand programs if they are to continue to serve theirpublic purposes.Some large federal credit programs, such as FHAsingle-family mortgage insurance and federal smallbusiness loans, began in the aftermath of the GreatDepression. These programs served as pioneerswhose success in making new kinds of loans tocreditworthy borrowers could demonstrate toprivate lenders that the market in such loans wasviable and profitable. The new information technologieshold out the possibility that federal credit6 Credit Scoring and Loan Scoring

programs once again can take on a pioneering role,by helping to serve those creditworthy but underservedborrowers who have been left behind intoday’s highly efficient credit markets.This research report is organized as follows:Section I is the introduction. Section II discussescredit scoring and loan scoring as used by lendersand mortgage insurers in the private sector. Thissection provides an assessment of the usefulness ofscoring-based systems for originating and servicingloans, monitoring lender performance, and otherpurposes. The benefits of scoring will vary accordingto the type of loan program and the nature ofavailable data about borrowers, loans, and repaymentexperience.Section III looks at larger strategic issues raisedby the dramatic increase in the use of scoringbasedsystems. Borrowers, providers of financialservices, and government credit programs all willbe affected. For some federal credit programs, thenew technologies are likely to place organizationalstructures under stress, as some of their existingfunctions become outmoded. Other programs maybegin to experience deterioration in credit qualityas new technologies accelerate the process ofadverse selection by private lenders.Section IV reviews policy issues relating to creditscoring and special considerations for federal creditagencies as they consider adopting scoring-basedsystems for purposes of loan administration. Inparticular, some federal credit agencies have beenwary because of concerns that use of a scoringbasedsystem might adversely affect minorities orother disadvantaged borrowers.Section V surveys the current use of scoring byfederal credit programs. The Federal HousingAdministration, Department of Veterans Affairs,and the Small Business Administration have begunto incorporate scoring into the loan originationprocess. The VA also is conducting an experimentwith the application of scoring-based systems toloan servicing. Other federal credit agencies arelikely to follow soon.permit federal credit agencies to develop new diagnosticand analytic capabilities. A federal creditagency could construct a financial early warningsystem to assure that adverse selection by privatelenders was not creating unacceptable levels offinancial risk in the new loans being originated forits programs. Another use would be to help federalagencies to monitor the performance of lenderswith respect to the credit quality of loans that theyoriginate or service for federal guarantee programs.For some programs, credit scoring can improvecost-effectiveness by helping to target underservedbut creditworthy borrowers who are most likely tobenefit from access to federal credit.Section VII is the conclusion of this report: Creditscoring and loan scoring are here to stay. Each federalcredit agency and its stakeholders must determinethe extent that scoring-based systems arechanging their strategic environment and how theyshould address the new risks and opportunities.The section also presents some recommendationsfor federal policymakers and program officials.Finally, a brief appendix offers some suggestions forfederal managers who may want to explore theapplication of scoring to their own credit programs.The author would like to thank the many people ingovernment and the private sector whose insightscontributed to this work. The author is especiallygrateful to reviewers of earlier drafts of this work,including Mark A. Abramson; David Brickman;Charles A. Capone, Jr.; Barry Dennis; Gary A.Miller; Nicolas P. Retsinas; Steve Robertson; andRobert S. Seiler, Jr. These reviewers provided manyvaluable comments. The author also wishes toexpress thanks to the PricewaterhouseCoopersEndowment for the Business of Government forfunding this work. Sole responsibility for thecontents of this report rests with the author.Section VI suggests options for additional applicationsof scoring-based systems to federal creditprograms. Perhaps most important, scoring canCredit Scoring and Loan Scoring 7

II. Credit Scoring and LoanScoring in the PrivateSectorA. Credit Scores and Loan ScoresScoring is a way to apply statistical modeling to arepresentative database and generate a numericalscore for each borrower or loan. The score can beused to classify individual borrowers or loans intorisk categories. A credit score is a number that isintended to predict a borrower’s propensity to repaya loan; a loan score expands upon the credit scoreto include variables relating to loan characteristics— for example, the loan-to-value ratio for a homemortgage — to create a numerical indicator of theprobability that a loan may default. 1Within the range of scores, lenders establish a cutoffpoint according to the amount of risk that theyare willing to take with respect to borrowers orloans. In the mortgage market, for example, borrowerswith low credit scores might be served byso-called subprime lenders rather than through themortgage lenders who serve borrowers with higherscores in the conventional mortgage market.As a technical matter, it is important to distinguishthe actual factors that these scores measure. The1See, e.g., Loretta J. Mester, “What's the Point of CreditScoring?” Business Review, Federal Reserve Bank ofPhiladelphia, September/October 1997, pp. 3-16; and RobertB. Avery, Raphael W. Bostic, Paul S. Calem, and Glenn B.Canner, “Credit Risk, Credit Scoring, and the Performanceof Home Mortgages,” Federal Reserve Bulletin, July 1996,pp. 621-648.most common credit score, developed by Fair,Isaac and Company, is known as the FICO score.Fair, Isaac developed the FICO model to predictthe likelihood of a consumer loan going into delinquencywithin two years of origination. By contrast,a mortgage score is designed to measure thelikelihood that a 30-year mortgage, with a seventoten-year average life, will default and cause aloss. 2Credit scoring found its first applications inconsumer lending. Starting in the 1960s, financecompanies, followed by retailers and credit cardcompanies, began to apply scoring-based systemsto assess potential customers and evaluate creditapplicants. Data management firms began toconstruct credit models based upon informationextracted from credit bureau reports on borrowerswho had taken out consumer loans. These firmsconstructed databases and mined the data for correlationsbetween credit-related information abouta borrower and that borrower’s statistical likelihoodof becoming delinquent on a consumer loan.One of these firms, Fair, Isaac and Company,created a range of scores to capture these probabilities,from a low FICO score of 200 to a high scoreof 800. As lenders expand the use of the score-2See, e.g., Gordon H. Steinbach, “Making Risk-Based PricingWork,” Mortgage Banking, September 1998, pp. 11-21.8 Credit Scoring and Loan Scoring

cards, Fair, Isaac and users respond to feedbackand refine the scorecards to improve the correlationbetween score and actual credit performance.Lenders applied such scorecards, and began todevelop their own proprietary models and scores,for credit card loans, installment loans, and automobileloans. Today, lenders routinely use scoringto determine whether to extend such consumercredit and on what terms.Private loan servicers also benefit from scoringbasedsystems. Credit scores can help in the processof devising cost-effective strategies for collection.Credit scores also can be combined with informationabout the loan and loan collateral to create anew servicing scorecard that relates directly to therisk of nonpayment by individual borrowers. 3Creditworthy borrowers with high scores can beexpected to cure their delinquencies, possibly withoutthe lender intervening at all. Other borrowersmay require prompt intervention. Database managementpermits servicers to target their interventionswhere they will be most effective.Not all loans benefit from credit scoring or loanscoring. If the probability of default varies largelybased upon factors other than the borrower’s individualcredit, then loan scores do not add muchvalue to credit administration. If loan data are notstandardized, or if sound historical data are unavailable,then loan scores will lack predictive value.Effective loan scoring requires large amounts ofhigh-quality data. Many different types of data arerequired, including information about loan originationand continuing loan performance, borrowercharacteristics, and the financial outcome, i.e.,whether the loan prepays, becomes delinquent,defaults, or pays in full on time. The data must beavailable for a long period of time so that importantbackground factors, notably the robust economyin recent years, can be taken into account in apredictive model.Moreover, the data must be complete and clean. Togather and clean the data can involve substantialtime and effort. Thus, to the extent that a federalagency cannot avail itself of a commercially availableand appropriate set of credit or loan scores,scoring implementation can impose significantchallenges and costs.Until someone develops a sound database and amodel with predictive value based upon that database,loan scoring will not be useful in making orsupervising federally guaranteed loans for FHAfinancedmultifamily rental properties, for example.By contrast, FHA single-family loans are scorablebecause information relating to defaults is standardizedand available from many years of experience. 4Finally, one other application of scoring deservesmention. This is the “institution” score that canhelp federal credit agencies to monitor the performanceof lenders or other participants in their programs.The Department of Education, for example,needs to monitor the default rates of schools thatoffer federal direct or guaranteed loans to their students.Critical institution variables relate to thequality of origination and servicing of federal loans.Depending upon an agency’s credit administrationneeds, it may be much easier to fashion an effectiveinstitution score than to create a new type ofloan score.By contrast to the loan score, which depends uponhistorical data, the institution score can be appliedby measuring the performance of lenders and otherinstitutions against their peers. Here, the key is toapply the score in such a way as to select anappropriate peer group for an institution’s performance.Because of its focus upon credit scoringand loan scoring, this report discusses institutionlevelscoring systems only to the extent that theyintegrate loan-level scoring into their analysis.3Larry Cordell, “Scoring Tools to Battle Delinquencies,”Mortgage Banking, February 1998, pp. 49-56.4For a cautionary note about scoring single-family loans, seeJim Kunkel, ”The Risks of Mortgage Automation,” MortgageBanking, December 1995, pp. 15-57.Credit Scoring and Loan Scoring 9

