fico bankcard score

[Anonymous]

I'm trying to figure out if anyone knows how the bankcard score acts as opposed to fico 8 scores. I just had my fico 8 scores go up roughly 50 points across the board from opening a credit card( i had none!!), and im curious if my bankcard score possibly went up like that as well. I dont want to spend the $47 for a new report!

and also does the auto and mortgage scores typicaly rise the same as the regular scores?

[Anonymous]

Bankcard Enhanced tends to track Fico 8 Classic. Mortgage version seems to do it's own thing, so that's anybody's guess. Don't know about Auto Enhanced.

[Revelate]

Mortgage scores are just FICO 04 and FICO 98, all classic, for 99.999% of the mortgages underewritten in the United States, maybe higher honestly.  No big deal there and very well characterized.  The same optimization pattern for FICO 08 other than installment utilization holds for FICO 04 (EQ and TU for the mortgage trifecta) and it's effectively 1:1 for Experian which is on the FICO 98 model on a trimerge.

Auto enhanced wants installment history, preferably at pretty utilization, under any version but absolutely under the FICO 8 auto enhanced industry option which is the dominant auto pull by a wide margin as I understand it.  People that don't have installment history get whacked by the infamous "first time buyer" penalty there.

[Thomas_Thumb]

---

@Anonymous wrote:

Bankcard Enhanced tends to track Fico 8 Classic. Mortgage version seems to do it's own thing, so that's anybody's guess. Don't know about Auto Enhanced.

---

Best I can tell is both Enhanced Fico 8 scores track together. Also, Classic and Enhanced Fico 4 scores track together. See below trends by credit bureau for reference. In general all the credit scoring models show correlation (move in same direction except Classic Fico 8 has no movement). After all, a majority of scoring factors are the same among models - just weighed differently. The main point is one model's score can not be used to accurately determine another's model's score. 

[Anonymous]

Cool charts! 

[Anonymous]

Hello TT!  Thanks for the charts.  If I am understanding them right, they are the scores of one particular individual (you) over a time period of 12 months.  Is that right?

That's suggestive and helpful... so thanks again.  On the other hand, it's a piece of data on only one person and furthermore that person happens to have an unusual profile (one that can generate a perfect 850 FICO 8).

Do you have any charts you could share with us that show aggregate data of any kind?  Immensely helpful would be a chart that shows percentile data for a huge cohort (hopefully representaive of the population at large).  Instead of the horizontal axis being time, it would be percentiles going from 1 to 99.  It would enable us to compare, for example, what an 83rd percenile means for:

     * FICO 8 Classic

     * FICO 8 Bankcard

     * FICO 8 Auto

     etc.

Just draw a vertical line at percentle 83 and you'd be able to see what score all the different models return at that point.

We know for certain that when you get high enough (96, 97, 98, 99) some of the different models would diverge, since they have different top ranges.  But it would be interesting to see at what point that happens, and whether an 80th percentile still shows numeric scores very close to each other.

The same kind of chart could have the axes flipped, which might be even more illuminating.  The horizinontal axis would be the score itself.  So you could see whether or not a "780" score, for example, means ultimately very close to the same thing on all the different models (i.e. returns approximately the same percentile ranking).  Eventually as the scores get high enough they diverge of course: in some models a perfect score is 818 (Equifax FICO 4) but with other models you don't hit a perfect score till you get to 900.

Does anything like this exist, to your knowledge?  I am especially curious to see how the models compare in the score range of 700 to 800.  It seems like something like this MUST exist somewhere, even if hidden, or else it would be hard for creditors to make decisions about what model to use.  (E.g.  company X is drawing a line at 730 for its best rate using FICO 8 Classic, but they are thinking about switching to FICO 8 Auto -- what might be an equivalent score number for the same percentile?)

[Thomas_Thumb]

Yes, the data is 3B results on my profile over time. I probably have been assigned to the same scorecard for the entire period. Thus, very little external noise.

I suspect most profiles that have not hopped scorecards would see fairly strong correlations given enough data. "IV" discovered that comparing VS3 to Fico in a recent correlation analysis.

In general populations as a whole, while useful in showing correlation, have too great a spread to develop/use a generalized correlative formula on an individual basis. The problem with that approach is the scorecard effect which strongly influences how factors are weighed. Looked at as a whole with clean/dirty in the mix creates a huge spread and models will certainly interact differently depending on scorecard.

I did come across a lengthy study published in 2012 comparing Fico4/Fico98 to VS1/VS2 and Educational scores:

"Analysis of Differences between Consumer- and Creditor-Purchased Credit Scores"

See link below. You will have to scroll down the list and then click on the article to view/download the pdf. A summary table from the article is also pasted

EDIT: Changed to direct link from indirect

http://files.consumerfinance.gov/f/201209_Analysis_Differences_Consumer_Credit.pdf

 Table from: "Analysis of Differences between Consumer- and Creditor-Purchased Credit Scores"

[Anonymous]

---

@Thomas_Thumb wrote:

Yes, the data is 3B results on my profile over time. I suspect I have been assigned to the same scorecard for the entire period. Thus, very little noise. Even so, I suspect most profiles that have not hopped scorecards would see fairly strong correlations given enough data. "IV" discovered that comparing VS3 to Fico in a recent correlation analysis.

