PUBLIC FORUM: THE CONSUMER AND CREDIT SCORING

• creditscoring.com: To not take too much of the nation's time with whining and long-winded stories, asked five succinct questions and even provided them in writing (indeed-- even on this web site prior to the forum). The idea was to get the questions in before the noon break: "... as promised we would like to give the audience a chance to raise some questions and issues so that we can be sure to cover them this afternoon." The questions did not survive lunch. Wasted plane ticket.
• FTC: "I can't think of a subject that more consumers have expressed to us their desire to understand exactly what it is."

• Fair, Isaac: The Wholesome, Brave, Hard-Working Pioneers (introduced as those who "pioneered" -- see PR)
  • Answers the big question from creditscoring.com: What horrible thing will happen if you release the consumer's score to him?
  • Fair, Isaac's presentation: an attempt to either bore you to death, or, if you do pay attention, confuse the issue. Pseudo technical babble about something completely secret-- so there is no way to challenge their statements or methodology, anyway. Here, in a national presentation, they run out of time and skip though their own material like they are ad-libbing a freshman book report. But, just shut up and take what they feed you, America-- it's PROPRIETARY (as they say, often).
  • On Non-Water Cooler Thinking-- Cold, Hard, Serious, Number Crunching, Numbers Don't Lie, Just the Facts, "It's better to count than to guess" statistics:
    • "We saw all of those numbers, so where do they come from? Well, we do not sit around the water cooler at Fair, Isaac and kind of pull them out of our ears, which again may come as a surprise to people... All of the numbers that are in those scoring models are the result of very painstaking, detailed statistical analysis of real data."
    • NON-WATER COOLER MENTALITY: "And so in fact what we've said is, we don't count any inquiries in the 30 days before this particular score is computed. So you can have a hundred inquiries in the month before the score is computed and they won't count at all. And you can have any number of mortgage or any number of auto loans in any --I'm sorry. It's a 14 day period at any time before the score is computer, and they'll only count as one inquiry each."
      So, with all this statistical, analytical, multiple regressionistic, yadda, yadda, yadda calculating going on, why didn't their algorithms spit out some unusual number of days-- like 12, 17, or 26?
  • "For most kinds of credit, 700 or maybe a little bit up in the 700's. Anything above that is considered golden for most kinds of credit. Anything below about 550 is considered awful."
• The national credit bureaus:
  • Answer the big question from creditscoring.com: What horrible thing will happen if you release the consumer's score to him?: "It's an issue I'm not going to be able to address today."
  • A BIG BREAK FOR THE CONSUMER; SCORE ONE FOR THE FTC:
    "MS. TWOHIG: Thanks, Ray. I have one question about something Peter Mahoney said this morning. When he was arguing that the black box was a myth, which, I think, a lot of folks in the room since disagreed with, he said that you can contact each of the three repositories and they will give you what he called a feedback list. I'm not sure if that's the same as a factor list. And I was just wondering if you know anything about that?
    MR. CRESCENZO: The list of factors that are part of --that could be construed as a full list?
    MS. TWOHIG: I assume that's what he was referring to. Yeah. That's what he was referring to.
    MR. CRESCENZO: Well, I have a full list right here.
    MS. TWOHIG: So that is the case, that those factors that are used by the --
    MR. CRESCENZO: The whole list is available.
    MS. TWOHIG: Okay. I just wanted to clarify that. And if anybody called and asked for that, would you mail it out to them?
    MR. CRESCENZO: I would.
    (Laughter.) MS. TWOHIG: Call Ray.
    MR. CRESCENZO: We have been also --one of our other staff members has been working with the scoring development folks to try to expand the comment factor codes --the explanation codes --so that there is an educational message besides just the statement of lack of payment. So it's, you know, more of the evolution process that continues.
    MR. MEDINE: Ray, do you want to make that part of the record of this, so that people can have access to it and have a better understanding of how the process works?
    MR. CRESCENZO: The list?
    MR. MEDINE: Yes.
    MR. CRESCENZO: Yeah, sure."

    Now, go to his web site and try to find the factors "that could be construed" as a full list. Please send email if you find them. Send it if you don't, too.

    And, try the national credit bureaus' web sites:
    Trans Union
    Equifax
    Experian (be sure to "Ask Max")

    Try
    Freddie Mac
    Fannie Mae

    Try Fair, Isaac:
    www.fairisaac.com/servlet/SiteDriver/Content/929
    http://www.fairisaac.com/servlet/SiteDriver/Content/397#4

    Links from this page to the list from these credible sources will be provided when they get around to publishing it on the Internet. We can rank them in chronological order.

    But, don't hold your breath.

    While you're looking, try to find links to the FTC's (or any other) transcript of the Credit Scoring Forum. It may be hard, since "... 3 days prior to the meeting, you could check the following group calendar of events and find no record of this critical public forum: Fair Isaac, Associated Credit Bureau, Equifax, Experian, Trans Union, Fannie May, Fannie Mac."

• MR. GOLDBERG: "Joe Goldberg from the Pennsylvania Attorney General's office and the Bureau of Consumer Protection. I have a very serious concern with what you said, and I also have a very serious concern with the credit bureaus. Nobody wants to take responsibility for the information that you are transferring. What Fair, Isaac is doing and what the other systems do, is making a statement about the character-- the financial character-- of a consumer, and you are conveying that to a third party. It's not purely internal. And there is a reliance on that statement for better or for worse. What you do need to do is, if you're going to take that step, I think you have the responsibility to tell the consumer the basis for the statement about the character, and the effect that changes in their behavior will have on that statement. I disagree that you don't tell them that. You are making a statement and conveying that to a third party. You have an obligation to allow this consumer to act. And by just giving a list of four factors, if that consumer adds to the consumer's detriment, in my opinion that's an unfair or a deceptive trade practice."
  (Applause.)(no response from the opposition)
• Fannie Mae:
  • Myth of the Down Payment: "Because no longer is a low down payment loan by itself the highest risk alone in the bucket. And in fact, it may be --there may be plenty of opportunity to expand into, you know, a greater population of low down payment loans by using credit scores as a compensating factor."
  • Omigod, has it come to Approval By A Computer?:
    • "However, just as with credit scores, the use of Desktop Underwriter is meant to compliment the underwriter's judgment. It does not replace the underwriter's judgment."
    • Then on the same page: "An approved recommendation, which means that the loan meets Fannie Mae's credit underwriting guidelines and does not need further review by an underwriter."

      Eyewitness account: read from a prepared script.

• Freddie Mac - "High Energy" and Humor (or Comedy):

  "Credit Scoring in the Mortgage Industry: Myths and Facts"

  1. "MYTH: Only One Type of Credit Score is Used in Mortgage Lending"
     
  2. "MYTH: Credit scores make lending decisions"
  3. "MYTH: Credit scoring models are a 'black box' and we don't know how they work"

     "You can contact each of the three repositories, and they will give you a list of the feedback messages --and there's a long laundry list of them --that each of the three repositories use when they send a credit score."

  4. "MYTH: Credit Scores are not based upon a representative population sample"
     • "... we have confirmed that generic credit scores --the stuff that's in FICO scores that we've recommended --are based upon a representative sample based on some of the stuff that Pete showed you, which is that there is a 6.7 percent on minority area representation versus an 8.3 percent of the overall population."
     • On an Experian page: "Developed from a representative national sample of roughly 1 million Experian consumer credit histories, the Experian/Fair, Isaac Model:... "
  5. "MYTH: Credit scoring has not been independently validated"

     "That is not an accurate statement. The Federal Reserve Board --actually Bob Avery is here, and you can ask him about this. They obtained some credit scores. They tested it against low income populations, as well as a broad sample of populations. Their findings were published in the July 1996 Federal Reserve Bulletin. I'm still looking for my autographed copy."

     Looking for any copy.
     Freddie Mac quotes the Federal Reserve.

     Checking with the Federal Reserve on that:

     • MS. WELSH: Hi. Kristy Welsh, K-R-I-S-T-Y, W-E-L-S-H, and I'm with creditinforcenter.com. And I had a question about what Mr. Cook said, that the Federal Reserve Board regulates or somehow evaluates the FICO scoring. How is this done? How is it tested? Are we going through a statistical analysis? Do we take case A, B, C and B and run them through the scoring? Hey, we know for sure that it matches what it should match. And if so --I mean, you said that this was done. Is this published anywhere? Can we get a copy of it?
       MR. COOK: Well, I may have mis-spoke. I don't think I said that we evaluate credit scoring systems per se. I will say that HUD has currently embarked on a study of the scoring systems of the GSE's --Fannie and Freddie. No. I do --I can tell you that in 1996 --and, Bob can correct me if I'm wrong, if he's still here --the Fed did undertake an analysis of credit scoring systems in the sense of trying to determine whether or not there were solid correlations between low scores and poor performance. And generally they found that there were. That credit scoring systems do tend to be predictive of behavior. But that leaves a lot of room, I think, for issues today.
       MS. WELSH: Right. So they haven't been evaluated for bias or discrimination or any of these kinds of things?
       MR. COOK: No. This was not an exhaustive study of each factor in a given scoring system and a relationship. I guess one of the questions that I didn't quite follow up on with Pete, you stated that the statistics are available for your clients to understand and appreciate the various risks associated with each of the factors, so that if they're called upon to justify the inclusion of a particular factor, it's predictiveness can be quantified for them based on the statistics that you've done. But let me ask you this, sort of a further question. Recognizing --sort of taking off on Debby's approach --that we're talking perhaps more than just the legal requirements here. Recognizing that probably you don't have a legal obligation to do the rest of the analysis, to do the third stage of a --what did we call it --disparate impact analysis, that while there may be a justification for using something like finance companies --a number of finance companies --is that the least --does it produce the lowest impact, or is there another set of factors that you could have used that would produce a lesser impact?
     • Federal Reserve: "Credit scoring models are proprietary and consequently public scrutiny is precluded. Suppliers such as Fair, Isaac & Co. provide substantial data to support the validity of their scores but do not publicly release information about how the formulas work."