B. Conducting ControlledExperimentsOnce a lender has access to an effective datamanagement system, managers can mine the datafor useful information. Access to appropriate loanlevelscores can permit a federal credit agencyto conduct such experiments with greater costeffectivenessthan if scores are not available.A number of promising practices in the privatesector use information generated from selectedsubgroups that then are used to run controlledexperiments. Officials of Fair, Isaac and Companyapply the term “adaptive control systems” todescribe the process of devising strategies forcredit management, using control groups to testoutcomes of alternative strategies, and modifyingoverall processes and strategies in response tothe learnings. 5The essence of a controlled experiment is the waythat it permits managers to test new approaches tocredit management. Each current approach — toorigination or servicing and collections of types ofloans, for example — is deemed the “champion.”The manager then selects subportfolios that areused to test alternative approaches, deemed“challengers.” When a challenger yields a betteroutcome (here, in terms of reduced delinquenciesor defaults) then it becomes the champion. Theresult is a continuous process of testing and refinementto move towards ever more valuable creditmanagement techniques.Controlled experiments can be valuable both inloan origination and in loan servicing and collections.In loan origination, for example, a lendermay decide to extend credit to a specified group ofnontraditional borrowers. Thus a mortgage lendermight relax underwriting criteria that would rateborrowers as high risk if they had failed to makecertain types of payments in the past. The lenderthen would track the delinquency and default ratesof these borrowers for a few years. Based upon thisexperience, the lender might decide that paymentrecords for such payments were not helpful inpredicting the reliability of borrowers’ mortgage5Mary A. Hopper and Edward M. Lewis, “Behavior Scoringand Adaptive Control Systems,” Fair, Isaac and Company,Incorporated, (San Francisco, CA: May 1992).payments and might omit such measures in thefuture, or perhaps weight them differently than inthe past.For loan servicing, the champion would be thecurrent approach to doing business. A varietyof challengers might be tested. For example, ifdata mining indicates that certain subgroups ofborrowers have problems making their very firstpayment, then one challenger might involve atelephone call to selected borrowers at the timethey receive their payment books to counsel themabout the first payment. It could turn out that thechallenger is cost-effective only for some types ofborrowers; data mining allows targeting of firstpaymentcalls to such borrowers rather than others.Another challenger might relate to meeting thespecial language needs of other types of borrowersand so forth.C. Automated Scoring-BasedSystemsAlthough the consumer credit industry has usedscoring systems for many years, new developmentsin electronic data interchange and processing meanthat lenders now can originate and service highvolumes of loans with unprecedented speed andaccuracy. A lender can build a sophisticated informationtechnology system, create a central database,and link it electronically to computers at thepoint of loan origination or servicing. This permitsa loan officer to input a new loan application andtransmit the data electronically to be matchedagainst the central database and scored. Thescoring-based decision is then transmitted backto the loan officer literally within minutes.The mortgage market began to adopt scoring-basedsystems in the 1990s. In 1994 Freddie Mac, followedby Fannie Mae, announced the applicationof scoring to mortgage loans. Using the new dataprocessingtechnologies that now are available, thetwo lenders soon demonstrated how new scoringbasedsystems could offer substantial operationaladvantages over old ways of doing business.Freddie Mac and Fannie Mae created automatedunderwriting systems that enable lenders to scoreloans and borrowers and accept large numbers ofapplications within only a few minutes. For themore creditworthy borrowers and higher-scoring10 Credit Scoring and Loan Scoring

loans, the systems permit mortgage lenders to makeimmediate “accept” decisions, lock in a mortgagerate, and quickly close the loan. For less creditworthyborrowers or lower-scoring loans, thesystems refer the application back to the lender.Programs are now available to provide guidanceto lenders about critical factors that the lendermight adjust to make the loan acceptable.Freddie Mac and Fannie Mae rolled out their automatedunderwriting systems in 1995; within twoyears these scoring-based systems accounted forabout 40 percent of mortgage originations. In1998, declining mortgage interest rates caused arecord number of refinancings in the mortgagemarket; automated underwriting systems permittedlenders to keep up with demand and to close arecord number of mortgage loans. Fannie Maeused automated underwriting to process nearly2 million loans in that year alone.New systems also permit companies to servicehuge numbers of loans. Mortgage servicers, such asCountrywide Home Loans, use their database systemsto generate borrower and loan informationwhen a borrower makes a telephone inquiry. Forexample, if a delinquent borrower calls, the systemrecognizes the caller’s phone number and routesthe borrower immediately to the collections department.Systems also can help to substitute forexpensive staff time:“The [Countrywide] system records recentinformation on an account and tries to guesswhy a call is being made, so a recordinganswers and tells the caller, for example, thathis or her monthly payment was receivedthree days earlier.” 6Using such systems, major mortgage servicingcompanies today process payments and collectionson immense volumes of loans. The top five mortgageservicers each serviced over $200 billion ofmortgages at year-end 1998; the top 10 companiesserviced a total of almost $1.6 trillion of mortgagevolume.6Ted Cornwell, “Technology Will Help Lenders Cope: DefaultsRise While Economy Booms,” Mortgage Technology,March/April 1998, pp. 31-36.Credit Scoring and Loan Scoring 11

III. Strategic Implicationsof the New TechnologiesThe new information-based technologies haveprofound implications for the credit markets.Consider the consequences for borrowers,competing financial services providers in theprivate sector and government, and the futurerole of some of the larger federal credit programs.A. Implications for BorrowersFor borrowers, a major consequence is increasingclassification of loan applicants into distinct creditcategories. These classes will be determined by twofactors: (1) creditworthiness, and (2) transparencyof information about their creditworthiness. 7Because of the increased ability of lenders tostratify borrowers into classes, scoring systemsheighten the importance of accuracy of theinformation in each borrower’s credit report.In the highest class are those borrowers who arecreditworthy on the basis of easily accessible information.The new technologies will serve theseborrowers through electronic loan transactions thatapprove and close loans almost instantaneouslyand at lower cost than ever before.The next class consists of borrowers who are creditworthybut whose past credit activities may fall7The issue of transparency versus translucent or opaque informationabout credit quality is explored in, William R. Emmonsand Stuart I. Greenbaum, “Twin Revolutions and the Future ofFinancial Intermediation,” Olin Working Paper 96-48,November 1996. The author also is grateful to Neil Conklin,Chief Economist of the Farm Credit Council, for his applicationof this model to rural credit markets in a talk at JohnsHopkins University, March 1999.outside of traditional statistical patterns. These areborrowers whose creditworthiness is translucent oropaque rather than transparent to the new scoringbasedsystems. These borrowers are likely to findthat they receive very limited terms on credit cardapplications. Their mortgage applications are likelyto fall within the category of applications that failto receive an automatic “accept” decision from anautomated mortgage underwriting system in theconventional market.Instead of receiving automatic approval, their applicationwill return to the lender for individualized,and more time-consuming and resource-intensiveconsideration. Such applicants may find that theyqualify for special private sector or governmentloan programs.Over time, the use of controlled experiments canhelp lenders to make the creditworthiness of someof these borrowers more transparent. Thus, analysisof payment records of renters might reveal thattimely payment of rent and utility bills correlateswell with reliable mortgage payments once theserenters become first-time homebuyers. Once thescoring system incorporated this new information,more loan applicants would be scored in the“accept” range than before because their creditworthinessnow would be clear to the scoringsystem.The bottom class of credit applicants includes twokinds of borrowers who are not creditworthy: those(1) with and (2) without a transparent credit history12 Credit Scoring and Loan Scoring

