win333 said:How about page 44 link

WOW I thought that would be illegal?

Yes, the minority "profiling" is also interesting... kind of like the mortgage industry redlining that occurs around some neighborhoods.

I am not sure how you guys read the minority study.

I don't think Fair Isaac was trying to do profiling. I think with that HMA study, Fair Isaac was trying to determine if their scoring model is a good predictor of delinquency in both population groups ("HMA" and "Total Minus HMA").

There is not enough data shown, but from the data shown, these are the conclusions that I can see:

(1) from page 45, "Median Household Income" is lower among the "HMA" population vs. the "non-HMA" populatation, where the word "population" is used in the statistical sense. This basically says that on average, "the households who live in areas that are prediminantly minority are poorer". Notice that the way that statement is worded, we can infer that if you are not a minority but live in a predominantly minority zip code, then you are likely to be poorer, as far as income is concerned. You might infer that "minorities are generally poorer", although that would require further study.

This, unfortunately, is one of the sad realities about this country.

(2) from page 46, "The FICO score statistical distribution" is different among the "HMA" and "non-HMA" group. It looks like the distribution of the scores is lower among the HMA, so average, median, etc, all those indicators are probably lower.

Again, this is a sad reality of the state of the country.

(3) from page 47. I am having trouble parsing this one, but it appears, that for the same credit score, the chance of default is slightly lower among the HMA population, than non-HMA. This is actually an interesting result. Basically, if I am reading this result correctly, then the scoring model that is entirely race-blind (and, yes, the score is race-blind) is unfairly penalizing the minorities. So in fact, adding the race, or a zip-code component that reflects the minority areas, into the score might help minorities get a slightly better score.

I could be wrong about the interpretation of this result, because the slide has hardly any words.

(4) from page 48. We simply see differences in the statistical distributions among the two populations (HMA and nonHMA). The following metrics: previous delinquency, utilization, oldest open account, and number of credit card account, are significantly different among the populations. The slide does not say which way, but due to a lower credit score in the HMAs, you can guess which way the difference is.

The remaining two tabs ("moderate difference") and ("little difference") are addressing other parameters.

I don't see an attempt at racial profiling. I see a scientific study which tried to understand if there are differences among the minority and non-minority populations, and it appears that these differences are sufficiently captured by the FICO score, which is entirely race-blind, as far as we know.

Now, if one or more bad bank uses the Census Bureau data and gives out unfair terms in High Minority Zip codes, then the appropriate government agencies should be on their case. But it appears that the credit score is a good enough indicator of creditworthiness, and there is no reason to use zip-code or racial profiling to identify risk level.

Auream, precisely the point I was trying to make in the post just above yours.

Either way, anything that looks at race color or zip codes should not be permitted.

If the numbers were worse, that would be like saying no one in that zip code could EVER be credit worthy. There is no way we can ever pigeonhole what makes people great.

Warren buffet looks like a normal person in every aspect, you NEVER could have guest that he would have been special.

This Country was founded by taking all the worlds losers and giving them a chance, I think it proved that there is no way to tell who will come out on top.

India has a really screwed up system from what I understand, the CAST(sp) system.

I'm not trying to be against the last 2 posts, just stating what I believe the system should not do EVER.

Let me know when you have the 2008 powerpoint is available. Thanks

win333 said:Either way, anything that looks at race color or zip codes should not be permitted.

If the numbers were worse, that would be like saying no one in that zip code could EVER be credit worthy. There is no way we can ever pigeonhole what makes people great.

Warren buffet looks like a normal person in every aspect, you NEVER could have guest that he would have been special.

This Country was founded by taking all the worlds losers and giving them a chance, I think it proved that there is no way to tell who will come out on top.

India has a really screwed up system from what I understand, the CAST(sp) system.


I'm not trying to be against the last 2 posts, just stating what I believe the system should not do EVER.

Just because the statue of liberty says so, does not make it so.