  6. "MYTH: Credit scoring is unfair to low and moderate-income borrowers"
  7. "MYTH: Credit scoring is unfair to minority borrowers"
  8. "MYTH: Credit scores are always predictive for every single person"
  9. "MYTH: People without credit scores cannot obtain a mortgage"
  10. "MYTH: Credit scoring reduces the availability of low-cost mortgage credit"

• Kirk Willison, Senior Vice President and Director of Government Affairs for Countrywide Home Loans:
  • "... And after listening to a lot of discussion today, I've become convinced that Congress was wrong, in that Congress should not have said that consumers really aren't entitled to their scores... "
  • SCORE ANOTHER ONE FOR THE FTC. The worst appears to be true-- it is contractual:
    "MS. TWOHIG: Kirk, I just want to make sure I understood one thing. Is it the case, though, that you don't give consumers their actual score when you explain to them the reasons why they might be denied?
    MR. WILLISON: Again, the agreement that we have with Fair, Isaac, is that we are not to release that score.
    MS. TWOHIG: Okay.
    MR. WILLISON: Our hands are tied."
• Marcia Griffin, founder and President of Home Free USA:
  • "They need to know their credit score, and they need to know it early... "
  • "... the first secret is that the process has changed. And the second secret is to get your credit score. Get your credit report immediately, not from the credit reporting agencies, because when you write, they don't give you your credit score. They need to really get their credit report with a credit score, so that they can understand where they are... "

TRANSCRIPT STARTS HERE

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 1
1 Page 2 3
2
1
2 P R O C E E D I N G S
3 -----
4 (9: 02 a. m.)
5 MR. MEDINE: Good morning, and thank you all for
6 coming. This is a tremendous turn out in the middle of
7 the summer on a very, very important issue to consumers.
8 And I would like to begin by introducing the Director of
9 the Bureau of Consumer Protection of the Federal Trade
10 Commission, Jodie Bernstein.
11 MS. BERNSTEIN: Thank you, David. Thank you for
12 the introduction. And I guess they thought I needed it,
13 because I see they've made two name tags for me, which
14 either means that I won't remember my name, or we need a
15 special introduction.
16 I'm delighted to be here this morning to welcome
17 all of you to what's going to be a very important forum.
18 And it's particularly important, as David said, that so
19 many of you were able to get here. I know a lot of you
20 travelled from out of town. And we particularly
21 appreciate it, because the diversity here in the room of
22 views and expressions is going to be critical to us.
23 I can't think of a subject that more consumers

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 2
2 Page 3 4
3
1 have expressed to us their desire to understand exactly
2 what it is. As someone said, when we get a call from
3 consumers --and we do get lots of calls about a variety
4 of subjects --it's as if there's a black box somewhere
5 and people --somebody somewhere is dropping information
6 into the black box and a score emerges. It's just almost
7 that mysterious to people, they tell us.
8 And then they want to know --and it really does
9 sound like the SATs --how can I improve my score. I
10 know that we won't --we won't produce a list that we can
11 tell consumers how to do that. It's much more complex
12 than that. But I do hope that with this group of experts
13 and the discussion we hope will occur today, that we will
14 all emerge with a good deal more knowledge and
15 understanding of what credit scoring is about. We know
16 it can be a huge benefit to consumers, and we will to be
17 able principally to be able to explain it to them and
18 ensure that they really do understand it and that they're
19 fairly treated.
20 So with the excellent panelists we have today,
21 again, I want to thank all of you for coming and for
22 agreeing to guide our discussion. I could not close
23 without noting that we have true experts to work with us

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 3
3 Page 4 5
4
1 this morning. And I believe our first speaker this
2 morning is the gentleman sitting to my right, am I right?
3 MR. MEDINE: He is. I wanted to make a
4 housekeeping remarks first. I'm David Medine, the
5 Associate Director for Financial Practices here at the
6 FTC. This forum, first of all, would not have been
7 possible without the two people to my left, Kellie
8 Cosgrove and Peggy Twohig. And I want to congratulate
9 them for all their tremendous effort in putting this
10 together.
11 (Applause.)
12 MR. MEDINE: Cynthia Lamb also was a tremendous
13 assistant in organizing this event, and she's sitting in
14 the back.
15 A couple of just logistical points before we get
16 started. First, there will be a transcript of today's
17 meeting, and as per Commission practice, a copy of the
18 transcript will be available on our web site at
19 www. ftc. gov once it's been prepared.
20 As Jodie mentioned, the purpose of today's forum
21 is to discuss a wide range of issues associated with
22 credit scoring. We want this to be a discussion in which
23 everyone has an opportunity to participate, although the

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 4
4 Page 5 6
5
1 first couple of sessions will be laying out some of the
2 background. But as the morning progresses, we're going
3 to encourage people to ask questions and to make comments
4 as we proceed, from the audience as well as from the
5 panel members.
6 And for those who are in the overflow rooms,
7 feel free to come by and join us with questions and
8 comments, although we may ask you to return to the
9 overflow rooms, given space considerations. But we don't
10 want people not in Room 432 to be denied the opportunity
11 to participate in today's discussion.
12 If you have any questions or need directions,
13 keep an eye out for the folks with yellow tags on them.
14 They're Commission staff, and they will be happy to
15 direct you to rest rooms, which are just outside and to
16 the left, or to our wonderful Top of the Trade, if you
17 are hungry and need a snack later on, on the seventh
18 floor.
19 It's my pleasure to start the morning off with a
20 good friend of ours, Pete McCorkell, to give us an
21 overview of credit scoring. Pete is the Senior Vice
22 President and General Counsel of Fair, Isaac and Company.
23 Jodie talked about the black box. Well, Fair, Isaac

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 5
5 Page 6 7
6
1 makes the box, and so this will be a wonderful
2 opportunity for us to get a peak inside the box and get a
3 better sense of how the scores --credit scores are
4 prepared.
5 I think we all discovered that credit scoring
6 had entered the mainstream when in an episode of Murder
7 One a couple of years ago, someone talked about their
8 FICO score as if everyone knew what that meant. And so I
9 think credit scoring moved into the mainstream at that
10 point.
11 Fair, Isaac, of course Pete's company, pioneered
12 the commercial development of empirically derived,
13 predictive models for the credit industry and has really
14 popularized and expanded their use. They are located in
15 California, and they have subsidiaries dealing with the
16 whole issue of decisions using scoring methodology.
17 Pete supervises Fair, Isaac's legal affairs. He
18 joined them in 1987, and has provided advice and
19 assistance to credit grantors in connection with consumer
20 litigation and regulatory proceedings. He is the
21 company's primary liaison with the federal credit
22 regulatory and enforcement agencies, including the FTC,
23 and has conducted seminars for us and other agencies to

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 6
6 Page 7 8
7
1 educate us about the scoring. And today he'll educate
2 you about the scoring as well.
3 Pete?
4 MR. MCCORKELL: Thanks, David. When I came in,
5 I used a line from Butch Cassidy and the Sundance Kid. I
6 said, can I move. So I'm going to get up and move.
7 Okay. I'm going to sort of approach credit
8 scoring a little bit like peeling back an onion. We've
9 got a lot to cover in the next hour, and so kind of put
10 on your seat belts and try to stay with me.
11 We're going to cover a number of topics, and
12 really in some sense the most important part of what I've
13 got to say is really bringing everybody's awareness about
14 what the nature of the credit decision is. Because an
15 awful lot of the alleged shortcomings of credit scoring
16 turn out to be problems that are inherent in the nature
17 of the credit decision and applied other ways of making
18 credit decisions.
19 I'm also going to talk about sort of the
20 differences between statistically based decisions and
21 judgmental decisions. I'm going to then go into a little
22 bit, hopefully in a very nontechnical way, about how
23 credit scoring systems are developed. I'm going to talk