that reveals their lack of creditworthiness. In thepast, the credit markets, and especially governmentprograms, might have extended credit to the lattergroup of noncreditworthy borrowers simply out ofan inability to screen them out.Over time, the application of scoring-based systemswill increase the transparency of information aboutsuch borrowers. Again, as lenders include newtypes of information in their databases, they mayfind transparent new reasons to exclude someborrowers from the “accept” category. To take theexample above, lenders might find that a poorrecord of renters in making rent and utility paymentswas statistically relevant in helping to predicttheir unreliability in making mortgage payments ifthey decide to become homebuyers. As such informationis validated, borrowers who lack creditworthinessincreasingly will be denied credit, bothby private lenders and by government programsthat are concerned about default rates.Given the heavy costs that default can imposeupon a borrower, and the value of denying creditto those borrowers who in fact are unlikely to beable to handle their debt burdens, this is generallya good outcome. Some government programs inthe past have done a grave disservice to someborrowers by extending credit that they could nothandle. 8Thus, the widespread application of scoring-basedsystems will stratify borrowers as never before. Themarkets will serve the most preferred class of borrowerthrough automated underwriting and lowcosttransactions. Lenders will serve the less transparentclass of creditworthy borrower with morecostly and time-consuming processes.The new scoring-based systems may permit theextension of some special and limited forms ofcredit, possibly at higher-than-average prices, toless creditworthy borrowers. The new systemsalso will increase the transparency of informationabout less creditworthy borrowers who in the pastreceived more preferred forms of credit than wouldbe merited by their actual circumstances.8Thomas H. Stanton, “Improving the Design and Administrationof Federal Credit Programs,” The Financier: Analyses of Capitaland Money Market Transactions, May 1996, pp. 7-21.B. Implications for FinancialServices ProvidersScoring-based systems also are exerting a profoundinfluence upon lenders and other providers offinancial services. The new technologies are drivingdown the costs of processing information relatingto loan origination and servicing. The result hasbeen to lower transaction costs and create excesscapacity. Several strategic implications help toprovide a context for the discussion here.First, an increasing proportion of business isshifting from traditional lenders and loan-relatedservice providers, who base many of their loansand services upon personal relationships withcustomers, to high-technology companies that substituteanalysis of information databases for olderways of doing business.Information databases have become importantcompetitive assets. In the residential mortgagemarket, for example, information-based systems arehelping to squeeze primary mortgage lenders andservice providers between emerging Internet-basedshopping and loan origination services, on the onehand, 9 and powerful secondary market databaseson the other. 10 The result is to shift substantial valueaddedand related profits, from primary lendersand providers of loan-related settlement servicesto institutions that use scoring-based systems tomanage risk and reduce transaction costs.Second, the new technologies are taking apart oldfunctions and reconstituting them in new ways.Thus, the new technologies have created economiesof scale and have increased the capacity of companiesto service loans effectively and inexpensively.The resulting increase in productivity has forcedconsolidation among mortgage servicing companiesand the dramatic growth in servicing volumes, asnoted above. Low costs permit centralized servicersto offer round-the-clock access to borrowers,instead of traditional face-to-face service at alender’s office only during business hours.9Kenneth A. Posner, "The Internet Mortgage Report: NewModels, New Opportunities," Morgan Stanley Dean WitterInvestment Research, February 4, 1999.10Office of Federal Housing Enterprise Oversight, "EnterprisesIntroduce Automated Underwriting Systems," 1995 AnnualReport to Congress, 1995, pp. 1-7.Credit Scoring and Loan Scoring 13

Technological change places a premium uponorganizational flexibility. Home computer terminals,traveling loan officers with laptop computers,and remote servicing facilities now can substitutefor the expense of brick-and-mortar offices thatonce were the proud hallmark of lenders and otherproviders of financial services.The pace of change makes organizational stabilityhard to sustain. As information-based technologiesbuild upon one another, increasing numbers andkinds of organizations find that new systems andprocesses make many of their functions obsolete,in whole or in part. Outmoded organizationalstructures find themselves subjected to unprecedentedstress as they suffer in the competition forresources to sustain themselves.Third, continuing waves of improvement in scoringbasedsystems mean that nimble participants in themarket are able to shift higher-risk loans to lesssophisticated competitors. These market dynamicsaccelerate the process of adverse selection, bothbetween more-adept and less-adept competitorsand between the private sector and government.The initial enthusiasm about scorecards and scoringbased systems has been followed by warningsabout the need to constantly update databases andadjust to feedback so that scorecards do notbecome outdated. 11For credit market participants, one important factoris the relative quality of their scoring-based systems.One recent article presents credit scores asa tactical problem; lenders that use more sophisticatedscorecards can sell higher-risk loans to unsuspectingbuyers whose scorecards do not detect theactual level of risk involved and who do not pricethe transaction correctly. The article concludes:“[A]s more and more institutions incorporatescore analyses into their decision making,the companies that don’t will find themselvesbeing adversely selected and, unknowingly,11See, e.g., Michael Todd, Robert Kennedy, and Colette Fried,“Ten Steps to Better Credit-Scoring,” The Journal of Lendingand Credit-Risk Management, October 1998, pp. 54-59;Mark H. Adelson and Linda A. Stesney, “Dispelling SomeCommon MBS Myths,” special report, Moody’s InvestorsService, December 12, 1997.12Dan Feshbach and Pat Schwinn, “A Tactical Approach toCredit Scores,” Mortgage Banking, February 1999,pp. 46-52, at p. 52.accepting more credit risk than theyplanned.” 12The new technologies also affect government creditprograms in many of these same ways.C. Implications for Federal CreditProgramsSeveral developments deserve discussion here.First, as the private market becomes more skilled atapplying scoring-based systems to loan originationand servicing, the gap in transaction costs betweenprivate loans and government loans (either directloans or loan guarantees) is likely to widen. Theimpact of this trend upon each particular federalprogram will vary according to the extent thatscoring has high predictive value in determiningthe likelihood that a loan will become delinquentor default.The reasons for this trend relate to the dynamicscreated by the rapid adoption of scoring-basedsystems in the private market. As the private marketbecomes more adept with such systems, it will usethem to provide credit to worthy borrowers whosecredit-related information is fairly transparent.Government direct loan and loan guarantee programsthen will be left with a higher proportion ofapplicants and borrowers with less transparentcreditworthiness. The benefits of information-basedsystems thus will accrue most directly to applicantsand borrowers with transparency, and lower theirorigination and servicing costs the most, comparedto the remaining kinds of borrowers.Also, to the extent that the new scoring-basedsystems permit the private market to accelerate theprocess of adverse selection, the government islikely to serve an increasing proportion of applicantsor borrowers of questionable creditworthiness. Theaverage loan application then will require morecareful processing than in the past and may requirededication of increased resources, such as financialcounseling, to make many borrowers creditworthy.To the extent that the average borrower becomesless creditworthy and the number of loan delinquenciesor defaults rises in a federal program,loan-servicing costs also will rise. Current loans areeasy to service, simply by collecting payments andaccounting for them. By contrast, greater servicing14 Credit Scoring and Loan Scoring

esources are required to deal with loans that aredelinquent or in default. Thus the disparity in loanadministration costs, between private loans andthose loans made directly or guaranteed by government,is likely to grow.Second, the new technologies are likely to applysignificant stress to many government agencies, toan even greater extent than occurs with privatefirms. Governmental organizations tend to be morerigid than are those in the private sector. Strongconstituent support may sustain many organizationalunits, such as an agency’s offices in key states orcongressional districts. Some organizational structuresmay be prescribed by law, and especially inappropriations acts. To the extent that technologicalchange makes consolidation or reallocation offunctions advisable, federal credit agencies may notbe able to respond.Government organizations also possess other rigidities.Budget constraints, civil service and classificationlaws, and pay differences between the governmentand the private sector may limit the numbersand affect the skills of people that staff an organization.Budget limitations, and especially the annualnature of the budget process for many federal creditagencies, may limit the amount of money that governmentcan invest in new systems or processes.Time-consuming procurement procedures and pressuresto accept a low bid rather than the most costeffectiveproposal also limit government.To the extent that technological change requires achange in business processes, many federal creditagencies may lack the ability to adjust either theirstaffing or their investments to keep up. Technologicalchange thus may apply far greater amountsof stress to the structure and operation of somegovernmental organizations than to the functioningof more flexible and adaptable private organizations.In this environment, federal credit agenciesmay find that the need for organizational changebecomes an integral part of the strategic planningprocess.It is notable that those federal credit agencies thatare authorized to operate as wholly owned governmentcorporations show signs of being more flexiblethan the usual government department oragency. The three wholly owned federal governmentcorporations that have a primary mission ofproviding credit are the Government NationalMortgage Association (Ginnie Mae), the Export-Import Bank of the United States, and the OverseasPrivate Investment Corporation.Third, especially some of the larger government creditprograms may find that advances in the private sectoraccelerate adverse selection to the point that seriousresponses are required to avoid taking unacceptablelosses. Tracking earlier warnings, an actuarialreport on the state of the FHA’s Mutual MortgageInsurance Fund suggests that new technologies canfacilitate a process of adverse selection:“[T]he improvement of the governmentsponsoredenterprises’ (GSEs’) ability to underwritehigh quality, high LTV loans might causean adverse selection effect. That is, withoutmodifying its underwriting rules, FHA mightend up with lower average quality loans…FHA is studying the development of its ownmortgage scoring system. However, becauseof the relatively low volume and short performancehistory, little information can be drawnto quantify any of these effects. The ongoingdevelopments in these areas should continueto be closely followed in the future.” 13The conventional mortgage market has made steadyincursions into the market share of federal mortgageprograms, and especially the market traditionallyserved by FHA. In 1970, FHA provided mortgageinsurance for 24.6 percent of the single-family mortgagesoriginated that year; VA provided loan guaranteesfor 10.8 percent of single-family mortgagesoriginated. By 1986, these numbers had droppedto 13 percent and 4.6 percent, respectively.A steady decline in market share has meant that FHAand VA at the end of 1997 served only 8.6 percentand 3.3 percent of the market, respectively. 14 While13Price Waterhouse, “Section I: Introduction,” An ActuarialReview for Fiscal Year 1997 of the Federal Housing Administration’sMutual Mortgage Insurance Fund: Final Report, February19, 1998, p. 7. See also, PricewaterhouseCoopers LLP, “SectionI: Introduction,” An Actuarial Review for Fiscal Year 1998 of theFederal Housing Administration’s Mutual Mortgage InsuranceFund: Final Report, March 1, 1999, p. 7.14Calculated from U.S. Department of Housing and UrbanDevelopment, “Mortgage Originations, 1-4 Family Units byLoan Type: 1970-1997,” Table 16, U.S. Housing MarketConditions, May 1999, p. 64. In part, the decline in VA marketshare reflects a reduction in the number of veterans eligible touse the program.Credit Scoring and Loan Scoring 15

