emerged as a result of the AU systems now in place. But no one outside the purveyors of these systems can say for sure.

The Fannie Mae AU system is known as Desktop Underwriter; the Freddie Mac version is known as Loan Prospector. Each system relies on range of indicators, including numerical scores, loan to value ratios and other data submitted by the borrower to calculate a mortgage score. These scores, in effect, represent the willingness to accept the loan application, or to refer it for further review through more costly manual underwriting. However, those customers that do not meet the required minimum cut-off scores are likely to pay a stiff price. If their loan is not approved, the borrower in all likelihood is relegated to the higher cost subprime market. Subprime interest rates, on average, range from two to three interest points higher than those charged for loans approved by the AU systems. Subprime loans also generally entail much higher points and fees for the borrower than do prime loans.

Subprime loans are typically refinancings of existing mortgages and are made disproportionately to lower income, elderly, and minority homeowners. AfricanAmericans homeowners are nearly three times and Latino homeowners almost two times more likely to receive subprime loans than their white counterparts. Thus the stakes are great for borrowers, which reaffirms the importance of ensuring for the accuracy and fairness of the scoring systems that are used for making these loan decisions.

Increasingly, the GSEs and lenders are using risk-based pricing to make loan decisions in both the prime and subprime markets. And in fact, the difference in the cost of credit some someone with a high credit score and someone with a low score can be quite substantial. At current interest rates, for a $100,000 mortgage made for a property in Maryland, an applicant with a credit score in the highest tier (720 or higher) will qualify for a loan with an interest rate of 5.564%, carrying a monthly payment of $572.00. In contrast, an applicant with a credit score in the lowest tier (under 559) will qualify for a loan with an interest rate of 7.945%, carrying a monthly payment of $730. This means that over the life of a 30 year loan, the higher rate will cost the borrower with the lower credit score $56,924 in additional interest payments.

As credit scoring and AU systems are increasingly used to determine the cost of credit (as opposed to access to credit), new questions arise about the relationship between risk and price. Research by Freddie Mac, for example, suggests that many customers in the subprime mortgage market are being charged interest rates that are higher than would be required to cover the risk they pose to their lenders. AU systems permit the GSEs' entry into the higher end of the subprime market, which can reduce costs for borrowers. However, this "expanded approval" comes with a price, applicants that do make the cutoff for prime loans find themselves paying a higher price for their loan.

Questions remain about the validity and accuracy of scoring and the models used for mortgage lending

Fundamental questions remain about the validity and accuracy of scoring systems being used. These questions linger, in no small part, because the systems and the algorithms they use are proprietary, and held closely by the companies that develop them.

For one thing, the accuracy of the credit score generated by any scoring system rests on the quality, consistency, and completeness of the credit information going into the system. A study published last year by the Consumer Federation of America in conjunction with the National Credit Reporting Association looked at credit scores and the information that went into formulating these scores. The study found wide variations in the scores assigned to consumers based on credit information from each of the three major credit repositories. As many as one in three files had a variation in credit score of 50 points or more, and one in twenty had a range of 100 or more points. This led researchers to conclude that one in five consumers is at risk of being mis-classified into the subprime market due to inaccurate information in the credit reports. (CFA/NCRA. 2002. Credit Score Accuracy and Implications for Consumers).

Further, other research has raised concerns about whether certain creditors may manipulate the credit reporting system to prevent competitors from enticing their best customers away. Some lenders have deliberately failed to report current and accurate information about their borrowers to the credit reporting agencies. The consequence for the borrowers involved has been to depress their credit scores falsely and artificially. Information that creditors were gaming the system led federal banking regulators several years ago to take steps to discourage this practice. However, it is not clear whether financial institutions that are not federally regulated continue to engage in this practice. Issues about the methods used for computing scores have also been raised. For example, some research has found that developing bureau credit-scoring models through national population samples may omit potentially important variable relating to local and regional economic conditions. The study suggests that credit scores calculated from samples not adjusted for local and regional economic conditions could result in inaccurate credit

scores.

Moreover, other important methodological issues regarding the accuracy and fairness of computing scores for mortgage lending purposes still remain. These concerns tend to be of three kinds:

1) Concerns that low-income, minority borrowers, and persons living in older urban areas may be underrepresented in the bureau files. Consequently, the information provided for computing scores may not accurately portray the creditworthiness of underrepresented groups in the applicant pool and thus, may result in inaccurate

scores;

2) Concerns that the scoring models used typically omit certain nontraditional indicators of credit performance, such as rent, utility payments, and other non-traditional credit histories which are important components of credit performance for many lowincome and disproportionately minority applicants. Conversely, there are also concerns that the models fail to adequately take into account important positives or compensating factors, such as the use of pre-purchase and post-purchase housing counseling which many experts believe can affect projected risk.