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 7
7 Page 8 9
8
1 about different types of credit scoring systems. I'll
2 focus a little bit on the credit bureau scoring systems,
3 because I think that's probably what most people have in
4 mind when they hear the term credit scoring or FICO
5 scores. And then finally, if there's enough time, touch
6 very briefly on some legal and public policy issues
7 related to credit scoring.
8 Actually, one more. Casey Stengel actually had
9 it about right. Making predictions about the future can
10 be kind of dangerous. But in fact, that's what credit
11 grantors have to do. They have to make predictions about
12 how borrowers will behalf, how borrowers will repay that
13 obligation. And when you come right down to it, there is
14 no way to predict with certainty how any individual
15 borrower is going to behalf, how they're going to repay.
16 But what you can do, is that you can predict how
17 groups of borrowers will repay with reasonable certainty.
18 You can predict that if I have a thousand customers with
19 these characteristics, 900 of them, or 950 of them, or
20 990 of them will repay in a satisfactory manner and the
21 others won't. But nobody can predict that this
22 particular borrower will or will not repay in a
23 satisfactory manner.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 8
8 Page 9 10
9
1 And so what we're talking about is not
2 predicting with certainty the behavior of an individual,
3 but really trying to estimate the risk that different
4 borrowers pose. And risk by its very nature is assuming
5 that you're looking at a group of people, because
6 ultimately every borrower is either going to behave in a
7 satisfactory manner or they're not. They're either going
8 to repay in a reasonably satisfactory manner, or they're
9 going to go delinquent. Maybe go all the way into charge
10 off.
11 So at the end of the day with any particular
12 borrower, you can say, yeah, this was an absolutely great
13 borrower, or this was a terrible borrower. I should
14 never have extended that loan. But when you're making
15 that decision, you can't say for sure that this borrower
16 will or will not repay. And that really gets into kind
17 of the meat of credit scoring and what it's all about.
18 Basically there are fundamentally two ways of
19 making credit decisions. The sort of old fashion
20 judgmental type of decisions, where somebody takes a look
21 at whatever information is available, and sits down and
22 scratches his or her head, or pats his or her stomach or
23 whatever they do to think great thoughts, and comes to a

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 9
9 Page 10 11
10
1 conclusion that this is a loan that I'm willing to make.
2 The other type of decision process is --and in
3 many cases the two are actually used jointly in any given
4 case. But fundamentally the other type is using the
5 statistical method of making decisions, and that broadly
6 is what I'm going to refer to as credit scoring.
7 Now, a little bit later on I'm going to get into
8 some terminology. In the mortgage industry, at least,
9 credit scoring has come to have, for at least some
10 people, a somewhat more narrower definition. But at
11 least for right now, I'm using credit scoring in kind of
12 the broadest sense of any kind of statistically based
13 credit decision system.
14 So what are the differences between those two
15 types of decisions? Well, actually what I really want to
16 talk about first is what are the similarities. Both
17 methods assume that the future is in some large measure
18 going to resemble the past. Both methods, you're taking
19 a look at the applicant that you've got to make a
20 decision on today, and comparing that applicant with
21 qualities or characteristics of borrowers that you've had
22 experience with in the past, and making the assumption
23 that if today's applicant looks like the people who

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 10
10 Page 11 12
11
1 generally repaid in a satisfactory manner in the past,
2 then this applicant is likely to repay in a satisfactory
3 manner.
4 The days of Mr. Bailey and the Savings and Loan
5 from It's a Wonderful Life are long gone. Very few of us
6 are known personally to loan officers. Most credit today
7 is extended by mail, sometimes over the phone and
8 increasingly over the Internet. It's probably been 15
9 years since I've had an up front personal encounter with
10 a loan officer, and I suspect that the same is true for
11 most consumers.
12 Today you just don't have that friendly
13 interaction across the desk, and even when you do, you're
14 likely to have --you're not going to have it with Mr.
15 Bailey who knows everybody in town and has known you and
16 your parents and your grandparents. You're going to have
17 it with somebody that doesn't know you from a hole in the
18 wall.
19 And so while your own past history is certainly
20 an important part of the credit decision, in many ways
21 whether somebody is making judgmental or statistical
22 decisions, they're comparing what they know about you to
23 what they know about other applicants and how those other

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 11
11 Page 12 13
12
1 applicants behaved as borrowers.
2 And then finally, it gets back to the idea of
3 trying to grant credit to acceptable risks. Now, a lot
4 of people making judgmental decisions may sort of fool
5 themselves into thinking well, I know for sure this
6 borrower won't repay, or I know that if I extended credit
7 to this guy, he would not repay. But in fact, you just
8 can't make those individual decisions with any degree of
9 certainty, so in either case you're really trying to hit
10 that acceptable degree of risk.
11 Okay. Now, what are the differences? Well,
12 credit scoring really defines that degree of risk. It
13 allows you to come up with a numerical measure of the
14 degree of risk that any borrower presents. It lets you
15 rank order those borrowers with respect to one another.
16 And I'll get into that a little bit later.
17 It explicitly lets you make those decisions
18 based on a degree of risk. A lender can say, I can make
19 money if I extend credit on this product to people where
20 10 of them will repay for every one that won't, and
21 therefore I have to have at least a 10 to one odds of
22 repayment in order to make money. So I'll extend credit
23 to everybody whose odds are 10 to one or better. If

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 12
12 Page 13 14
13
1 they're worse than 10 to one, I'm not going to extend
2 credit to those people.
3 It lets you track how those borrowers perform
4 based on the criteria that you're using in the credit
5 scoring system. It lets you make adjustments in your
6 policy. If somebody is making judgmental decisions and
7 the boss comes in and says, our losses are too high,
8 we've got to tighten up, every credit officer is going to
9 have a different idea of what tighten up means. Or if
10 the boss says, gee, our market share is slipping, we're
11 losing business to the competition, we've got to loosen
12 up, every loan officer has to figure out what loosen up
13 really means.
14 With credit scoring somebody could come in and
15 say, we need to tighten up. Let's raise the cutoff by
16 five points. And because of the statistical work that
17 goes into the system, the management knows that if they
18 raise the cutoff by five points, they'll reduce the
19 acceptance rate by X amount and also reduce the
20 delinquency rate by Y amount.
21 And then finally, credit scoring allows the
22 automation of a lot of credit decisions, which I think
23 some of the other folks are going to talk about, but

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 13
13 Page 14 15
14
1 which has really been a big part of the acceptance of
2 credit scoring. Certainly in small business lending, for
3 example, probably more people have gone to credit scoring
4 for the benefits of automation.
5 In small business lending, for example, the
6 typical statistics we hear are that it takes 10 to 12
7 hours of loan officer time to process a small business
8 loan judgmentally. Using credit scoring and automated
9 underwriting, it can be done in 15 to 30 minutes. Now,
10 if you can take 10 or 11 and a half hours out of the
11 process --and that's not data entry time. That's fairly
12 high paid time. If you take that much time and salary
13 expense out of the process of making a small business
14 loan, you can make a lot more small business loans,
15 especially small dollar small business loans, than you
16 could make if you were doing it judgmentally.
17 One way to sort of think about the differences
18 is that probably in most cases, even somebody making a
19 judgmental decision has a list of factors that they're
20 going to look at when they are reevaluating an applicant.
21 And, again, to just get some lingo in here, what we call
22 characteristics are really the different types of
23 information that are looked at the decision process.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 14
14 Page 15 16
15
1 Another way that you can think about
2 characteristics is that you might think that each
3 characteristic implies a question that you could ask and
4 then get the answer, either from a credit application or
5 from a credit report.
6 Next slide, please.
7 And so in this case, at the top of the page
8 we've got some information that you might get from a
9 typical credit application, and then down towards the
10 bottom below the blanks are a number of characteristics
11 that you might get off of a credit bureau report. And
12 somebody making a judgmental decision is going to be able
13 to go through that list of characteristics, and probably
14 about the best that they're going to do is be able to say
15 well, this guy looks pretty good on this, but not so good
16 on that.
17 In this case, time at present address looks
18 good, and time on job looks good. Oh, but this guy is a
19 renter and not a homeowner, so that's not so good. Debt
20 ratio is good. He's got a good bank reference. The N/ A
21 after age doesn't mean that there's no age. It means
22 that legally age is not supposed to be used in judgmental
23 credit decisions because of ECOA in Rate B. But they do

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 15
15 Page 16 17
16
1 allow it to be used in scored decisions.
2 And so you just go through the list like that,
3 and at the end of the process judgmentally you say, oh,
4 this is an acceptable borrower, and that again implies
5 that this is a borrower who poses an acceptable level of
6 risk. You couldn't begin to site the odds of that
7 borrower if you evaluate that borrower using judgmental
8 processes.
9 With credit scoring --and again here is another
10 thing to keep in mind. You're probably going to go
11 through very much the same list of factors that somebody
12 would go through judgmentally. And that's not just sort
13 of, you know, conjecture on our part. One of the things
14 that Fair, Isaac does, is when we build a credit scoring
15 system for somebody that hasn't used credit scoring
16 before, we also build a statistical model of their prior
17 judgmental decision process.
18 And so we can actually find out how --what
19 factors at least appear to be important in that
20 judgmental decision process, and almost invariably will
21 find that very much the same factors are being considered
22 --at least apparently considered --in the judgmental
23 process as wind up being considered in a credit scoring

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 16
16 Page 17 18
17
1 system.
2 And so the list of factors, the types of
3 information that are considered, are very similar.
4 They're not --again, other than age, which for the legal
5 reasons can't be --or least maybe more accurately, isn't
6 supposed to be used in judgmental decisions. Other than
7 age, the types of information considered are going to be
8 pretty much the same.
9 But with a credit scoring system, for each of
10 those questions that are implied by the characteristics,
11 each possible answer to that question is going to have a
12 number of points assigned to it. And so you could add up
13 all those points and you get the overall score. In this
14 case, you get the same decision, that is an acceptable
15 borrower. In other words, a borrower that poses an
16 acceptable level of risk.
17 But you get one very important additional piece
18 of information, and that is, you get told this borrower
19 is 11 to one. In my example earlier, I said we could
20 afford to extend credit to people at 10 to one odds.
21 This guy is 11 to one, so we'll take him.
22 Next slide, please. This is a sample --and let
23 me stress a sample --of what a credit --the guts of a