the conventional mortgage market grew by over 70percent between 1986 and 1997, the number oforiginations of FHA and VA loans grew only slightly.At the same time, the private mortgage market hasbeen able to use new technologies to improve thecredit quality of conventional mortgages comparedto those insured by FHA. In 1986, FHA mortgageswere 1.9 times more likely and VA loans 1.8 timesmore likely than conventional mortgages tobecome 90 days past due. By the end of 1998,these ratios had jumped to 4.7 times higher forFHA and 4.2 times higher for VA, compared toconventional mortgages. 15Adverse selection will affect different credit programsdifferently. Some programs such as the federal directand guaranteed student loan programs may offerspecial terms that the private markets may not beable to match, except at the margins. Such programsare likely to be somewhat protected against largescaleadverse selection in the near future. Also, thosetypes of loans for which scoring has not yet beendeveloped, such as farm mortgages and multifamilymortgages, will not be subject to this form of adverseselection. By contrast, the FHA and VA single-familyprograms and possibly other federal programs suchas the business loan programs of the Small BusinessAdministration would seem to be more susceptibleto such developments.These changes in market dynamics ultimately mayrequire that some government credit programsundergo substantial business process reengineeringif they are to continue to serve their public purposes.New technologies are generating competitivepressures on market players, and organizationalstrength and flexibility will become increasinglyrelevant to programmatic success. Pressure also canarise from the increased perception that the privatesector is doing a substantially better job of loanadministration than a government agency.For some federal agencies organizational redesignmay be an essential part of gaining the capacity togather and process loan information. An interestingexample comes from the Rural Housing Service(RHS) of the Department of Agriculture. Thisprogram was under pressure from capable privatesector servicers who wanted the Congress to privatizeRHS loan servicing. The RHS responded with amultiyear program to centralize servicing of ruralhousing direct loans in St. Louis, to add new technologicalcapability to field offices to help withloan origination and servicing, and to relocate anddownsize staff to accommodate the changes. TheRural Housing Service obtained a multiyear commitmentfrom the Office of Management andBudget of the funds needed for the new technologyand staff training.All parties lived up to their commitments, and thenew office is now operating on the basis of state-ofthe-artservicing technologies. The result of centralizedservicing has been to begin to create thecapacity of the Rural Housing Service to gatherhigh quality loan level data and to monitor loanperformance in a timely manner. When a borrowerbegins to become delinquent, the RHS now is ableto intervene early and allocate its scarce staffresources more effectively, trying to avert a default.Program changes can involve more flexible andstronger organizational structures or changes in theform of government credit support. In the mid-1990s, FHA Commissioner Nicolas Retsinas soughtto enhance the organizational capacity of the FHA.He proposed legislation to transform the FHA into aFederal Housing Corporation. A HUD report at thetime suggested that the new wholly owned governmentcorporation, to be known as the FederalHousing Corporation, would be a nimble organizationthat could gather and respond to informationin a timely manner.“[U]nlike the existing FHA, [the new corporation]would function through consolidated,flexible product line authority and new operationalflexibilities so that it can easily adaptto market demands and customer needs.” 16Opposition from some constituencies meant thatthe Federal Housing Corporation idea failed toreceive serious congressional consideration.15Calculated from U.S. Department of Housing and UrbanDevelopment, ”Mortgage Delinquencies and ForeclosuresStarted: 1986-Present,” Table 20, U.S. Housing MarketConditions, May 1999, p. 68.16U.S. Department of Housing and Urban Development, HUDReinvention: From Blueprint to Action, March 1995, p. 49.See also, “Federal Housing Corporation Charter Act (Draft),”pp. 60-72.16 Credit Scoring and Loan Scoring

Especially if adverse selection pushes a federalcredit program into a small niche, the governmentmay want to consider transforming the form ofcredit support that it provides for underserved borrowers.One idea along these lines was containedsome years ago in an OMB proposal to revise theFHA single-family mortgage insurance program.OMB proposed that FHA extend credit through aprogram of providing credit enhancements forpools of high loan-to-value and other high-riskmortgages securitized by Fannie Mae, FreddieMac, or other securitizers.The enhancement, in the form of a loss reserve,was to be designed to assure that the cash flow toinvestors would not be interrupted by defaults. Asin the current program, FHA was supposed to continueto charge borrowers a fee to fully fund theloss reserves and cover its administrative costs. 17This proposal too met with substantial objectionsand was not refined to the point of addressingsome of the significant policy issues that it raised.The point here is that the transformation into aprogram of credit enhancement may be one usefulway that a federal credit agency such as FHAultimately could address the problem of adverseselection that otherwise could cause intolerablelosses to a federal credit program. The OMBproposal, for example, would have allowed FHAto budget for its credit risk each year in actualdollars provided for credit enhancement, ratherthan maintaining an open contingent liability.It would be appropriate to consider a range ofoptions for preserving a program such as the FHAhomeownership program that serves such importantpublic purposes. Increasingly, the value added fromfederal credit programs is likely to involve the provisionof information to facilitate the flow of creditto underserved borrowers, rather than merely theprovision of credit itself.17The Office of Management and Budget, “FY 1996 Passback:Department of Housing and Urban Development,” November21, 1994, pp. 21-22.Credit Scoring and Loan Scoring 17

IV. Public Policy IssuesRelating to Adoption ofScoring by Federal CreditAgenciesGiven the strategic significance of scoring-basedsystems for the delivery of federal credit, it is reasonableto argue that many federal credit agenciesought to incorporate appropriate systems into theadministration of their credit programs. However,scoring-based systems cannot be applied blindly tofederal programs.Officials in a number of agencies express a varietyof reservations about credit scoring and loan scoring.These may include concerns about the consequenceof measuring the low creditworthiness ofborrowers in some programs, and the possible disparateimpact of credit scoring and loan scoring onminority borrowers. Other program managers mayfear that disclosure of credit and loan scores mayhave negative effects upon the availability of federalcredit in the future to the types of borrowers orloans that score least well.The issue of possible disparate impact on subgroupsof borrowers raises special concern. In one studyconducted over twenty years ago, statistical analysisrevealed that predictive factors relating to repaymentof SBA small business loans differed significantlybetween white and African American borrowers.Applying one group’s factors to the other groupresulted in a significant increase in poor loan underwriting.Not only were some creditworthy borrowersprecluded from obtaining a loan, but loans weremade to less creditworthy borrowers who had a significantprobability of failure. As the author pointedout, both results were harmful; in the latter case,“[o]nce the black borrower has failed in business,he must meet his loan repayment obligations unlesshe pleads bankruptcy.” 18A recent study of the residential mortgage marketraises similar concerns. One issue relates to omittedvariables in a credit-scoring model. Important omittedvariables for home mortgage loans wouldinclude payment histories for rent and utilities,which typically are not included in credit-scoringmodels. The omission of such variables can make itdifficult for creditworthy renters to score appropriatelywhen they apply for mortgage loans.Another critical issue relates to the application tonon-random subsets of the population of credit historiesthat had been developed for a more generalpopulation:“A particular concern in this respect is thatminorities and lower-income individuals maybe systematically underrepresented in the18Timothy Bates, “An Econometric Analysis of Lending to BlackBusinessmen,” The Review of Economics and Statistics, vol.LV, No. 3, August 1973, pp. 272-283, at p. 282.18 Credit Scoring and Loan Scoring

