@cashnocredit wrote:I'm pretty sure FICO 9 and VantageScore 3 use payment data in addition to balances. FICO pointed out how predictive this payment data was at least a year ago. Doubht they would have subsequently introduced a new score that didn't include it. The data has been reported long enough to have been included in the logistic regression FICO and others do to establish score algorithms.
FICO 9 maybe though I'm not sure on that one as some major lenders like Amex don't report it... though possible it's just a further refinement, if it has that information score accordingly, if not skip.
VS 3 though was developed in 2012 and released early 2013: I don't even think TU/EQ had the historical payment information that point on consumer credit reports? It's possible as Experian I know did looking at my 2011/2012 reports, I should go check some of the older ones off my old system which I haven't tossed out yet.
@Anonymous wrote:Well SB, one set of concerns you have has to do with How do the current credit scoring models work? I.e. do any of them use the "high balance" field? Do any of them use the other "memory" data, the fields that describe your history of past CC balances and amounts paid? In case I haven't been clear, the answer is definitely NO. (I was 99% sure before this thread started, but FICO's own language made me wonder a little.)
So all of your credit related decisions should be made based on the idea that FICO utilization has no memory. So sure, if you put a bunch of money on a 0% interest card and want to pay it off slowly, fine. Do whatever is in your interest. Assume that there is no problem, now and next year, in carrying a balance, if that is what you want to do, Eventually, perhaps several years down the road, I think FICO might be starting to look at whether people have a habit of PIF vs. balance carrying, but if they start to do that, you can always start making sure you switch to PIF.
The real reason to get in the habit of PIF is not because there might one day be some FICO benefit in so doing. The real reason, as I see it, is because when anyone is carrying a CC balance it almost certainly means he is living paycheck to paycheck. Otherwise why not PIF? And when a person is dependent on the next paycheck to pay the rent or buy groceries or whatever, that's a very precarious position to be in, as so many folks discovered during the meltdown of fall 2008 or the Great Depression in 1929. The truth is, we should not only have at least 4 weeks of salary in a cash account, but a lot more.
PS. The other thing I can certainly tell you is that the current and past FICO models are indeed hyper obsessed with your CC utilization and also with the number of cards reporting a balance. But those things have nothing to do with carrying a balance. So yes, you will doubtless discover that your FICO score will get much much better when your total U gets down to 1-3% and when at least half of your cards have a $0 balance.
So good to now they haven't decided to start calculating our highest balance yet. I do believe eventually they will. (always tweaking) I agree with you on PIF versus carrying a balance. Just experimenting myself. But was taking advantage of the same as cash since it was free for 12 mos. My UT is over 10% right. No where near 20 or 30%.. I would lose it!lol I will bring it back down in a month or so? Just hoping to shake things up and come out with a higher score. And yes, most ppl should have a good 6 months put away. 
I'm also wondering about everyone hitting those credit limit increases Wondering how they watch/calculate those? Since it's a sp, from what point do they start another hp? Did they put that on there to see what most would do? Like go crazy with it? Just another thought.
@Revelate wrote:
@cashnocredit wrote:I'm pretty sure FICO 9 and VantageScore 3 use payment data in addition to balances. FICO pointed out how predictive this payment data was at least a year ago. Doubht they would have subsequently introduced a new score that didn't include it. The data has been reported long enough to have been included in the logistic regression FICO and others do to establish score algorithms.
FICO 9 maybe though I'm not sure on that one as some major lenders like Amex don't report it... though possible it's just a further refinement, if it has that information score accordingly, if not skip.
VS 3 though was developed in 2012 and released early 2013: I don't even think TU/EQ had the historical payment information that point on consumer credit reports? It's possible as Experian I know did looking at my 2011/2012 reports, I should go check some of the older ones off my old system which I haven't tossed out yet.
In my reports EX started reporting payments from wf spring of 2010. EQ was reporting and TU by the end of 2010. However, it took a while before most issuers were reporting in these fields. By summer of 2011 most of them except Citi were. Interestingly, Chase initally reported this info but stopped early last year. I wonder if CRAs were marketting this info to banks seeking transactors and Chase is probably one of the largest issuers to transactors.
So the data has been around a good while but, from some distribution charts I've seen in a recent FICO presentation most people that carry a blance pay within 1.2 times the minimum or less. There are only a handful that pay more than that over the minimums. The rest are transactors.While the people that pay, say, 1/3 the balance have a much lower risk the numbers are small and probably don't affect overall score accuracy much. However, the difference between transactors and revolvers is very significant. There may well be heuristics to try to estimate which people do that from overall balance changes alone. I've seen some indication of that.
How and if this stuff is incorporated into scores appears to be stuff, like balance trending and criteria for scorecards, they don't want to discuss specifically. FICO execs are on record that the payments made field is not used in standard FICO scores through at least FICO 8.
Even if highly predictive, there are regulatory concerns as well and these tend to take more time to test. For instance if the distribution of revolvers v transactors changed with consumer's age they would have to make sure that such shifts did not act as a proxy for age. At least adversely in the case of older folks.