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 17
17 Page 18 19
18
1 credit scoring system, the credit scorecard itself, might
2 look like. It is not an actual credit scoring scorecard,
3 so don't go home and try to score yourself on this and
4 get into a major tiz about your score.
5 I'll just use it to illustrate a couple of
6 points here. First of all, in this case we've got the
7 characteristics down the left hand column. Again, those
8 questions --those types of information. And now across
9 in each row next to the characteristics, we've got the
10 different attributes. The different possible answers for
11 each of those questions, and you'll see that each of them
12 have a different number of points assigned.
13 A couple of things that I want to point out on
14 here, you'll see --and again above the jagged line are
15 pieces of information typically that you get from a
16 credit application, and below the line are
17 characteristics that you would typically get from a
18 credit report.
19 For all of those application characteristics,
20 and actually implicitly also for the bureau
21 characteristics, at the very far right of each column --
22 each row --there is a box labelled NI. That stands for
23 no inform, which is Fair, Isaac speak for no information

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 18
18 Page 19 20
19
1 value. That may be different than the questions not
2 answered.
3 If you look at the fifth line down, Department
4 Store/ Major Charge Card, you'll see there is a box for no
5 answer. Sometimes when a question is asked on an
6 application and the applicant leaves that blank, the fact
7 that they left it blank tells you something. On the
8 other hand, in a lot of cases credit lenders change
9 applications, or for one reason or another a particular
10 piece of information may not be available for a given
11 individual.
12 We have figured out that it would not be fair to
13 assume that that means that person should get zero points
14 for that piece of information. What they get is what we
15 call the no inform points, which if you go to the next
16 slide, is really can of the neutral value. No inform
17 means that there is no information value to say that this
18 applicant is either better or worse than the average
19 applicant based on that particular piece of information.
20 So if you have a characteristic like age, where
21 you've got, you know, kind of a nice, smooth numerical
22 scale in most cases, there is going to be some age value
23 that's exactly equal to the no inform points. For a

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 19
19 Page 20 21
20
1 given applicant, not because the applicant refused to
2 answer something, but because you didn't ask the question
3 on an application, if you're missing a piece of
4 information, it will go to the no inform points.
5 In other words, it says we don't have a piece of
6 information on this factor that tells us that this
7 applicant is likely to be risky or less risky than the
8 average applicant who walks through the door. So that's
9 what the NI or no inform points are about.
10 Okay. We saw all of those numbers, so where do
11 they come from? Well, we do not sit around the water
12 cooler at Fair, Isaac and kind of pull them out of our
13 ears, which again may come as a surprise to people. I
14 read something just last week that talked about the
15 scoring model assumes that. Well, scoring models don't
16 assume anything. All of the numbers that are in those
17 scoring models are the result of very painstaking,
18 detailed statistical analysis of real data.
19 The first thing that we have to do in developing
20 a scoring model is to get a sample of borrowers where we
21 can say, this borrower's performance is known to the
22 extent where I'm willing to say that if I knew when I
23 made the decision what I know now about how that borrower

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 20
20 Page 21 22
21
1 has actually performed, I could say either I would make
2 the same decision to extend them credit, or if I had
3 known then what I know now, I wouldn't have extended
4 credit.
5 And that's what we mean by goods and bads here.
6 I'll use that term again probably throughout the rest of
7 my hour here. That's not a moral judgment or anything
8 else. In some senses, the good and bad refers to the
9 credit grantor's own decision. Did I make a good
10 decision in extending credit to that borrower, or did I
11 make a bad decision. Or in a sense in kind of the way we
12 define it, if, again, knowing how that borrower actually
13 performed you would still extend them credit, then that
14 counts as a good. If you knew how they would perform and
15 you wish you hadn't extended them credit, then that
16 counts as a bad.
17 For the people that you've been accepting under
18 --and let's assume this is somebody that's been using
19 judgmental decisions. For the people that have been
20 accepted before, of course it's very easy to figure out
21 who the goods and the bads are. You just wait a couple
22 of years, depending on the type of credit product, and
23 you see how they perform. You ask yourself, am I glad I

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 21
21 Page 22 23
22
1 extended credit or not.
2 And so you can count the known goods and bads
3 out of that Accepts on the right hand side there. Again,
4 just to use nice round numbers here, we're assuming we
5 start out with 10,000 people. Under the old decision
6 process, 7,000 were accepted and 3,000 were turned down.
7 For the 7,000 Accepts, figuring out the goods and bads is
8 just a matter of counting based on how they actually
9 performed.
10 But in building a scoring system, we don't want
11 to ignore these 3,000 rejects. The example I gave of
12 somebody saying I've got to have 10 to one odds to make
13 money means that somehow if you could improve your
14 decision process, if you could make sharper decisions
15 about individual risk level, for every 11 people you turn
16 --if you were turning down people that were below 10 to
17 one, for every 10 people you turn down, in certain groups
18 at least, as many as nine of them would have turned out
19 to be good customers if you would have accepted them.
20 And, again, that's an important concept to grasp
21 for lots of types of --in fact, virtually any kind of
22 credit product. To set a cutoff score for economic
23 efficiency, the credit grantor is going to wind up

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 22
22 Page 23 24
23
1 setting the cutoff score at a point where more of the
2 people that are turned down would have been good, and
3 sometimes by a very large majority. Sometimes it may be
4 10 to one. It may be even more than ten to one. Odds
5 are the point where it becomes unprofitable, because the
6 loss for one bad customer is a lot greater than the
7 profit from any single good customer.
8 And so when somebody says, you know, our cutoff
9 is 415, that doesn't mean that everybody who scores 414
10 and lower is going to go delinquent, and in fact at 414,
11 an awful lot of those people would probably have
12 performed just fine if you had accepted them. But it's
13 not economical given the facts of that particular
14 portfolio to extend credit to those people whose odds are
15 worse than that.
16 We don't want to ignore them, because in fact we
17 know that there is good business in those prior rejects.
18 And so what we go through is a process that has a number
19 of names. The one that I think is most descriptive is
20 Reject Inference. For those of you who paid attention to
21 your high school math classes, you will remember a
22 process called extrapolation, where you could infer where
23 a point was on a line where you didn't actually have any

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 23
23 Page 24 25
24
1 sample points. It went beyond the ends of the line when
2 you're putting stuff on graph paper.
3 Remember when you were a junior or sophomore in
4 high school, you were putting stuff on graph paper, and
5 if you only had samples, they were sort of in the middle
6 of your graph. You could nevertheless extrapolate to
7 where some points would be outside where you had actual
8 samples.
9 Well, Reject Inference, you can think of that as
10 kind of a multi dimensional form of extrapolation, where
11 again we look at the rejects. We see how they compared
12 on all of the relevant characteristics to the Accepts,
13 and we infer from that, we estimate from that, we
14 extrapolate from that, how they would have performed.
15 And indeed we see here, that of those rejects,
16 the odds were about nine to one. They're almost --
17 they're about nine times as many inferred goods as there
18 are inferred bads in the Rejects. So to build a scoring
19 system, we fold them back in with the known goods and the
20 bads, and, again, hopefully the prior decision process
21 made some sense.
22 And so the odds from actual performance of the
23 people that were accepted, the overall odds for that

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 24
24 Page 25 26
25
1 group --not the odds at cutoff. But the odds for that
2 group as a whole are 24 to one. The odds for the Rejects
3 are 8.7 to one. And again, I sort of use these numbers
4 because they will work out well for my next example. The
5 overall odds for that population turn out to be 16 to
6 one.
7 So if all that I know is that somebody has
8 walked in the door in this population, I can say, well,
9 if I extended credit to everybody that walks in the door,
10 for every 16 people that paid in a satisfactory way, one
11 of them wouldn't. And that's actually pretty --that's
12 pretty high odds. More typically, in most populations
13 we're going to see something --in the population odds,
14 we're going to see something on the order of ten to one,
15 eight to one and something like that.
16 But as I said, that 16 to one number happens to
17 work well for my next slide and actually several slides
18 after this. So we've now sort of divided the world into
19 the goods and the bads. There aren't any uglies, just
20 goods and bads. So now we continue the process of a very
21 careful analysis of actual data, and we see how people --
22 look at each characteristic, and we look then at each
23 attribute of each characteristic, and we see how those

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 25
25 Page 26 27
26
1 people perform.
2 And my first example here is age. And we'll see
3 that for people under 30, that group had 10 percent of
4 the goods, but 40 percent of the bads. And so the
5 information value for that attribute is one to four. And
6 again, these numbers here are in percentages, not in
7 absolute numbers. But if all I knew about somebody is
8 that they're in this population and that they are under
9 30, I can multiple the population odds of 16 to one by
10 the information odds for that one isolated piece of
11 information of one to four, and the first line at the
12 bottom there, I find out that this person is four to one.
13 It's the 16 to one multiplied by one to four.
14 On the other hand, if I have somebody who is
15 over 50, you've got exactly the reverse situation.
16 Again, this is just a sample --an example. There 40
17 percent of the goods are over 50 and only 10 percent of
18 the bads. So the information odds there are the other
19 way around. It's four to one. Multiple that by the
20 population odds of 16 to one, and I can say people over
21 50 --my best guess, again, just based on this one piece
22 of information, age, are that they're 64 to one.
23 Well, now I've started spreading people out away