aseline populations used to develop thescoring models. As a result, members ofthese groups, as they move into traditionalcredit markets, may be evaluated by modelsthat may not accurately reflect their repaymentpropensities.” 19The unfairness that can result from applying nonrepresentativedatabases to subgroups of borrowers,and especially minorities, has not yet been fullyresolved. 20 Acting Assistant Attorney General BillLann Lee, head of the Civil Rights Division of theU.S. Department of Justice, has addressed thequestion of disparate impact upon minority loanapplicants and borrowers without fully answering it.Mr. Lee spoke to the Mortgage Bankers Associationabout credit scoring and fair lending. He stated,“Well designed and fair credit-scoring systems holdgreat promise for objective decisions on any formof loan application.” 21However, even though sound credit-scoring modelsthemselves may not discriminate, Mr. Lee did seeseveral prospects for discrimination in the lendingdecision itself. First, law enforcement agencies havefound discrimination in lenders’ failure to provideminority applicants with the same level of assistancethat they provide to white applicants in securingtheir loans. The most common areas are the failureto help with explanations of negative credit historiesand with documentation of applicant income.Second, Mr. Lee pointed to lenders who overridecredit scores more often for white applicants thanon behalf of minority applicants. Such overrides caninclude both approval of applicants despite a failingcredit score and denial of applicants with a passingcredit score. Finally, Mr. Lee expressed concernabout price discrimination, in the form of lenderrequirements mandating that minorities pay higherpoints or interest rates than those required of whiteborrowers.All of these issues are receiving attention fromfederal officials. 22 The Federal Reserve Board, inRegulation B under the Equal Credit OpportunityAct, precludes creditors from using credit-scoringsystems unless they are “empirically derived,demonstrably and statistically sound.” 23Recently, the Department of Housing and UrbanDevelopment (HUD), through the HUD Office ofFair Housing and Equal Opportunity, requested andobtained information from Fannie Mae and FreddieMac about their automated underwriting systems,presumably to determine whether they meet Mr.Lee’s standard of being well-designed and fair to allapplicants and borrowers. 24 The Federal HousingAdministration is developing its own FHA scorecard,partly to try to assure that FHA scoring-basedlending decisions would be fair to all applicantsand borrowers.In other words, the state of knowledge concerningequal credit opportunity would seem to suggest thefollowing. First, credit-scoring systems can beconstructed to be sound and fair to all applicantsand borrowers. Second, the advent of credit scoringhas not removed opportunities for lenders and creditorsto discriminate access to loans and pricing.Third, as will be discussed below, there may be animportant role for federal credit agencies in assistingwith the development of databases that increase theaccess of disadvantaged borrowers to credit byincreasing the transparency of information abouttheir creditworthiness.The important point for federal credit programsis to provide reassurance that, so long as federalscoring based systems are well-designed, withsensitivity both to statistical validity and to fairnessissues, there seems to be little evidence that suchsystems would contribute to an increase in unfaircredit practices. Scoring-based systems are toolsto be used, as are any other tools of credit administration,with safeguards to prevent their misuse.19Robert B. Avery, Raphael W. Bostic, Paul S. Calem, and GlennB. Canner, "Credit Scoring: Issues and Evidence from CreditBureau Files," Board of Governors of the Federal ReserveSystem, February 23, 1998, p. 8.20See Florence Lockafeer and David Rosen, Credit ScoringSmall Business Loans: Capital Access or Bias? David PaulRosen and Associates, July 1996.21Bill Lann Lee, remarks before the Mortgage BankersAssociation Fair Lending Conference, March 23, 1998,p. 3 (prepared text).22See, e.g., Office of the Comptroller of the Currency, OCCBulletin 97-24, “Credit Scoring Models Description:Examination Guidance,” May 20, 1997.2312 CFR Sec. 202.2(p)24US Department of Housing and Urban Development, AStudy of the GSEs' Single Family Underwriting Guidelines,Final Report, Office of Policy Development and Research,April 1999.Credit Scoring and Loan Scoring 19

V. Current Use of Scoring-Based Systems by FederalCredit AgenciesA review of federal credit agencies reveals thatat least three have begun to permit some use ofscoring-based systems. The Federal HousingAdministration, Department of Veterans Affairs,and the Small Business Administration each permitlenders to use scoring-based systems in loan originationfor some programs. The VA is experimentingwith the application of scoring-based systems toloan servicing. No federal agency yet applies creditor loan scoring as a diagnostic tool to ascertain theperformance of lenders or the credit quality of portfoliosof direct loans or loan guarantees.Both FHA and VA now permit mortgage lenders touse approved automated underwriting systems tooriginate their loans. Both agencies undertookassessment of Freddie Mac’s Loan Prospector systemand have approved its use. FHA and VA arenow testing Fannie Mae’s Desktop Underwritersystem, including the PMI Mortgage InsuranceCompany’s pmiAURA scoring system, in a similarpilot program.After gaining experience in the Freddie Mac pilotprogram, FHA announced that it would developits own scorecard. FHA contracted with Fair, Isaacand Company to develop a distinct FHA singlefamilyscorecard. Preliminary indications are thatthe FHA automated system approves a much higherpercentage of applicants than are approved byautomated systems in the conventional market.Also, some lenders contend that automated underwritinggenerates approval for a significant proportionof FHA loans that would not have beenapproved under traditional FHA guidelines. 25The Small Business Administration also contractedwith Fair, Isaac and Company to obtain scorecardsto help the SBA underwrite and quickly decidewhether to approve loans under the SBA LowDocumentation (LowDoc) Loan Program. SBA hasmade a commitment to attempt to respond to aLowDoc guaranty request within one-and-a-halfbusiness days.The SBA applies scorecards based upon creditscores, rather than loan scores. The credit scoresprovide information about the propensity of thebusiness proprietor to repay, on the theory that thisis the single most important factor affecting thecredit quality of a smaller-sized small businessloan. The LowDoc program is limited to loans upto $150,000. SBA directs lenders to submit largerloans and loans with more complicated financialissues to the standard SBA business loan programfor underwriting according to usual procedures,rather than to LowDoc.The Department of Veterans Affairs has a specialconcern about the welfare of VA borrowers and theneed to avoid foreclosures whenever possible. VA25“What’s the Score at FHA?” ABA Banking Journal, AmericanBankers Association, October 1998, p. 54.20 Credit Scoring and Loan Scoring

is working with Fannie Mae and four servicers ofVA mortgages to test the application of FannieMae’s Risk Profiler system when servicing VA mortgages.This test is a controlled experiment that VAis conducting at no cost to the government.Starting May 1, 1998, the four servicers began toapply Risk Profiler to their servicing portfolios ofVA loans. About 500,000 loans will be included inthe test, out of a total VA guaranteed loan volumeof roughly 3 million loans outstanding. The fourservicers range in size from one of the largest to aservicer of more modest size.The servicers will perform a champion-challengertest on the 500,000 loans. Half of the loans will beserviced in the traditional manner. The other halfwill be scored and serviced with stratified techniquesthat involve special attention for early delinquencyloans with weak scores. VA and the servicerswill follow the performance of the loans forat least six months. Fannie Mae plans to use informationfrom the test as feedback to revise andimprove the VA loan segment of the Risk Profilerdatabase.The four servicers are willing to undertake theexperiment because of the savings that they willachieve in managing delinquencies of high-scoreloans (i.e., those that tend to reinstate with no or littleintervention) and in reducing defaults that otherwisewould require costly attention. If the pilot programis successful, then VA could revise its servicingregulations to permit all VA loan servicers to benefitfrom more flexible requirements if they use RiskProfiler or, eventually, other acceptable systems.The three pioneers in federal use of credit scoring forloan origination and servicing have been able to usesystems that could be purchased, with or withoutadaptation, from private vendors or applied throughprivate lenders. It should be pointed out that theadoption of a scoring-based system represents onlythe beginning of a process; to be successful, FHAand SBA will need to develop the capacity to obtainfeedback from the application of their scorecardsand to make adjustments from time to time. In theworld of constantly evolving systems, any marketparticipant that fails to learn from experience andadapt will find itself absorbing unanticipated losses.Credit Scoring and Loan Scoring 21

