Who creates FICO scores?

[3darcher2]

I know there are dozens of FICO scoring models but for the sake of argument, lets say we are talking mortgage.

I think it is typical that lenders us the 2004 version of scoring, right? 

So is the 2004 version of EX, TU and EQ all the same algorithm and calculated by Fair Isaac?  Does a lender call FI and ask them for a score and FI pulls your credit and calculates, or do the CRAs own these formulas?  I am guessing they own at least part of them or their rights - otherwise you could just authorize MyFICO to access your report and calculate your score, right?  So MyFICO either doesn't own the algorithms outright or there is some other agreement, right?  If that's the case, then why are the CRAs FAKOs so far off - they would know exactly the factors and their affect on scores.

If there is another thread describing this just point me to it - I can't find it.

Additionally, if all information is exactly the same, are the algorithms the same so scores would be the same?  Could you get different scores with the '04 versions with exactly the same information in all three?

[llecs]

I'm not sure exactly when the versions wre created, but it is safe to say that the most popular versions out there, TU04, Beacon 5.0 for EQ, and Fair Isaac Risk Model v2 were all created around 2004, give or take a couple of years. But I wouldn't necessarily call them a 2004 version though because they were made 2-3 years apart.

It is not the same formula across all 3. Even comparing FICO08, which was made within a year of each other, the formula differs across all three. It's a custom-made formula for each CRA.

FICO owns the formulas. In terms of mortgage lending, they'll sell rights to it to mortgage lenders and third-party mortgage report companies (e.g. Kroll, CBCInnovis, etc.). Each time a FICO score is pulled, a fee is sent to FICO. I don't know the specifics, but I assume the formula is kept at FICO, because the formula is dependent on everyone else's pulls (hence scoring buckets). In terms of us buying the score from here, myFICO pays a fee to the CRA(s) for access to your report info.

Why are FAKOs so far off? Because they can only guess what the FICO formula might be. They are off because they have different score ranges. They factor in stuff that FICO does not, and vice-versa. FAKOs score items that FICO never will score. And in terms of the CRAs providing FAKOs, they do so for the $$$. Experian back out of their agreement with myFICO, I bet, because Experian wanted to sell their PLUS score. Instead of only getting a little $$$ from myFICO each time we pulled our EX FICO, Experian can now charge more $$$ and not have to pay any fee to FICO and sell their crappy score instead (PLUS).

[3darcher2]

Thanks, I sort of understand, but why in the world would there be different formulas for different CRAs?

You would think that the same information would yield the same score.  That would seem fair.  If FI makes the formulas, what would possibly be the point of different ones for different CRAs? 

[crunching_numbers]

---

@3darcher2 wrote:

Thanks, I sort of understand, but why in the world would there be different formulas for different CRAs?

You would think that the same information would yield the same score.  That would seem fair.  If FI makes the formulas, what would possibly be the point of different ones for different CRAs? 

---

I have been wondering that same thing myself....

[llecs]

Speculating only, I think it's a licensing issue between the CRA and FICO. They required their own unique score.

[marty56]

---

@crunching_numbers wrote:

---

@3darcher2 wrote:

Thanks, I sort of understand, but why in the world would there be different formulas for different CRAs?

You would think that the same information would yield the same score.  That would seem fair.  If FI makes the formulas, what would possibly be the point of different ones for different CRAs? 

---

I have been wondering that same thing myself....

---

Could score bucketing be a fatcor?

[RobertEG]

FICO cant score without data, and the holder of the data cant produce a score without the algorithm.  Both are dependent upon the other.

Thus, they have cross-licensing agreements.  FICO licenses use of its algorithm, and when Brother Isaac wants to vend scores, they must obtain them from the CRAs, where the actual number crunching is done.

Just as Henry Ford relented in selling only black cars, customer demand ultimately carries the commercial day.  CRAs dont want black cars, so they get the color they want.

[cashnocredit]

---

@crunching_numbers wrote:

---

@3darcher2 wrote:

Thanks, I sort of understand, but why in the world would there be different formulas for different CRAs?

You would think that the same information would yield the same score.  That would seem fair.  If FI makes the formulas, what would possibly be the point of different ones for different CRAs? 

---

I have been wondering that same thing myself....

---

Here's a snippit from FICO's banking analytics blog:

http://bankinganalyticsblog.fico.com/2011/12/score-differences-across-credit-bureaus-reflect-true-da...

Because the CRAs define, structure and maintain data differently, FICO develops its scoring models for each CRA to get the maximum predictive value from that CRA’s unique data. This process yields a more valuable score for lenders and borrowers compared to the alternative approach of bypassing data differences and focusing only on those data characteristics that are common to all three CRAs. A score based only on the lowest common denominator between CRAs cannot extract the full predictive value from each CRA’s data.

Since each CRA has different structures and some differences in reporting FICO scores are optimized for each CRA to most accurately predict risk with the information and structure available at that CRA. This can, on occasion, produce different scores even if the generation of FICO is the same and often they aren't.

Note that the last sentence from the snippit addresses the claim to fame of VantageScore which makes an effort to produce similar scores across the CRAs. This may seem to produce a more valid score but in actuality it is somewhat less accurate because it must, of necessity, use less information than scores optimized for predictability with each CRA's individual database.