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 26
26 Page 27 28
27
1 from that overall population odds of 16 to one. I've got
2 the younger borrowers over here riskier than the general
3 population, and the older borrowers a lot less risky than
4 the general population.
5 Of course, we don't look at just one factor. We
6 may throw in another factor, on the next slide, such as
7 own or rent, or residential status. And again, not
8 terribly surprising, the owners are better than average.
9 They're two to one. The renters are worse than average.
10 They're one to two.
11 And then on the next slide, I put those two
12 factors together and things started to get really
13 interesting, because now if I have an under 30 renter,
14 I'm taking again the population odds of 16 to one, the
15 information odds for under 30 of one to four, the
16 information odds for renter of one to two, and I multiple
17 that out, and now that under 30 renter is only two to one
18 odds. So if I lend money to three under 30 renters, I
19 can expect two of them to repay in a satisfactory manner
20 for every one that doesn't.
21 On the other hand, if I take my over 50
22 homeowner and multiple that out, now I've got odds of 128
23 to one. So I'm pulling different groups of that

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 27
27 Page 28 29
28
1 population sort of further away from that center line by
2 adding the information value from these various factors.
3 Now, this example makes the assumption that all
4 of these factors are independent. Those of you who
5 remember the O. J. Simpson trial and the DNA evidence,
6 you'll remember there were some pretty staggering numbers
7 that the DNA experts cited of 9.5 billion to one odds
8 that these DNA markers could have belonged to anybody
9 other than O. J. Simpson.
10 Well, they were going through very much the same
11 process with a large number of individual DNA factors --
12 DNA characteristics, if you will. And based on the
13 assumption that they were independent, that's how you get
14 those kind of staggering numbers like 9.5 billion to one.
15 And if all of these factors in a credit decision were
16 also independent, you might get some pretty staggering
17 numbers.
18 That's not the way it works in credit. And in
19 fact, what we'll find is that there are correlations --
20 in some cases very high degrees of correlations --among
21 the different factors that we look at, so we've got to
22 correct for that. We don't assume that they're all
23 separate. We certainly wouldn't ever make a credit

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 28
28 Page 29 30
29
1 decision using just age, or just own or rent, or just
2 those two factors, because in fact one of the things we
3 know --and it's pretty intuitive if you think about it
4 --is that as people get older, they're more likely to
5 become homeowners rather than renters.
6 You know, those of you who are over 30, you
7 know, remember back when you were in your 20's. Probably
8 most of us spent a few years as renters in our 20's, and
9 the chances of becoming homeowners get a lot higher as we
10 get older. And also, except for those who are in their
11 40's and still living with mom and dad, the time at
12 address is also going to go up as you get older.
13 And, you know, in a lot of cases, about the same
14 time that you get old enough to be legally able to borrow
15 anything, you're likely to leave home to go to college or
16 join the Navy and see the world or whatever, and so your
17 time at address takes a big hit. And then as you get
18 older, you're likely to be at the same address longer.
19 Homeowners, of course, are less likely to move than
20 renters, so there is a large correlation between time at
21 address and home ownership, as well as with age.
22 So these overlapping circles here, if you think
23 of the size of each circle independently as the

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 29
29 Page 30 31
30
1 information value in that particular characteristic, the
2 overlap just suggests that because of these correlations,
3 while each of those factors taken individually may have a
4 fairly large information value, as you start adding these
5 correlated characteristics, you then have to correct for
6 that overlap.
7 And on the next slide, I throw in one more
8 factor. In this case, time on job. Well, of course,
9 time on job is likely to be correlated with those other
10 three factors. Again, unless your parents forced you
11 into child labor in the family business, if you're in
12 your 20's you're probably not going to have been at the
13 same job for very long. Your time at address and time on
14 job frequently go together, because if you get a new job,
15 in a lot of cases you've got to move, or maybe you have
16 an opportunity to move, and again, the home ownership.
17 So while time on job standing on its own has a
18 reasonable information value, what this represents simply
19 is that where you've already got those other factors --
20 kind of stability characteristics, age related
21 characteristics in the system --adding one more factor
22 of the same kind just brings very little additional
23 predictive value.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 30
30 Page 31 32
31
1 And that's important for a number of reasons,
2 because typically in building a scoring system, we'll
3 start out with maybe as many as 50, and maybe as many as
4 a hundred, characteristics that independently have a
5 reasonable amount of information value. But because of
6 these correlations --these overlaps --as we actually
7 start building the system and we start getting a few
8 characteristics with a lot of information value in there,
9 a lot of those additional characteristics won't get into
10 the final system because of these overlaps.
11 Nevertheless, all of the information value there is still
12 captured by the characteristics that are in the system.
13 And so frequently somebody that's not sort of
14 familiar with the concepts of credit scoring comes along
15 and says, well, gee, this system doesn't have time on job
16 in it. I know that's predictive with credit performance.
17 There is something wrong with your credit scoring system.
18 No, actually the information value is in there, but it's
19 just picked up by those other factors that have
20 correlated with time on job. And so sometimes somebody
21 will come along and say, well, but you didn't consider
22 such and such. It's not in your scoring system. You
23 didn't consider such and such about me.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 31
31 Page 32 33
32
1 Well, we may have considered that implicitly
2 because of all of these other factors that are in the
3 scoring system, and so for a number of reasons, they're
4 characteristics that people try to keep out of scoring
5 systems. One of them --something that, you know, sort
6 of might be a very obvious characteristic is income. It
7 sounds like a nice numerical characteristic. It's
8 probably got something to do with how people are going to
9 be able to pay their bills.
10 But it turns out that in practice income is
11 difficult to work with, because no matter how explicitly
12 the application asks the question, people insist on
13 answering it in different ways. You may ask for monthly
14 pre-tax income and you get semi-weekly, before tax and
15 after tax, and you get with and without child support and
16 alimony. And you get the rental property, even though
17 there is another box on the application for that. And it
18 actually turns out to be a very messy characteristic.
19 And so somebody says, well, you didn't ask for my income.
20 How could you possibly make an intelligent credit
21 decision. Well, it's because of these correlations.
22 If we go to the next slide, sort of the goal of
23 all of this, as I said, we're trying to separate --move

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 32
32 Page 33 34
33
1 those people apart on the risk scale away from that
2 average that you know just because they walk in the door.
3 And what we really want to do in building this credit
4 scoring system is come up with a system that will
5 separate the eventual goods from the eventual bads by as
6 much as possible, so that --now there will always be
7 some overlap in the distributions.
8 Actually this doesn't quite do it justice,
9 because in reality the distribution of goods would go all
10 the way down here to the bottom of the scale, and the
11 distribution of bads would go all the way up to the top
12 of the scale. But what we're trying to do is move those
13 two distributions as far apart as we can, so that the
14 bulk of them in my nice bell shaped curve here --and of
15 course life isn't usually quite that perfectly drawn, but
16 it kind of approximates that.
17 What we're trying to do is move those two curves
18 as far apart as possible, so that we can set a cutoff
19 score that turns away as many of the bads as possible,
20 while turning away as few of the goods as possible. Now,
21 these are in percentages. Hopefully in any population
22 the goods would be a lot smaller in absolute numbers --
23 the bads would be a lot smaller than the goods in

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 33
33 Page 34 35
34
1 absolute numbers. So these are in percentages, which is
2 why the curves are the same size.
3 But you can see at the top where the two curves
4 are a little further apart, you can set a cutoff that
5 cutoffs most of the bads and only a very small percentage
6 of the goods. Down at the bottom, the two distributions
7 are closer together, and so to turn away the same
8 percentage of bads, you've got to cut off twice as many
9 of the goods.
10 Now, that's the goal of the credit scoring
11 system. There is no absolute right or wrong, and we call
12 that measure divergence, just how far apart those two
13 distributions are. There is no right or wrong answer for
14 that. The goal is to get those two distributions as far
15 apart as possible in any given case, because then it lets
16 you make more efficient decisions at cutoff.
17 For example, for a college student population,
18 college students, you know, all look at lot alike in
19 terms of credit factors. Again, unless they're still
20 living at home, they're all renters. They've all been at
21 the address for a short time. If they've got a job at
22 all, they've been on the job for a very short time. They
23 have very little credit history, etc. So you might --if

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 34
34 Page 35 36
35
1 you had a college student population, you simply couldn't
2 get the two curves as far apart as you could for a more
3 general population.
4 Next slide. Now, technically what credit
5 scoring systems do is to rank the order of consumers by
6 risk, because they don't consider things like economic
7 changes. If the economy goes south, people that scored
8 200 in this example that equaled odds of 20 to one, if
9 the economy takes a real downturn, those people that were
10 scoring 200 may find themselves with odds of 18 to one or
11 15 to one.
12 And so we don't purport to provide an absolute
13 odds quote, but, of course, what every credit grantor
14 wants to do is to turn that score into an odds quote.
15 And there are ways that they can do that, both from the
16 development statistics and also in terms of watching how
17 their own population performs.
18 So we get through all of this, and somebody is
19 probably still out there saying, oh, gee, but why do you
20 think scoring can do a better job of predicting credit
21 behavior of estimating risks than judgmental decisions.
22 It's real simple. It's better to count than to guess.
23 And fundamentally that's the difference between