VI. Options for AdditionalApplications to FederalCredit ProgramsCredit scoring and loan scoring offer the opportunityfor some federal credit agencies to devise scoringbaseddatabase management systems for a broadrange of purposes. Once they include such scoresin their databases, federal managers can use theinformation to manage the credit risk of some largeprograms much more effectively than in the past.The opportunity to develop such approaches isgrowing; prices are dropping and new systems areconstantly coming to market.Several opportunities suggest themselves for governmentuse of scoring: (1) in financial early warningsystems, (2) to monitor performance of lenders thatparticipate in government loan guarantee programs,(3) to improve the servicing of federal direct loans,and (4) to help improve targeting of federal creditprograms. In other words, government credit agenciesmay find that, besides helping to enhance loanorigination and servicing, other important and immediatebenefits of scoring-based systems relate to theuse of scores for diagnostic and analytic purposes.A. Scoring-Based Financial EarlyWarning SystemsScoring-based systems can help federal credit agenciesto measure the creditworthiness of their portfolios.For some credit programs, this will be useful,but not imperative. Other federal credit programsface strong private sector competition and growingevidence of adverse selection. To protect the financialhealth of some such programs, these agenciesmay find it essential that they construct and maintainfinancial early warning systems andcontinuously refine them.In today’s markets, information is the basis forfinancial performance. The FHA and VA now sharetheir portfolio data with the private markets andpermit private firms to construct scorecards fromthe information. This information will assist theconventional market to mine the governmentguaranteedmortgage market for creditworthyborrowers. Issues of adverse selection, both amongborrowers and among lenders, are real and must beaddressed. Similarly, the growing use of scoring insmall business lending and even rural lending maysuggest the usefulness of financial early warningsystems for agencies such as the SBA and someloan programs of the Department of Agriculture,for example.The term “financial early warning system” can havea variety of meanings, depending upon the context.Here the term is meant to apply to a system thatmeasures the credit quality of portfolios of directloans or loan guarantees and that provides an earlywarning signal of deterioration in credit quality.For such financial early warning systems, creditscores and loan scores can add considerable value.Using statistical sampling, a federal credit agency22 Credit Scoring and Loan Scoring

can construct a database of current loans or loanguarantees, adjusted as new borrowers enter orleave the program. The federal agency then canscore the direct loans or guaranteed loans forwhich a score can be obtained.Initially, at least, programs that extend credit toindividuals such as homebuyers are likely to findcredit scores easier to obtain than loan scores.Interviews with mortgage lenders, for example,indicate that they now routinely obtain creditscores on all mortgages that they originate. Theyare unlikely to object if a federal credit agencysuch as FHA or VA required submission of suchscores as a part of the documentation for each loanthat the government insures or guarantees.The function of the financial early warning systemis to establish a baseline average score for the portfolio,or selected subsets of the portfolio, and toprovide a prompt report of changes in that averagescore or in the distribution of scores. If the systemsounds an early warning, the question becomesone of selecting an appropriate response. Federalcredit agencies may want to review their legalauthority and prepare a “what-if” handbook ofmeasures that are available in response to a signal.If an agency lacks sufficient authority to act,discussions with the relevant congressional committeesmight be useful. A repertoire of responsesis desired, as a way of permitting an agency todeal with an early warning signal and avoid theHobson’s choice between applying a draconianmeasure — say, to shut down some part of aprogram altogether — or waiting passively to seewhether the signal might have been premature.B. Adding Credit Scores or LoanScores to Lender MonitoringSystemsLender performance is an essential part of anagency’s ability to maintain the financial soundnessof a loan insurance or guarantee program, such asthose of the FHA, VA, SBA, and Department ofEducation, for example. Credit scores or loanscores now can offer some of these agencies newcapabilities for monitoring the quality of loans thatlenders originate and service.A private sector model is instructive. One privatemortgage insurer monitors its risk position as follows:1. Once insurance is in force, the company runsall loans through an automated underwritingsystem to obtain a mortgage score for eachloan. The company also obtains FICO creditscores for each loan.2. The company categorizes the mortgagesaccording to credit quality.3. The company then conducts an automatedreview of the quality of loans in groups andsubgroups: companywide, insured througheach of the company’s insuring offices,originated or serviced by each lender, andoriginated by each office of each lender.4. The company pays special attention to earlypayment defaults as an indicator of lenderperformance.5. For large lenders, this system can detectproblem patterns within 90 days of originationor sooner. It takes longer for patterns to emergewith respect to lenders that originate smallernumbers of loans.Once the information is included in the database,the company can stratify it according to particularvariables. These could include geographical location,type of loan (e.g., fixed rate or variable rate),term, and loan-to-value ratio, for example.The mortgage insurer uses its performance informationto visit lenders whose performance is unacceptablylow. The company reports that the FICOscore seems to correlate especially well with earlypayment defaults. The FICO score is also useful tohelp explain performance problems to lenders. Thecompany reports that most lenders respond positivelywhen they are approached with evidencethat average FICO scores on their originations havelagged those of their peer group.The private mortgage insurer follows up with carefulmonitoring of average FICO scores for originationsof mortgages by the lender (or by the officeof the lender) that have been singled out for review.Alternatively, the insurer may specify that it will notaccept originations of mortgages from that lender oroffice that fall below a specified FICO score.Credit Scoring and Loan Scoring 23

A private mortgage insurer tends to have greaterleverage over lenders than does a federal agency.Laws and regulations may require a significantprocess, for example, before a federal credit agencymay apply sanctions or refuse to do business with alender that originates loans with a high propensityto default.Nonetheless, as has been demonstrated by one federalagency, Ginnie Mae, an effective and reliablelender monitoring system can yield many dividends.Ginnie Mae too would benefit from adding a creditscoreor loan-score dimension to its successfullender monitoring systems, known as IPADS (IssuerPerformance Analysis Database System) and CPADS(Correspondent Performance Analysis DatabaseSystem). 26 Scoring systems based upon loan-levelinformation can help to monitor the performance oflenders or other institutions in originating soundloans and also in servicing them.C. Scoring to Improve Servicing ofFederal Direct LoansCredit scores and loan scores help lenders to testcurrent approaches, e.g., to loan servicing and collections,against experimental alternatives. The VAcurrently is running experiments to test applicationof scoring-based systems to servicing of VA homeloans, and the practice is likely to spread to servicingof other federally guaranteed loans.Federal credit agencies may find that scoring-basedsystems also can help with the servicing of federaldirect loans, such as Rural Housing direct loans.Some officials at the Department of Education alsohave begun to explore application of scoring-basedsystems to the servicing of direct student loans. Forsuch direct federal loans, champion-challenger testscan help to refine and target loss mitigation techniques.Combined with electronic data interchange,the government may be able to save money by targetingdefault prevention and loss mitigationapproaches to the types of borrowers who willbenefit most.Credit scoring provides a valuable starting place fordetermining whether intervention might be neededand what form it might take. Again, this is adynamic process that should evolve as it is applied.Credit managers can document correlationsbetween credit scores and the forms of interventionthat work most effectively in bringing a borrowerinto current status on repayments.A credit manager may find, for example, that earlycounseling is most effective for borrowers in acertain range of credit scores, but that it is notneeded for borrowers with high scores and is nothelpful for borrowers with low scores. Such insightcan help credit managers target such counselingto those cases where it is most likely to be costeffective.Credit managers may also begin to experimentwith new approaches to targeting borrowersfor whom such counseling was not helpful; perhapsanother form of intervention, or intervention earlier,may be more useful and cost-effective.By mining the databases of the contractors thatservice federal direct loans, credit managers canbegin to construct their own credit scores and thenrefine those scores as new correlations emergefrom the data. Such scores would reflect factorsthat experience shows to be relevant to predictinga borrower’s propensity to repay the particular typeof federal loan.Using experimentation and a process of comparingoutcomes, credit managers can determine whethertheir own credit scores are superior to credit scoresthat can be purchased off-the-shelf from privatevendors. It also may be possible eventually to combinecredit scores with other information to createan empirically derived loss mitigation score thatwould have even greater predictive power.The point of the exercise is to use scores as a toolto identify groups of borrowers for whom someinterventions are more valuable than others. Thishelps to direct a federal credit agency’s loan managementactivities so that they are the most usefulin protecting borrowers from the unpleasant consequencesof delinquency that could lead to default.26See Thomas H. Stanton, “Managing Federal Credit Programsin the Information Age: Opportunities and Risks,” TheFinancier: Analyses of Capital and Money MarketTransactions, Summer/Autumn 1998, pp. 24-39.24 Credit Scoring and Loan Scoring