In general populations as a whole are not overly useful in showing strong correlations. The problem with that approach is the scorecard effect which strongly influences how factors are weighed. Looked at as a whole with clean/dirty in the mix creates a huge spread.

I did come across a lengthy study a couple years ago comparing Fico to VS2 and other Educational scores. See link below. You will have to then click on the article to get thr pdf download.

https://search.yahoo.com/search?p=analysis%20of%20differences%20between%20consumer%20and%20creditor%...

---

Well, it depends on how the word useful is being used.  If a person is curious to see where he falls on a percentile ranking, the aggregate data that I described is exactly what you need.  Tracking how one's own scores are fluctuating over time doesn't help him get a sense of that, because all he is seeing is his own single bit of data.

It's very common for a person to get a score of some kind and then want to know where that puts him compared to the other people in the class, so to speak.  Somebody gets a 700 on the verbal SAT.  He can even take the SAT several times and find out how much his score is likely to fluctuate, or compare how much higher or lower his Verbal score is compared to the Quantitative score.  And that is useful in one way.

But it is also useful -- in another way -- to want to know what your score means compared to other people.  Knowing that (as a percentile) gives him some kind of common yardstick in assessing how two different scores on two different tests might be viewed by an outsider (e.g. a college admissions board).

Bear in mind that a typical outside creditor when he pulls your credit score just sees your score.  He doesn't see your score and then analyze its meaning based on also knowing what scorecard you are in.  He never knows your scorecard.  Therefore from a typical credtor's perspective, he's seeing thousands of random people thrown into a big pot each with a score, which his company ranks from lowest to highest and then draws various lines in the sand.  So from a consumer's perspective it is actually helpful to know roughly what his score means as a percentile, because it is fundamentally the same kind of thinking that the creditor is engaged in when he evaluates him.

Thanks for the link.  I will try to look at it later today.

[Thomas_Thumb]

---

@Anonymous wrote:

---

---

It's very common for a person to get a score of some kind and then want to know where that puts him compared to the other people in the class, so to speak.  Somebody gets a 700 on the verbal SAT.  He can even take the SAT several times and find out how much his score is likely to fluctuate, or compare how much higher or lower his Verbal score is compared to the Quantitative score.  And that is useful in one way.

But it is also useful -- in another way -- to want to know what your score means compared to other people.  Knowing that (as a percentile) gives him some kind of common yardstick in assessing how two different scores on two different tests might be viewed by an outsider (e.g. a college admissions board).

Bear in mind that a typical outside creditor when he pulls your credit score just sees your score.  He doesn't see your score and then analyze its meaning based on also knowing what scorecard you are in.  He never knows your scorecard.  Therefore from a typical credtor's perspective, he's seeing thousands of random people thrown into a big pot each with a score, which his company ranks from lowest to highest and then draws various lines in the sand.  So from a consumer's perspective it is actually helpful to know roughly what his score means as a percentile, because it is fundamentally the same kind of thinking that the creditor is engaged in when he evaluates him.

Thanks for the link.  I will try to look at it later today.

---

Score distributions are quite useful in comparing how you fit in with a population as a whole. However, that's different than correlation. I previously posted a few score distributions for VS and Fico to help one see how they compare. Distributions are fairly readily available so no real need for correlation to get a ranking. 

Per your analogy, I have seen a few SAT to ACT comparison tables which aim at correlating the scores. It's interesting and helpful. However, at the end of the day, actual score distributions are available for each exam nationally and by state. Colleges also provide summary statistics for both tests. Thus, no need to attempt a correlation for ranking

Getting back to Fico credit scores - Experian has published percentile score tables for all the 3B Fico scores  (Fico 9 flavor excluded). They also have tables for Fico NextGen and Fico bankruptcy available. If you do enough searches, you can find links to all of them. Unfortunately, links are not consolidated.

Key point is: Use specific score distributions for ranking not correlations

Correlation comes into play when attempting to estimate how score on one model would translate to score on another. That's where using a generalized correlation model for a specific individual lacks validity. The spread is much to wide for reasonable precision.

[Anonymous]

Yeah, it sounds like we are just talking about two different things, both useful and related, but different.

If a specific individual wants to know how his different scores are likely to be related to each other (and this might be as specific as model-CRA) then the best way to do that is for him to track his own scores over a period of time, looking to see if his EQ FICO 8 Auto score tends to be 10 points higher than his EQ FICO 8 Classic (say).   And even this will only work when he is confident that his profile is fairly stable and has remained within the same scorecard (for a given model).  I am guessing this is what you mean by correlation: the ability of an individual to guess what model X will produce for him if he knows his score value under model Y.

The ironic thing about that is that a person can only develop that indivdualized predictive framework for himself by actually pulling all his scores and doing so repeatedly, e.g. as you did over a year period.  And if the person is doing that, then he has less of a need to infer the likely value of one score from another -- because he actually has the true value of every single score.  But I bet it would have a lot of practical value for a person who was willing to do that for a while, develop the framework, and then simply use his free FICO 8's from his credit cards.  I think that's probably the direction I am headed. 

I was interested in a different thing, which is what percentile ranking a certain score might have. E.g. if we focus just on the value of 780, do the various models give pretty close to the same percentile ranking for it?  What happens if I changed that to 790 or 770?  Etc.  This is how a lender would be viewing a score -- as a way of ranking you compared to other consumers.  Getting a sense of whether all my models are giving me somewhat close to the same percentile is useful in that sense.