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 35
35 Page 36 37
36
1 how somebody using a credit scoring system is making
2 decisions. They're counting very carefully the prior
3 performance. They're taking a very hard look at the
4 prior performance of different kinds of people,
5 separating that into individual characteristics, and then
6 comparing that in a very disciplined way to what you know
7 about the next applicant through the door.
8 If somebody is making judgmental decisions,
9 they're trying to keep all of that in their head and, you
10 know, there is no way that you can juggle 10 or 12
11 different characteristics, each of which has three to six
12 different attributes. There is simply no way that you
13 can juggle that much information in your head, because if
14 you have 10 or 12 characteristics with three to six
15 attributes each, you have 10 to 20,000 possible
16 combinations in there. There is no way that somebody
17 judgmentally can keep straight 10 or 20,000 different
18 combinations of attributes.
19 And so to a large extent, they're guessing.
20 They're taking --making an informed guess, hopefully in
21 most cases. But really they're guessing at how today's
22 applicant compares to the applicants that they have seen
23 in the past and for whom they have actual performance.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 36
36 Page 37 38
37
1 Okay. Real briefly I want to talk a little bit
2 about different types of credit scoring systems.
3 Historically most credit scoring systems were built for a
4 particular credit grantor, what we call custom
5 scorecards. The next category of borrowed scorecards is
6 really not terribly relevant these days, although in a
7 few cases it might be possible for one credit grantor to
8 borrow a system from a different credit grantor.
9 What is becoming very common, and again probably
10 what most people think of when they hear about a credit
11 scoring system, are credit scoring systems built from
12 data across multiple credit grantors. Now, of course,
13 the biggest example of that are the credit bureau
14 systems, that I'll talk about in more detail, and what we
15 call pooled system.
16 And then in a very few cases, somebody may come
17 along and want to do something kind of entirely new and
18 different. It doesn't happen in the United States,
19 again, very much these days. We do this more overseas.
20 Sometimes somebody will come along and want to do
21 something so new and different. They say well, we don't
22 have any data on this. We don't have any data on similar
23 portfolio. Fair, Isaac, you've been in this business for

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 37
37 Page 38 39
38
1 40 years, what can you do for me.
2 And then in that case we can sit around the
3 water cooler and say, well, based on our 40 years of
4 experience in lots of different systems, here are some
5 guesses we can make about the kind of borrowers you're
6 going to see. We can build a judgmental --what's really
7 a judgmental scorecard, or what we call a launch
8 scorecard, just to get them going until they've got some
9 data.
10 Next one. The information used in a credit
11 scoring system can come from a variety of sources. Of
12 course, the most common are credit reports and credit
13 applications. A bank may have prior experience with a
14 particular customer. Sometimes demographic information
15 may be useful. For existing customers the issuer's
16 billing file has a lot of good information. And in
17 secured lending, the terms of the deal, the amount down
18 and things like that, can be very useful.
19 Next. Credit scoring can also be used to make a
20 variety of different types of decisions. Last year there
21 were about three billion pre-screened credit card
22 solicitations mailed in the United States. Probably 2.9
23 something billion of those were credit scored. Believe

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 38
38 Page 39 40
39
1 it or not, not everybody in the world gets those pre-
2 screened solicitations, and typically one of the screens
3 they go through is credit scoring.
4 New applications, again, is probably what most
5 people think of in terms of credit scoring, making a
6 decision on an application for new credit. On the other
7 hand, if you pay attention if you own a credit card or
8 two, you'll notice that these days the issuers, for most
9 of you at least, try to do nice things once in a while.
10 They send you that letter that says because
11 you're such a wonderful customer, you don't have to make
12 the minimum payment this month. Or because you're such a
13 wonderful customer, we're going to increase your limit.
14 Or if you should happen to be standing at the cash
15 register with a purchase that's going to put you over the
16 limit, they want to know whether or not they're going to
17 authorize that transaction.
18 And finally when that piece of plastic expires,
19 they want to decide should we reissue it at all. If so,
20 should we reissue it for six months or for three years.
21 And so they're using decisions to manage those existing
22 --they're using credit scoring to manage those existing
23 accounts. They may also be using credit scoring to try

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 39
39 Page 40 41
40
1 to decide whether or not to try to sell you another type
2 of product, based on what they know about your behavior
3 on the one you've got.
4 And finally, in certain kinds of lending,
5 especially mortgage lending, lenders frequently resell a
6 lot of the portfolio. The lender is using credit scoring
7 to decide which parts of the portfolio it wants to sell
8 versus which ones it wants to keep. And the buyer or
9 investor, on the other hand, may be using credit scoring
10 to decide which loans it is interested in buying.
11 There are also a number of different performance
12 definitions. Up until now I've sort of talked about
13 good-bad or really a scoring system designed to predict
14 any credit delinquency. We can --and there are systems
15 designed specifically to predict the risk of bankruptcy,
16 rather than other types of delinquency.
17 If somebody is making a mailing --a pre-screen
18 mailing --they want to know whether somebody is likely
19 to respond, as well as if they do respond and you issue
20 them a card, will they pay. There are a lot of people
21 who are absolutely great credit risks who are literally
22 not worth the price of a stamp for a pre-screen mailing,
23 because they are so unlikely to respond.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 40
40 Page 41 42
41
1 And actually I'm kind of in that category. My
2 wife gets a lot more pre-screened mailings than I do,
3 because in fact we have sort of different patterns of
4 credit uses. And so somebody looking at her credit
5 bureau file is more likely to say this is somebody that
6 might be interested in another credit card. McCorkell
7 hasn't gotten another credit card in 15 years. We're not
8 going to waste a stamp on him.
9 Okay. Really just kind of an intro for the next
10 panel, scoring today in the mortgage industry, which is
11 probably as David and Jodie mentioned, is really how
12 scoring came to kind of the front of public
13 consciousness. Scoring is really used in a couple of
14 different ways. And, again, there is some terminology
15 here.
16 A lot of the mortgage industry refers --when
17 they say credit scores, they're referring to scores built
18 just on credit bureau data that are installed at the
19 credit bureau. When you order a credit report, a score
20 is generated. And when people in the mortgage industry
21 talk about credit scoring, they may mean what I would
22 probably refer to as bureau or credit bureau scores.
23 And most of those credit scores or credit bureau

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 41
41 Page 42 43
42
1 scores used in mortgage lending, interestingly enough are
2 what I labelled here as generic scores, which are
3 designed to predict the risk of any credit delinquency.
4 There are also some mortgage-specific credit bureau
5 scores that were designed specifically to predict the
6 risk of mortgage delinquency. Actually what we found is
7 that the generic scores do a very good job of predicting
8 mortgage at risk, as well as other types of credit risks.
9 And then finally, what I've labelled
10 comprehensive mortgage scores, I think a lot of folks
11 just refer to as mortgage scores, where they are looking
12 not just at the credit bureau --the credit history piece
13 of information, but they're also looking at application
14 information and the deal terms --the amount down and
15 that kind of thing --in making a more comprehensive
16 decision. I think the next panel will get into that a
17 little more.
18 So what I would like to turn to next are
19 specifically the credit bureau scores. And here, first
20 of all just kind of broadly, are the types of information
21 that are considered by the credit bureau scores. Now,
22 these are scores that Fair, Isaac and others --there are
23 other scoring developers out there. I don't know why

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 42
42 Page 43 44
43
1 anybody would use them, but some people do.
2 So there are other flavors of credit scores, and
3 occasionally somebody will use a FICO score to mean any
4 kind of credit score or any kind of credit bureau score,
5 even though FICO happens to be a trademark of ours. It's
6 a little like Kleenex. You don't always look at the
7 brand label when you walk into the store and buy a box of
8 Kleenex. You just want tissues to blow your nose. FICO
9 scores have sort of gotten the same kind of generic use
10 of the brand name.
11 These are scores that are running just on the
12 credit bureau data, and they're looking generally at five
13 types of information. And they're really listed here in
14 the order of importance. First is previous credit
15 performance. Has this person performed well, or have
16 they gone delinquent on other obligations. Not
17 surprisingly, that's the most predictive piece of
18 information we've got.
19 Secondly, now those for people where the answer
20 is yes, all of their prior credit they've performed in a
21 satisfactory way in the sense that they've met all of the
22 contractual obligations and they've never been late,
23 etc., that accounts for about 85 percent of the

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 43
43 Page 44 45
44
1 population. And so there are only, again, rough numbers,
2 15 percent of the population that's got serious
3 delinquencies noted on their credit report at any given
4 time.
5 The other 85 percent of the population does not
6 all present the same degree of risk in terms of future
7 credit behavior. And so we've got to look at other types
8 of information as well to see how those folks are likely
9 to perform in the future.
10 And the next most predictive piece of
11 information is what's labelled, current level of
12 indebtedness. In one sense, you can sort of think of
13 that as how close to the edge are these folks. Is it
14 somebody --you know, somebody who has got five credit
15 cards, all of them they have borrowed right up to the max
16 and they're making minimum payments, are a lot riskier
17 than somebody that their balance --the limit ratio is
18 down in the 10 to 20 percent range.
19 The amount of time that credit has been in use.
20 And actually that line is a little bit mislabelled. It
21 probably ought to say sort of the amount of credit
22 history available, which is really a combination of the
23 time that credit has been in use and also the number of

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 44
44 Page 45 46
45
1 trade lines that are on the file.
2 Now, trade lines can be on the file that aren't
3 active any more. In fact, again, those of us who have
4 been around for a while, sort of after we got into the
5 credit world, we probably have a lot of accounts that
6 we've paid off and closed. That good information, in
7 most cases, will stay on the file kind of indefinitely.
8 And so somebody that's got 20 or 30 years of
9 credit experience and any kind of reasonable number of
10 trade lines, maybe a lot of which have been paid off --
11 you know, student loans that have been paid off.
12 Mortgages that have been off. You know, your first
13 credit card may have been a gas card or a store card,
14 because that was the only place you could get a credit
15 card when you, you know, first got out of school.
16 So those have been paid off. All of those
17 historical pieces are likely to still be on the file, and
18 so we look at that. It's really the amount of credit
19 history available.
20 The next category we look at is the pursuit of
21 new credit. This is where the inquires come in and also
22 new account openings. Somebody that has a lot of
23 inquiries generated by the consumer --and I'll get into