D. Scoring to Improve Targeting ofFederal Credit ProgramsUltimately, some programs may be able to applyscoring-based systems to help improve the targetingof federal credit. The most suitable borrowers mightbe defined as those who both (1) need the federalloans or guarantees because they lack access tononfederal credit on reasonable terms, and (2) arecreditworthy enough to be able to repay their federalloans without risking unacceptable levels ofdefault. To the extent that improved targeting isacceptable to a program’s stakeholders, somefederal programs could complement the privatecredit markets with much more precision than everbefore, and therefore with greater cost-effectiveness.Especially in credit markets that are being transformedby widespread use of scoring-based systems,the availability of credit scores and loan scores canhelp federal agencies to devise more flexible underwritingcriteria that enable the federal program torespond promptly to market developments. Aslenders increasingly adopt automated underwritingsystems, they will be able to adjust their own loanorigination systems to follow such changes withoutdifficulty. Again, it should be noted that creditscores or loan scores might be inappropriate forsome federal credit programs, especially where theavailable information is inadequate.For appropriate programs, scoring can be used toreduce access of borrowers who are unlikely tobe creditworthy, while increasing access of creditworthybut nontraditional groups of borrowers. Aswith any new way of doing business, the effects ofscoring must be evaluated in terms of the publicpurposes that a credit program is supposed to serve.Some federal credit programs may be able to applyscoring to help to screen out borrowers who areunlikely to be able to handle their debt burdens.Candidates for this application of credit scoreswould seem to include some federal direct loanprograms that serve borrowers who potentiallymight be at the lower end of creditworthiness.The SBA disaster loan programs, for example,might run quick credit scores on applicants for SBAloans after a disaster. Those people or businessesthat were not creditworthy before the disaster areunlikely candidates to be able to repay their federalloans afterward.Increased use of scoring-based systems also canprovide an opportunity for some federal agencies tocomplement the behavior of private lenders andincrease access to credit for new creditworthy borrowers.Consider again some of the limitations ofscoring based systems: credit scores and loanscores are valuable only as applied to people andloans whose credit-related characteristics were capturedin the sample of borrowers and loans thatwere used to develop the scoring database.Some analysts complain that credit scores and loanscores may not properly reflect the creditworthinessof nontraditional borrowers. They point to factorsthat reduce a borrower’s credit score, such as thenumber of credit inquiries, whether the borrowerborrows from a finance company rather than abank, or whether the borrower has a large numberof small credit balances outstanding. While thesefactors may have predictive value for the averagemiddle class borrower, there is a chance that someof them are not statistically relevant to the creditworthinessof some subgroups of borrowers.Needed is an opportunity to conduct controlledexperiments to determine whether some such factorsmight be relaxed, or replaced by other morepredictive factors, when applying credit scores tosome nontraditional borrowers. The federal governmentis ideally suited to support such experiments.A federal credit agency such as FHA might developa risk-sharing program with private lenders toextend mortgage credit to nontraditional borrowersand monitor their propensity to repay the loans.The FHA already possesses statutory authority toenter into risk-sharing arrangements.The resulting loan performance data could beshared with the private market as a way to facilitatethe extension of credit to creditworthy borrowerswho previously could not have been served on reasonableterms. In the terminology of Emmons andGreenbaum, the government would help to maketransparent the previously opaque creditworthinessof these nontraditional borrowers.Ultimately, in the words of one analyst, taken froma different context, FHA could become “the BellLaboratories for housing finance.”Credit Scoring and Loan Scoring 25

“…FHA must test conventional perceptionsof risk. It must challenge current orthodoxy.It must design new mortgage products, providefor new ways of support … and continuallypush all participants … to providecheaper, more efficient and more responsivecredit products to the marketplace.” 27Whether Congress will allow federal credit agenciesto take on this role is an open question. In anyevent, a major lesson of the information age isthat federal credit agencies need to increase theirvalue by creating information rather than merelymanaging the credit risk of large government director guaranteed loan portfolios.27David Rosen, Target Markets and Key Products: Conclusion,Recommendations and Findings, FHA Multifamily HousingBusiness Strategic Plan, September 29, 1995, at p. 7.26 Credit Scoring and Loan Scoring

VII. Conclusion: The NewWorld of Scoring-BasedSystemsCredit scoring and loan scoring are here to stay.The conclusions of this research can be highlightedin terms of recommendations for federal programofficials and policymakers about ways to addressthe opportunities and risks that the new technologiespresent. The opportunities that come fromscoring-based systems relate to new operationalimprovements that can enhance the capacity offederal credit programs to serve their public purposes.The risks are more strategic in nature andrelate to the ways that some agencies will need torespond if their programs are to keep pace in thenew world of scoring-based systems.These systems offer opportunities for many federalcredit programs to increase their institutionalcapacity to manage credit risk:• Major federal credit programs that involveloans of a type for which scoring is suitableshould create scoring-based systems to providefinancial early warning of a deterioration incredit quality of loans being originated underthe program.These programs include the single-family mortgageprograms of FHA, VA, and the RuralHousing Service. Because loans from all of theseprograms are securitized by Ginnie Mae, itwould be cost-effective to add loan-level scoringto a Ginnie Mae system that builds upon today’sIPADS and CPADS lender monitoring systems tocreate a financial early warning database as well.Credit scores would enable the combined systemto have access to loan-level data that couldbe sorted by geographic region and loan type, aswell as by the lender that originates or servicesthe loan. Discussions with persons in the mortgageindustry indicate that lenders are unlikelyto object to reporting FICO scores with theirloan origination information.Other federal credit agencies also shouldexplore the use of credit scores to create financialearly warning systems. These include theSBA business loan programs and the guaranteedstudent loan program. Especially the SBA 7(a)business loan program may need to be concernedabout adverse selection by lenders.• Major federal guarantee programs, includingthe single-family mortgage programs of FHA,VA, Rural Housing Service, and Ginnie Mae,and the guaranteed loan programs of the SBAand Department of Education, should adoptloan-level scoring systems to monitor the qualityof performance of lenders who originate andservice their guaranteed loans.The application of credit scoring, and eventuallyloan scoring, to lender monitoring promises considerablefinancial payoff because of the way itpermits federal agencies to allocate their scarceresources, preventing unnecessary defaults thatcan occur from poor origination or servicing.Again, the Ginnie Mae IPADS/CPADS systems,which been pioneers in this regard, need to beenhanced through application of loan-levelCredit Scoring and Loan Scoring 27

scores to provide information about credit quality.The Department of Education also shouldapply scoring to the monitoring of participatingeducational institutions and perhaps to a reviewof the credit performance of student loan guaranteeagencies.• Federal direct loan programs, including federaldirect student loans, the SBA disaster loan programs,and rural housing direct loans, shoulduse scoring to help conduct controlled experimentsin improved approaches to loan servicing.Good servicing is especially important in assuringthe repayment performance of many federalstudent loans, for example. The development ofchampion-challenger experiments in servicing,even if rudimentary at first, could help theDepartment of Education to experiment withapproaches to servicing that increase the loanrepayment performance of different types of borrowers.Perennial questions, such as whether itis better to send monthly statements or a loanpayment coupon book, then can be answeredon the basis of empirical information.Good servicing also is important in enhancingthe performance of the many less creditworthyborrowers who may receive credit from an SBAdisaster program or the RHS direct home loanprogram. In all of these applications, loan-levelscores can provide an important diagnostic toolto help devise and learn from experiments.• Federal credit programs including the FHA single-familyand SBA business loan programsshould use credit scoring and loan scoring toexperiment with improved targeting of creditworthyborrowers for whom traditional creditscores may be inappropriate.Scoring can be used to expand access to federalcredit. Federal credit agencies could add significantvalue by conducting out-of-sample experimentsto determine whether modifications tocommercially available scores might improvethe access to credit of some creditworthy butnonstandard types of borrowers.The federal agency first would experiment,either through a direct loan program or throughrisk sharing with lenders in a guaranteed loanprogram. Then the agency could use the informationto adjust its own underwriting standardsand guidelines. Once again, high quality informationand systems are needed for such effortsto succeed.The rapid deployment of scoring-based systems inthe private sector means that some federal programsmay be at risk if they continue to do businessin the old ways:• Federal policymakers, especially at the Officeof Management and Budget (OMB), shouldencourage appropriate federal credit agenciesto make multiyear commitments of resourcesto adopt scoring-based systems in the contextof well-designed strategic plans.The Office of Management and Budget hasplayed a major role in facilitating cooperationamong the federal housing credit agencies tobegin the creation of a shared-data warehousethat can be used to test new approaches to creditmanagement. OMB also has worked with particularfederal credit agencies, for example theRural Housing Service, to provide a commitmentof multiyear funding to help develop technologybasedimprovements in loan management.Working through the Federal Credit PolicyWorking Group, an interagency group chairedby OMB, the Office of Management and Budgetcan help agencies to share approaches to creditscoring and to develop cost-effective plans. Theindividual agencies then will need a commitmentof multi-year funding to assure that the systemsare adopted and integrated into day-to-dayoperations.• Federal credit agencies should devote neededresources to assuring that they remain wellinformed about technological developmentsand the implications for the markets in whichtheir programs operate.By staying informed of technological developments,federal credit agencies can learn ofopportunities to improve their managementpractices. The VA experiment with scoring-basedservicing of VA home loans is an excellent exampleof the way that a federal agency can use itsknowledge of industry practices to support itsown improvements in program administration.Federal credit agencies also need to remain alertto the possibility that new technologies can hastenthe obsolescence of some of programs orpractices. For example, a federal direct loan programcould be affected adversely if the agencyenters into long-term servicing agreements withcompanies whose practices and systems lag theindustry standard.28 Credit Scoring and Loan Scoring

