4 ms·

The newest FICO models already exclude this debt, though this is a much more significant step since most lenders and loan types you care about use very old FICO

The newest FICO models already exclude this debt, though this is a much more significant step since most lenders and loan types you care about use very old FICO models that do include it and are unlikely to be updated any time soon.

nradov 3y ago
Do you know why lenders still use very old FICO models? Is this just inertia, or are the new models more expensive, or do the old models predict default risk better for those loan types?
- FireBeyond 3y ago
  More expensive, for one. That's why CreditKarma and other 'free score' services often use Vantage, rather than FICO.
- dmoy 3y ago
  I think up until last year it was federal policy for Fannie Mae & Freddie Mac to use fico 5, and the rest of the industry kinda follows. Now I think they can use a new model, but it's probably some inertia on changing.
- user3939382 3y ago
  Just a complete guess. Banks and related insurance are using historical data with prediction models to forecast risk. Once the FICO version changes all that data is apples to oranges and the forecasting is less accurate.
- InTheArena 3y ago
  The big reason is that the different scores come out of the models. That may seem counter-intuitive, but models tend to be very tied to specific populations in specific ranges, and stability is often as valuable as predictability. Different model behavior - up or down - means that financial institutions have to retest their entire strategy to make sure that it doesn't screw up the segmentation and models that they have layered on top of various credit scores. This is not a trivial or short process - It's pretty common to see 6 month and 12 month champion/challenger tests. Even then, assuming that there is not a significant lift in predictability (unlikely as the FICO score is good at what it does) or addressability (do people who previously had a thin file now have enough data to make a decision) then you also end up looking and saying that it's not worth the changes to move to a newer score. Also, scores are incredibly regulated. Proving that the model you layer on top of the FICO score doesn't have a disparate impact (something standing in as a proxy to a protected bit of information such as ethnicity) is expensive. VantageScore doesn't really change this; it's a copycat of the FICO score that came about because the credit unions don't like having the score aspect out of their control and would rather people just pay them instead of an independent third party instead. The only thing that really would change the score in any way is additional data - but social media mining for scores is not acceptable outside of China (where their score predicts compliance with government not credit risk).
- supertrope 3y ago
  Consumer credit is heavily regulated. Not only is discriminating based on protected class illegal, the bank has to be able to explain its credit risk model to regulators. They can't use stuff like machine learning because that would be an unexplainable black box. Credit risk modeling is literally the core of their business. Often banks don't use FICO directly. They run their own model. They want to pick up as many customers who are likely to successfully finish their loan payments without picking up too many customers who are relatively more likely to default if there's a recession. A few basis points here and there is a big deal. There's models for mortgages, car loans, credit cards, boat loans, etc. The licensing cost of a new model, or just the upgrade costs must be justified by some benefit. Banks are stereo-typically slow moving.

nradov 3y ago

Do you know why lenders still use very old FICO models? Is this just inertia, or are the new models more expensive, or do the old models predict default risk better for those loan types?

FireBeyond 3y ago
More expensive, for one. That's why CreditKarma and other 'free score' services often use Vantage, rather than FICO.
dmoy 3y ago
I think up until last year it was federal policy for Fannie Mae & Freddie Mac to use fico 5, and the rest of the industry kinda follows. Now I think they can use a new model, but it's probably some inertia on changing.
user3939382 3y ago
Just a complete guess. Banks and related insurance are using historical data with prediction models to forecast risk. Once the FICO version changes all that data is apples to oranges and the forecasting is less accurate.
InTheArena 3y ago
The big reason is that the different scores come out of the models. That may seem counter-intuitive, but models tend to be very tied to specific populations in specific ranges, and stability is often as valuable as predictability. Different model behavior - up or down - means that financial institutions have to retest their entire strategy to make sure that it doesn't screw up the segmentation and models that they have layered on top of various credit scores. This is not a trivial or short process - It's pretty common to see 6 month and 12 month champion/challenger tests. Even then, assuming that there is not a significant lift in predictability (unlikely as the FICO score is good at what it does) or addressability (do people who previously had a thin file now have enough data to make a decision) then you also end up looking and saying that it's not worth the changes to move to a newer score. Also, scores are incredibly regulated. Proving that the model you layer on top of the FICO score doesn't have a disparate impact (something standing in as a proxy to a protected bit of information such as ethnicity) is expensive. VantageScore doesn't really change this; it's a copycat of the FICO score that came about because the credit unions don't like having the score aspect out of their control and would rather people just pay them instead of an independent third party instead. The only thing that really would change the score in any way is additional data - but social media mining for scores is not acceptable outside of China (where their score predicts compliance with government not credit risk).
supertrope 3y ago
Consumer credit is heavily regulated. Not only is discriminating based on protected class illegal, the bank has to be able to explain its credit risk model to regulators. They can't use stuff like machine learning because that would be an unexplainable black box. Credit risk modeling is literally the core of their business. Often banks don't use FICO directly. They run their own model. They want to pick up as many customers who are likely to successfully finish their loan payments without picking up too many customers who are relatively more likely to default if there's a recession. A few basis points here and there is a big deal. There's models for mortgages, car loans, credit cards, boat loans, etc. The licensing cost of a new model, or just the upgrade costs must be justified by some benefit. Banks are stereo-typically slow moving.