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 45
45 Page 46 47
46
1 that in a little bit --or has opened a lot of new
2 accounts recently, is more risky than somebody that
3 hasn't been out there actively pursuing a lot of new
4 credit.
5 If I could actually then skip down to I think
6 number --yeah, number 33. There are a few --I'm
7 skipping a few slides in the presentation, because we're
8 kind of running out of time here. Keep going. Keep
9 going. Bingo. Okay.
10 I mentioned inquiries. This has been a topic of
11 some controversy. Fair, Isaac about a year ago
12 redesigned our bureau systems to do what we call inquiry
13 de-duping. We know that in a lot of cases somebody
14 shopping for a mortgage or for an auto loan will wind up
15 with multiple inquiries from the same transaction. And
16 of course not many people buy more than one house in any
17 given month, or more than one car.
18 And so we've sort of been able to say, if we
19 count all of the auto loan inquiries and all of the
20 mortgage loan inquiries within any given 30 day period as
21 a single inquiry, because it's probably only one auto or
22 one house that they're shopping for, that that is really
23 a better representation of their credit behavior.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 46
46 Page 47 48
47
1 And then also, again in mortgage lending
2 especially, people may wind up shopping around for some
3 period of time looking for the best rates, going to
4 different lenders or go through a mortgage broker who
5 shops at the different lenders. And so in fact what
6 we've said is, we don't count any inquiries in the 30
7 days before this particular score is computed.
8 So you can have a hundred inquiries in the month
9 before the score is computed and they won't count at all.
10 And you can have any number of mortgage or any number of
11 auto loans in any --I'm sorry. It's a 14 day period at
12 any time before the score is computer, and they'll only
13 count as one inquiry each.
14 Okay. If we could slip I think two slides.
15 It's just an example of the de-duping. One more. One of
16 the things we've done with the bureau scores is to
17 actually use multiple scorecards rather than a one size
18 fits all scorecard. And that will become relevant later
19 in the afternoon when we talk about what ought to be
20 disclosed.
21 And I think the question that Jodie posed is,
22 how can I raise my score. Because of the multiple
23 scorecard design, it's not possible to tell somebody, if

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 47
47 Page 48 49
48
1 you do this, your score will go up by X, because in fact
2 the credit bureau information is a moving target.
3 There's always information coming into the credit bureau.
4 Essentially on a daily basis credit bureaus are getting
5 information from credit grantors.
6 Also, our systems consider the age of certain
7 types of information. A delinquency last week is a lot
8 worse than a delinquency five years ago. A new account
9 that was opened up last month is more significant than a
10 new account that was opened up three years ago in terms
11 of pursuit of new credit.
12 And so even if you don't do anything with your
13 credit. You don't use your credit card in a given period
14 of time and you don't make a payment. And of course if
15 you don't make a payment for too long, something will
16 happen, because the credit grantor will now start
17 reporting you as delinquent. But, you know, you don't
18 miss a payment, but you don't make a payment. Even if
19 you don't do anything with your credit history, just the
20 passage of time is going to affect some of those things.
21 And so I could sit there with somebody's credit
22 report and say well, if you close this account, or if you
23 paid off that account, here's what would happen to your

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 48
48 Page 49 50
49
1 score if nothing else changed. But something else will
2 always change. And then one of the things that may
3 change, is that by doing something you may wind up on a
4 different scorecard, because the way that the system is
5 split, is first of all it looks for the presence of
6 serious delinquency information, either in the trade
7 lines or in the public record part of the credit file.
8 And so we have sort of the goods and the bads --
9 the previous goods and bads now separated using different
10 scorecards, because if we used a one size fits all
11 scoring system, anybody with any delinquency would get a
12 terrible score. But, in fact, not everybody with a prior
13 delinquency presents the same degree of risk, so we can
14 do a better job.
15 Now, the people with prior delinquencies as a
16 group are riskier than the people with no delinquencies
17 as a group. But we can do a better job by this split
18 scorecard design of finding the folks with prior
19 delinquencies that today pose a reasonable risk versus
20 the ones with prior delinquencies who are still just
21 terrible risks going forward.
22 And then depending --and then after we sort of
23 make the split based on the existence of delinquency,

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 49
49 Page 50 51
50
1 then we also look at some other information: the
2 thickness of the file, in a sort of philosophical sense
3 how many trade lines, etc., there are, the age of the
4 file, how long has this person had established credit.
5 Again, somebody who is 50 years old and has 30
6 trade lines, most of which are now closed, that may be a
7 very typical pattern. Somebody who is 22 years old and
8 has 50 trade lines, and they've only been in the file for
9 24 months and they've got 50 trade lines already, that's
10 probably kind of a strange bird.
11 And so we're using different scorecards to
12 evaluate those different groups of people and the recency
13 of new trade lines opened. Again, somebody who is kind
14 of out there in a steady state mode who hasn't taken out
15 a lot of credit recently versus somebody who is kind of
16 out there shopping actively, we're using different
17 scorecards to evaluate them.
18 Now, the same score still means the same thing.
19 If you get a 700 on any of those score cards, it still
20 means the same degree of risk. But we get there by
21 looking at different information and maybe weighing
22 information differently, depending on which of those sub-
23 populations somebody is in.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 50
50 Page 51 52
51
1 Okay. If you could again skip two. Just very
2 briefly, sort of the legal public policy issues, I guess
3 I'll save most of that for the panel this afternoon about
4 is it fair. I really just want to describe some of the
5 other slides that are in your packet.
6 We did two studies --and if you would go to the
7 next slide, please. We did two studies which are
8 labelled the LMI Study --low to moderate income. We
9 took data where we had --where we were developing custom
10 scorecards for specific credit grantors, and in many
11 cases where there were credit grantors who were going
12 from a judgmental decision process to a scoring
13 environment.
14 We looked at that data. We said sort of how do
15 the low to moderate income folks do when they're scored
16 versus how they were doing when the credit grantors were
17 making judgmental decisions. What is different about the
18 low to moderate income population versus the general
19 population, and sort of is scoring really an effective
20 way of making decisions on those populations.
21 A lot of people assume that for a lot of low to
22 moderate income borrowers, scoring just can't do a very
23 good job of evaluating credit risks, because they look

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 51
51 Page 52 53
52
1 different than the rest of the population. Well, in
2 fact, what we found out was that not surprisingly the
3 lower income borrowers as a group scored lower. That's
4 why that yellow line is lower than the other lines in
5 this graph.
6 If you go to the next slide, what we also found
7 out is that at any given score, the risk of those low to
8 moderate income borrowers was in fact at least as great,
9 and usually in most cases just a little bit greater, than
10 the general population. So that if a score in this case
11 of 200 equaled the 20 to one for the general population,
12 for the low to moderate income population it probably
13 equalled 19 to one or 18 to one. In other words, they
14 were just a little more risky than the general population
15 in any score. So, in fact, the scores were doing a very
16 good job of predicting the risk for that population.
17 The other thing that we looked at --again, if
18 you would skip two. This is why they were different. If
19 you skip to the next one, there are two pages that look
20 very similar in your set, and unfortunately they are
21 mislabelled in the printed set. They're right on the
22 screen.
23 Slide number 42 should say "Maintain acceptance

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 52
52 Page 53 54
53
1 rate; compare bad rates." And what we did here, is we
2 took the acceptance rate for the LMI population under the
3 prior judgmental decision process. We looked at the bad
4 rate that was coming up. And in fact, you see in some of
5 these cases --in fact, in a lot of these cases that bad
6 rate is terrible. It's horrible. The decisions that are
7 being made judgmentally on that LMI population are bad,
8 terrible decisions.
9 You know, the first couple of lines you've got
10 an acceptance rate of 26 percent and a bad rate of almost
11 11 percent. The next line you've got an acceptance rate
12 of 24 percent and a bad rate of 18 percent. Nobody is
13 going to live with an 11 or 18 percent bad rate in a bank
14 card portfolio. If you kept making judgmental decisions
15 there, what's going to happen is somebody is going to
16 come along and say, wait a minute. Our bad rate for low
17 income borrowers is totally unacceptable. We just can't
18 lend money to low income borrowers at all.
19 So what we did is say, well, okay. If you keep
20 the same acceptance rate and now you make those decisions
21 using scoring, what happens to the bad rate --and in
22 fact you'll see in every case the bad rate goes down, and
23 in many case very dramatically.