America was not founded on 'losers' But hard working people that in one part or another did not share in the belifs/rules of their home country. They came to america because it presented a chance at both freedom and a fairer shot at making something for themselves, and making their lives and their childrens better.

A loser would have never crossed the atlanic ocean on a 3 month voage, where the chance of death was high... Just to come to a country where in the early years labor was extremly hard.

Thats even true today, most people coming to america, legally or not... dont come in as 'losers'... They might come un-educated... But still see america as a place where they can make a buck, and a better life for themselves... by working hard. I wouldn't call anyone... )expect the live with the parents pot head kid...) that works hard at any legit job a loser.

This gets posted every so often, and IIRC it was for a new credit scoring model they were pimping, not the one they currently use.

blok said:I run credit everyday and there are certain zip codes i run in which i would be willing to bet a good amount of money that it will come back bad. theres mathematical proof that some zip codes contain more people with poor credit then others.

That's not the point. The point is that judging people on criteria which they can't control is unethical and misguided. Every person has a right to rise to their potential and judging them based on circumstantial evidence robs them of that opportunity.

If you ask me FICO and similar methods are a complete farce. An ability test which doesn't notify the subject he's being tested can never be valid. That's why FICO fails miserably at analyzing the risk of people who perform application sprees.

A much more ethical (and probably more accurate) way to score people would be to ask people to perform specific tasks: Save half your paycheck for six months, write an essay detailing your plans for paying back our loan, etc, etc. The key is that the subject needs to know the consequences of his actions ahead of time. Looking at things such as past credit utilization is meaningless because without also knowing motivation it tells nothing about ability (ie high utilization does not indicate a lack of control or responsibility if that's the goal the subject was working towards...it indicates the opposite).

chimeer said:If the judging is accurate does it really matter if it is "fair"? If a lending model gives an accurate description of the risk level for a paticular borrower it shouldn't matter if the factor(s) taken into account were based on factors beyond the control of the borrower, if they are still accurate predictors of risk.

First of all, it is not accurate. Any model which neglects to consider feedback when facing intelligent adversaries is inherently flawed. This fact should be painfully obvious to anyone that spends any amount of time around this forum. If it were accurate then it would correctly interpret AORs, business credit usage, rate chasing, etc, etc. On this subset of borrowers it fails miserably. The ability of the model to remain accurate overall is nothing but a gamble that the fickle majority will remain ignorant and apathetic to such manipulation techniques. You see, because the underlying assumptions employed are inherently unstable, the supposed "accuracy" is not a sign of an excellent predictor as much as it is a sign of a fortuitous guess and continued success depends entirely on the unlikely scenario that that guess remains fortuitous. As the saying goes, "past performance is no guarantee of future results" and foolishly drawing your attention to superficial circumstantial evidence in lieu of the obviously inherent ability of your intelligent adversary guarantees that this saying will be inevitably fulfilled.

Second of all, legally speaking, sometimes it does matter if it is "fair". Some types of unfairness are illegal such as basing decisions on race, religion, sexual preference, etc, etc. The thing that most people fail to realize is that, mathematically speaking, it's a very very fine line between basing decisions on these things and other kinds of legal factors which people don't have control over. It's all too easy to take factors like zipcode, place of birth, income, etc, etc to build an extremely accurate predictor of race, religion, etc, etc. Statistically speaking, they are linked and if using the legal factors gives nearly the same results as using the illegal factors then whose to say if the lender isn't purposely choosing his particular legal factors as an intermediary to obscure the true illegal basis for his decision? Moreover, even if the lender isn't doing it purposely, if his choice of factors are closely correlated to illegal factors then whose to say his choice isn't subconsciously discriminatory (which is just as illegal)? Statistics is a slippery snake and one day our society is going to have to come to grips with the fact that you can't crunch numbers on people as if they were cards in a poker deck while simultaneously preserving their right to autonomy.