• Federal policymakers in the executive branchand Congress should consider structuralchanges to credit programs and organizationsto increase their flexibility and capacity torespond to the many technology-driven changesthat affect their ability to serve their publicpurposes.To achieve useful programmatic and organizationalchanges requires more than the application ofsound principles of design. Also needed is thedevelopment of a consensus among a program’sstakeholders. The FHA’s attempt to create aFederal Housing Corporation benefited fromgood design but lacked the necessary consensus.Leaders of some federal credit agencies, andespecially those that benefit from significantexternal support, may need to begin a processof dialogue with stakeholders as a way to enlisttheir participation in a consensus that might bebuilt around the need to improve the design ofsome programs and organizations.take advantage of available systems and technologiesand private-sector practices that can effectivelybe absorbed into the organizational context of afederal program.Policymakers may need to redesign the form ofsome government programs so that they can complementrather than be undercut by a dynamicprivate sector. Government organizations themselvesmay need to become much more nimble ifthey are to continue to serve their public purposeseffectively in today’s rapidly changing environment.In the provision of credit, federal programs haveneither the resources nor the policy freedom tooperate at the leading edge of available technologies.On the other hand, the adoption of newpractices in the private sector ultimately maymake it untenable for some federal programs tocontinue doing business in the old ways.Credit scoring and loan scoring offer many possiblebenefits, but they also raise public policy concernsthat must be addressed. Federal credit agenciesalready have begun to partner with private lendersto apply scoring-based systems to the originationand underwriting of government-insured or guaranteedloans. Other important applications include theuse of scoring-based systems to develop or enhancefinancial early warning systems, to monitor lenderperformance, to improve the targeting of federalcredit to the most appropriate borrowers, and toexperiment with lending to subgroups of underservedborrowers in an effort to increase theiraccess to credit.Because they lack the requisite data, federalagencies may need to adopt and adapt off-theshelfscores and scoring systems to their ownprogram needs. Conversely, government agenciesare at special risk of making misjudgments if theyapply private-sector scorecards to subgroups ofborrowers in their programs who act differentlyfrom the general population for which the scorecardsmay be appropriate.As with many technology-based issues, the governmentwould benefit from working through the policyand organizational decisions relating to newways of doing business. Sound approaches need toCredit Scoring and Loan Scoring 29

AppendixScoring: Where to BeginCredit scoring and loan scoring can be valuabletools for many federal credit programs, but notfor all of them. New data management systemsare increasingly available at a lower cost thanever before. Here are some suggestions for federalofficials who may wish to explore the possibilityof applying scoring to their own programs.Gather InformationUseful sources of information include knowledgeableofficials at federal agencies that have begunto apply scoring to their programs, private sectorfirms that offer scoring systems, and private sectorfirms that have applied scoring to their own creditmanagement practices. Substantial literature onscoring now exists in trade, finance, and academicpublications.Develop a Modest PlanConsider starting small. Start with scoring as adiagnostic device, such as an early warning systemor to monitor lenders, rather than as an operationaldevice, e.g., to screen loans and applicants.Technology absorption is easier to implement if itcan be done without disrupting the work processesof multiple organizational units. Also, as in the VAservicing experiment, start with scoring as a supplementto existing practices, rather than as a substitute.A tentative business plan would include a dispassionateassessment of benefits and costs of the newscoring-based system and the time and resourcesthat are required for development, testing, andimplementation. Commercially available scoresare likely to be much less expensive than speciallytailored scores and may provide a good startingplace for many program uses.Critique the PlanTest the assumptions of the plan with officialsinside the agency and with people at other agenciesor private firms who have had experience withthe proposed application. Scrutiny of cost and timeestimates is especially important to assure thatimplementation won’t be stopped if the allocatedfunds prove insufficient.Try an ExperimentA small experiment can test assumptions and producevaluable feedback. For example, credit scoresmay turn out to be useful predictors of credit qualityfor some types of federal loans, but not for others.This is best tested before an agency tries toapply scoring to a large part of its business. Othertypes of experiment can help an agency to determinecost-effective approaches to carrying outfunctions such as the servicing of direct loans.Dissemination of the results of an experiment canprompt helpful comments from other federal agencies,the private sector, and perhaps economists orother academics. The development of a scoring systemis an iterative process based upon continuingmodifications that respond to the lessons and informationgained from experience.Try to Obtain Commitment from the TopGiven the resource constraints at many federalagencies, it may be important to obtain supportfrom senior agency officials before beginning toexperiment with credit scoring. Career-level officialsof the Department of Education, for example,were testing new approaches to avoiding loandefaults when funds suddenly were restricted sothat the experiment (not involving credit scoring inthis particular case) had to be terminated before theresults were complete. Commitment from the top ofan agency, and perhaps also from OMB, may beuseful to assure that when experiments begin, theyare permitted to run to completion.30 Credit Scoring and Loan Scoring

About the AuthorThomas H. Stanton is a Washington, DC, attorney. He is a fellow at the Center for the Study of AmericanGovernment at the Johns Hopkins University, where he teaches graduate seminars on the law of publicinstitutions; government and the American economy; and government and the credit markets.Mr. Stanton’s academic, legal, and policy experience relates to the capacity of public institutions to deliverservices effectively, with specialties relating to federal credit and benefits programs, government corporations,and financial regulation. In recent years, he has provided legal and policy counsel regarding thedesign and operation of a variety of federal government programs.Mr. Stanton is chair of the Standing Panel on Executive Organization and Management of the NationalAcademy of Public Administration (NAPA) and has helped to teach NAPA’s seminar on government enterprises.He is a former member of the Senior Executive Service.His writings on government and the financial markets include a book on government-sponsored enterprises,A State of Risk (HarperCollins, 1991), and numerous articles. He is a member of the Advisory Board of ajournal, The Financier: Analyses of Capital and Money Market Transactions.Mr. Stanton earned a B.A. from the University of California at Davis, an M.A. from Yale University, and aJ.D. from Harvard Law School. The National Association of Counties has awarded him its DistinguishedService Award for his advocacy on behalf of the intergovernmental partnership.Credit Scoring and Loan Scoring 31

Key Contact InformationTo contact the author:Thomas H. StantonFellowCenter for the Study of American GovernmentJohns Hopkins University801 Pennsylvania Avenue, NWSuite 625Washington, DC 20004(202) 965-2200e-mail: TStan77346@aol.com32 Credit Scoring and Loan Scoring

ENDOWMENT REPORTS AVAILABLEManaging Workfare: The Case of the Work ExperienceProgram in the New York City Parks Department (June 1999)Steven CohenColumbia UniversityResults of the Government Leadership Survey:A 1999 Survey of Federal Executives (June 1999)Mark A. AbramsonSteven A. ClyburnElizabeth MercierPricewaterhouseCoopersCredit Scoring and Loan Scoring: Tools for ImprovedManagement of Federal Credit Programs (July 1999)Thomas H. StantonJohns Hopkins UniversityTo download or order a copy of these reports, visit the Endowment website at: endowment.pwcglobal.com

About PricewaterhouseCoopersPricewaterhouseCoopers (www.pwcglobal.com) helps itsclients develop and execute integrated solutions to buildvalue, manage risk and improve their performance.Drawing on the knowledge and skills of 150,000 peoplein 150 countries, we provide a full range of businessadvisory and consulting services to leading global,national and local companies and to public institutions.Our objective is to have measurable impact. Our goal isto work with clients to make the world a better place.For additional information, contact:Mark A. AbramsonExecutive DirectorThe PricewaterhouseCoopers Endowment for theBusiness of Government1616 North Fort Myer DriveArlington, VA 22209(703) 741-1077fax: (703) 741-1076e-mail: endowment@us.pwcglobal.comwebsite: endowment.pwcglobal.comThe PricewaterhouseCoopers Endowment forThe Business of Government1616 North Fort Myer DriveArlington, VA 22209-3195Bulk RateUS PostageP A I DPermit 1112Merrifield, VA