Your zip code and the zip code of said transactions play a role in fraud.

Credit score likewise plays a role.

There are many many attributes taken into account when discussing fraud. Some models use methods which take in all attributes. There’s many kinds of fraud as well, some of which will be impacted by credit score (for instance if you have more credit, and use your credit card more often, you’re more likely to be a target for a “hacker”. This means fraud will be more likely to occur on your account).

For reference: I’ve built fraud detection models being used at banks.

They also don't necessarily correlate correctly with behavior. If you actively avoid debt, for instance, you are essentially penalized with regard to credit score.

That’s a feature, not a bug, the most profitable people are the ones that go into debt, never miss payments but also never pay it off.

I've read this as the explanation multiple times, but I think there's an even more obvious explanation: people that go into debt, never miss repayments, and never pay it off are more likely to continue paying their debt off on time than somebody who never takes on or retains debt.

As but one component to this a person who never takes on debt has, tautologically, managed to live a life without any need of debt. That implies he would have little difficulty continuing to live a life without access to credit. Because of this, he will generally have substantially less concern about his credit rating than a person who depends on credit for their basic livelihood.

I'm not sure that doesn't correlate.

If I avoid debt for a decade, suddenly seeking debt should be a bit of a red flag, and I should have a correspondingly low credit score.

But the problem is credit scores are used as proxies for other purposes.

Credit scores measures behaviour in the same way that financial standing measures character or skin colour tells you something useful about a person. It try's to proxy one with another and fails, because really they're two independent variables and life is complex. Someone's ability to pay down debt at any given time is not behaviour, at least it's a tenuous argument.

> Credit scores measures behaviour in the same way that... skin colour tells you something useful about a person

Yeah no.

Credit scores, being a composite set of observations of recent financial behavior, are literally the opposite of an immutable genetic characteristic fixed from birth.

Credit scores are far, far, from perfect, but if you have a better way of judging the future actions of a human than compiling a list of their recent actions do speak up.

I don't dissagree, but the nature of them being "far, far, from perfect" includes things that don't fall under the actions or behaviour of a person and end up being fixed for a period of time, making the comparison in some ways apt. Assuming you don't explore further, knowing someone's number and using it to characterize a person is at least close to as stupid as age discrimination or racial. I'd grant you that there is a higher liklihood than others that the number is informed by poor decision making, but it's equally likely that the number is low due to unforseen circumstances.

>It try's to proxy one with another and fails, because really they're two independent variables and life is complex.

The use of data from credit reporting agencies is not to predict the outcome of lending to a specific person, but to predict the outcome of many, many loans to many, many people. For which I would say it succeeds, as the entities utilizing it seem to be economically successful due to making a smaller proportion of bad bets than entities who would not, for example in the linked article.