3) Concerns that scoring models result in disparate impacts for protected classes and fails to adopt less discriminatory alternative measures. Disparate impact may occur in a credit scoring system when a variable used in the scoring system is facially neutral and applied evenly, but the variable disproportionately adversely affects a segment of the population protected by the fair lending laws (such as racial minorities).

This point is a particularly sensitive one since all parties - credit score providers, lenders, and the GSEs, quietly acknowledge that racial minorities, on average, have significantly lower credit scores than whites in the scoring models that are employed. In essence, the lower distribution of scores for minorities in means that credit scoring being used today disproportionately rejects minority applicants or means that on the whole they tend to pay more to obtain mortgage credit.

In response to these issues, the credit scoring industry and the proprietors staunchly defend that their systems are predictive of future loan performance and that scoring increases the accuracy of risk assessment. They insist that that they do not explicitly use prohibited factors, such as the borrower's race, ethnicity, age and gender in formulating scores. They point to some research that suggests that scoring can serve the interests of borrowers by expanding credit opportunities for many and improving efficiency of the credit review process.

Nevertheless, the key scoring models used in mortgage lending today, such as the Fair, Isaac & Co. and GSE systems have never been subject to independent verification to ensure that are indeed fair and unbiased and consistent with the nation's fair lending laws. The formulas for these models are closely guarded secrets and therefore, the methodological questions of the type that I have discussed have not been adequately addressed.

Discrimination has been a persistent problem in home finance markets in the United States. To be sure, the mere existence of disparate impact resulting from the application of scoring methodologies does not necessarily constitute the existence of discrimination or illegal treatment. However, given the legacy of lending discrimination, we believe that a high level of scrutiny should be required to ensure that the scoring models used today in mortgage lending are working a manner that is fully consistent with fair lending requirements.

Finally, let me emphasize that my testimony today focuses on the accuracy and fairness of scoring. I do not touch on a host of other real problems that may result from the improper use by creditors of scoring models. These include creditors that do not perform ongoing and effective oversight of the credit scoring model's performance. It also includes improper application of credit scoring models on products, particular subset of applicants, or geographic areas for which they were not developed and the inconsistent use of credit scoring models, including excessive overrides. All of which are real problems that may occur in today's marketplace.

What needs to be done?

De-mystify credit scoring by removing the veil of secrecy that currently pervades this industry.

First, we strongly recommend that Congress mandate the establishment of an effective and meaningful federal agency oversight process of all statistical scoring systems, including automated underwriting systems that are used for mortgage lending purposes. These reviews should be conducted on a regular basis and should focus on the fairness and validity of these systems. The results of these reviews must be released in a timely fashion.

HUD, which has oversight responsibility for Fannie Mae and Freddie Mac, is the only agency we know of that has undertaken a comprehensive review of the automated underwriting systems operated by the GSEs. Unfortunately, the results of this study which were completed more than two years ago is long overdue.

Second, we believe that consumers must have greater access to the scores that are being used to make credit decisions than they have now. Lenders may reveal to a consumer the score that is being used to evaluate their mortgage application, but this is generally too late for the consumer to do much about that score. In response to the California law that requires lenders to give customers a copy of their credit score, Fair, Isaac & Co. reversed its policy several years ago and began selling consumers their own scores. Customers may also obtain their scores from several of the three major credit repositories. Yet there is some question as to whether scores and the scoring model consumers are provided with represents the same ones that a lender may be using at any given time.

Lastly, we concur with the recommendations for improving the credit scoring industry contained in the recent CFA report on credit scoring accuracy. These include the following:

Require creditors to provide borrowers with a copy of the report resulting in adverse action on a consumer's credit standing.

Require the automatic re-evaluation of any adverse information resulting in a reduced credit score to determine its accuracy.

Strengthen requirements for complete and accurate reporting of account information to credit repositories with added oversight and penalties for non-compliance.

In sum, providers of credit scores should be required to share responsibility for ensuring the accuracy of the underlying data, for correcting that data, and for disseminating the correct information if requested by the consumer. Removing the mystery about credit scoring should be on everyone's agenda.

This concludes my testimony. I will be happy to answer any questions that you have.