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 53
53 Page 54 55
54
1 The next slide, again, is mislabelled in the
2 printed set. It's okay up here. It should say "Maintain
3 bad rate; compare acceptance rates." Now, again,
4 maintain the bad rate for most of these is probably not a
5 realistic scenario. But using scoring, what you see is
6 if you were willing to live with that bad rate --and
7 actually the third and the fourth ones, the bad rates may
8 be acceptable, 2.3 and 4.3 percent.
9 But the acceptance rates judgmentally were very
10 low for those folks. If you evaluated that same
11 population using credit scoring and you're willing to
12 live with that two and four percent bad rate, you could
13 accept in the one case --in bank card number four you
14 could accept four times as many applicants as you were
15 accepting judgmentally. And that's again very consistent
16 across that.
17 The next study that we did was using credit
18 bureau information. And there, rather than looking at
19 the low to moderate income population, we looked at
20 consumers in zip codes with a high concentration of
21 minorities, blacks and hispanics. And I could get into a
22 lot detail about why we chose the various categories we
23 did. But basically we looked at zip codes with those

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 54
54 Page 55 56
55
1 high minority concentrations. One of the things we found
2 was that those zip codes represented a little bit less of
3 the credit bureau database than they did of the entire
4 population, although certainly they weren't totally
5 unrepresented, as I think a lot of people assume.
6 And indeed there are folks in that population
7 that simply don't use mainstream credit. They are never
8 going to have a credit report. But they're never going
9 to go to a lender that would pull a credit report in the
10 first place, and secondly the degree of under
11 representation isn't so great that it would really affect
12 the ultimate scorecard design.
13 And we see there --if you would go to the next
14 slide --that in fact that zip code definition winds up
15 looking a lot like the low to moderate income definition,
16 because those zip codes have approximately --the average
17 household income in those zip codes --and this was a few
18 years ago now --is about two thirds of the average --
19 or, I'm sorry --of the household income of the remainder
20 of the population. So not surprisingly in economic
21 terms, those high minority zip code populations look a
22 lot like the low to moderate income populations, and so
23 not surprisingly we came up with very much the same

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 55
55 Page 56 57
56
1 results.
2 If you go to the next slide, again you'll see
3 the score distribution is a little bit lower, which again
4 means that at any given cutoff score, you're going to
5 accept fewer of those folks. But if you go to the next
6 slide, you'll see again exactly the same pattern. In
7 this case the red line is the minority population. The
8 dotted yellow line is the rest of the population. At any
9 given score those folks represent a slightly higher
10 degree of risk.
11 So again, the scorecard is doing an effective
12 job of rank ordering risk, and to the extent there is any
13 discrepancy in the risk estimation, it actually favors
14 those high minority areas of borrowers just as it did the
15 low income borrowers. Again, for the credit grantors
16 listening, that difference is very slight. It's not
17 worth getting excited about. But to the extent there was
18 any difference, it was consistently in favor of the low
19 to moderate income and the minority zip code borrowers.
20 Again, we looked at how those two populations
21 differed, and finally we asked ourselves, well, could we
22 build --next slide. Could we build a scorecard
23 specifically for that minority population that worked

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 56
56 Page 57 58
57
1 more effectively than the scoring systems that are being
2 used now.
3 Well --and next slide, please. What we found
4 out was that if we used a one size fits all credit bureau
5 score, in fact then we could build a more effective
6 minority zip code scoring system. Now, I suspect that
7 the folks from the Fed and some of the other regulatory
8 agencies might want to have a conversation with us if we
9 did that. But this was a research exercise. We found
10 that it would be possible.
11 But what we also found was that the existing
12 Fair, Isaac credit bureau score using the multiple
13 scorecard approach, where you're looking at different
14 sub-populations separately, where you're looking
15 separately at the folks with prior delinquency, and
16 you're looking separately at the folks that don't have a
17 lot of established credit, that that multiple scoring
18 system that we were using was in fact more effective in
19 predicting risk in that minority zip code population than
20 a scorecard --a single scorecard developed specifically
21 for that population.
22 I've run a little bit over my time. I thank
23 everybody for your indulgence, and I'll see you this

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 57
57 Page 58 59
58
1 afternoon. Thank you.
2 MR. MEDINE: Thank you, Pete, for an excellent
3 overview of a very complex issue.
4 (Applause.)
5 MR. MEDINE: Now we're going to take a minute
6 and get our next panel set up here, and then we'll resume
7 in about two minutes.
8 (Brief recess.)
9 MR. MEDINE: We just heard from Pete McCorkell
10 on the issue of credit scoring generally. This next
11 panel is going to focus on the use of credit scoring
12 specifically in the mortgage industry. Obviously home
13 purchases are a consumer's largest purchase. And so how
14 credit scores interrelate with that decision is of
15 critical importance to consumers. And the presentations
16 will also include information about mortgage scoring and
17 the automated underwriting process.
18 We're going to start from my extreme right, so
19 I'll first introduce Pamela Johnson from Fannie Mae. Pam
20 is Vice President of Single Family Mortgage Business for
21 Fannie Mae. She is responsible for new product
22 enhancements to Fannie Mae's automated underwriting tool,
23 the Desktop Underwriter, and is responsible for credit

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 58
58 Page 59 60
59
1 research and data acquisition for Fannie Mae.
2 We're thrilled to have her here this morning.
3 Thank you, Pam.
4 MS. JOHNSON: Thank you. Can everybody hear me
5 okay? All right. Are we ready? Let me start first of
6 all by stepping back for a minute, because I think I want
7 to clarify and make sure everybody understands where we
8 are here.
9 Peter did a fabulous job walking through how
10 does one develop a scorecard. But I want to be clear
11 that there are many different types of scorecards out
12 there. There are many different --they're used in a
13 variety of industries, not just in the mortgage industry.
14 In fact, the scorecards have been around a lot longer,
15 and used for a longer period of time, than they have in
16 the mortgage industry.
17 The consumer industry --for example, credit
18 cards, etc. --have been using scorecards, I don't know,
19 how long, Peter, 20, 30 or 40 years. But in the mortgage
20 industry it's only been recent that we've adopted the use
21 of credit scoring as a tool in underwriting.
22 And I want to be clear about which scorecard we
23 recommend, which is the use of the credit bureau

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 59
59 Page 60 61
60
1 scorecards. So Peter spoke a little bit about this in
2 his presentation. And I wanted to point out that the
3 characteristics in the credit bureau scorecard are those
4 listed on page 24 of Peter's presentation, and not the
5 ones that he talks about early on, which could relate to
6 other scoring models.
7 So in case you missed it, he listed the five
8 characteristics that are most prominent in a credit
9 bureau scorecard. And those were previous credit
10 performance, current level of indebtedness, amount of
11 time credit has been in use, pursuit of new credit and
12 types of credit available. So just as a way of
13 clarification.
14 First of all, let me say I'm delighted to be
15 here. I thank David and team for inviting me. I think
16 this is a critically important issue, and I think this is
17 a terrific forum.
18 Credit scores and automated underwriting are a
19 very important tool in the mortgage industry. At Fannie
20 Mae we spent a considerable amount of time researching
21 and analyzing credit scores. And long before we
22 determined that they would be useful in mortgage
23 underwriting, we went through a lot of work to answer

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 60
60 Page 61 62
61
1 some very important questions. Those questions are: do
2 they work, are they predictive of risk as it relates to
3 mortgage lending, are they fair, do they work for all
4 borrowers in the same way, and when used correctly, will
5 they expand home ownership opportunities.
6 We answered yes to all of those questions after
7 a lot of research, and I want to take you through some of
8 our research and some of the analysis we did to get
9 there.
10 The first slide, please. So today I'm going to
11 talk to you about credit scoring, mortgage scoring --
12 because they are different --and automated underwriting.
13 I'll also discuss how scores should be used to originate
14 and underwrite mortgage loans.
15 Next slide, please. As Peter indicated, credit
16 scores are empirically derived in a statistical method of
17 assessing risks. A credit score --and specifically,
18 again, we're talking about a FICO credit bureau score --
19 is based solely on the information in the borrower's
20 credit report. The evaluation of the information in the
21 borrower's credit report, meaning how the borrower
22 manages credit, has always been an important factor in
23 mortgage underwriting. Credit scores when used correctly

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 61
61 Page 62 63
62
1 are based on information in the credit report that has
2 been proven to be predictive of credit risk and loan
3 performance.
4 Next slide, please. Mortgage scores, on the
5 other hand, are slightly different. A mortgage score is
6 also based on information that has proven to be
7 predictive of credit risk. A mortgage score is based on
8 information about the borrower obtained from the loan
9 application and the credit report, as well as information
10 about property value.
11 Mortgage scoring supports a comprehensive
12 analysis of both the borrower's ability to repay a loan
13 and the borrower's management of credit. A mortgage
14 score helps an underwriter perform a more consistent,
15 objective and accurate evaluation of this information
16 than traditional underwriting does.
17 Next slide, please. Fannie Mae recommends that
18 lenders use credit scores, specifically the FICO credit
19 bureau scores, in their manual underwriting process. In
20 addition, we have developed a more comprehensive
21 customized mortgage scoring model for use in Desktop
22 Underwriter, our automated underwriting system.
23 Next slide, please. Okay. So why should we use

For The Record, Inc. Waldorf, Maryland
(301) 870-8025 62
62 Page 63 64
63
1 credit scoring in mortgage lending.
2 Next slide, please. Credit scores are in fact
3 predictive of credit risks for all loans and all
4 borrowers. A borrower with a credit score below 620 is
5 in the neighborhood of two and a half and sometimes three
6 times more likely to default on a mortgage loan than
7 someone with a credit score between 660 and 699.
8 Next slide, please. Okay. This slide
9 illustrates this point. And if you'll look at the green
10 box in the middle, which indicates a score --a loan with
11 a score range of 660 to 699, and you compare that to a
12 loan with a score range of less than 620, you can see the
13 dramatic difference in the likelihood of default
14 performance.
15 Next slide, please. Similarly, this sl