So I can't say for sure whether the payment data is included in FICO 9 or VS3. It has been around long enough but just barely and may not have been around long enough to satisfy regulatory issues. However, FICO 9 and VS3 claim substantial improvements in Gini scores while eliminating paid collections as a score input. Tossing out a previous predictor has to be replaced with something more predictive. Unless including paid collections was simply an error.
@Anonymous wrote:Hi CnC. Sounds like there are four issues, which are probably worth distinguishing. They are logically distinct and also temporally distinct: i.e. they are stages in time.
- The CRAs begin having rich enough data for exploratory studies to be conducted. (Month-by-month balance and payment data for CCs.)
- Exploratory studies begin to be conducted to assess how predictive these data are. (E.g. How often does a consumer carry a CC balance and by how much? How does that alter his risk?)
- Owners of credit scoring algorithms (e.g. FICO, Vantage, etc.) decide how they want to change their models (based on #2). Developers begin programming the changes in the new upcoming model.
- The company releases a new model with changes in it.
We hit #1 a few years ago, though I think that Revelate may have observed a week ago that not all CRAs and not all CCCs were supplying the data.
We have also crossed over #2 (though not necessarily finished it) Here is an interesting piece by John U where TransUnion describes some of its own exploratory studies on the data.
https://blog.mint.com/credit/guess-what-else-is-on-your-credit-report-now-0514/
It's not at all clear, however, that we have even started on #3. That is, we don't know yet that FICO or Vantage have decided exactlly how they'd want their models changed, based on the exploratory studies they have begun doing in #2. And assuming that they have finalized how they want their next model changed, #3 still involves many months (or years) of programming and testing.
Then finally they would need to release that model (#4). That takes time for PR and other issues as well. (Then finally there is a 5th stage would be the actual adoption by lenders of the model.)
Since VantageScore 3.0 was released (phase 4) in 2013 and FICO 9.0 was released in 2014, I am doubtful purely in terms of timeline whether they could have finished with stages 3 and 4. I am not sure they are even finished with stage 2 yet, though there's no question that some exploratory work has been done.
As I mentioned yesterday, however, there's another reason to be doubtful, and that is that there have been scores of articles published on ways that FICO 9 differs from FICO 8, or ways that Vantage 3.0 differs from FICO 8, and I haven't seen any of these articles mention that the newer model is assessing whether a consumer is a revolver vs. a transactor. Not one. This is such an important development that it would be very high on somebody's list to discuss. So that's another reason I am confident that the earliest we'll see this is in FICO 10 or in VS 4.
I am curious to hear you explain your thoughts more about the regulatory concerns you think might be involved. You mention that it's important that any factor FICO uses needs to be shown that it can't act (for example) as a proxy for age. The thing is, FICO and VS already employ factors that can act as a proxy for age. The most notable one is "age of oldest account" which is highly correlated with a person's chronological age. I.e., if a person has that factor as < 2 years, there is a strong likelihood that he is in his 20s (though not a certainty). Likewise if that factor is (say) 30 years, then it is almost certain he is in his 50s and likely to be in his 60s.
John's article in Mint refers to a study from TU published in 2013. It stresses the high predictive value of the payments made data. Payments made have been reported since 2010 by all CRAs. As of 2011 most issuers were reporting this data. At about the same time as John's article FICO's banking blog has a similar entry. There has been silence AFAIK since then. However, the report from 2013 only evaluated a 6 month time frame and while this information was highly predictive for that frame. General credit scores would have to be tested over a 2 year frame. The dichotomy between transactors and revolvers itself is highly valuable. It can be used to more accurately make income estimates and credit capacity estimates and for this purpose the time frame is not relevant. I suspect FICO and the CRAs are selling these specilized metrics and probably have been for over a year.
As to PR, I don't think either banks, CRAs, or FICO would discuss this aspect or at least want to be first out of the gate to talk about it without a NDA. However, the overall quality of a score is a major selling point and VS3 has been pretty vocal about its improved score quality. They have also been the most explicit in revealing specific factos. For instance VS3 factors in whether a card is secured or not and whether the CLs are "high." In the case of CLs scores can be impacted negtively by having lower CLs as can be seen in a few of their "reason codes" while nothing in the reason codes relates to whether a card is secured or not in spite of their specific statement that secured status is factored into scoring. There is also nothing in their reason codes that relates to payments on revolving accounts.
As for age, there are considerable numbers of retired people aged 65 and up that have long, excellent credit but because CRAs only go back so long and most people by far with long histories are still working I don't believe scoring algoritms use credit age as a proxy and they certainly cannot use legally use it to produce more negative scores. It's going to become more of an issue in 10 years or so.
In any case regulators have to make sure that any algorithm change does not disparately affect a protected class. This is a complicated regulatory area and is an issue in most countries though each country has their own restrictions. For instance many countries allow age to be used without restriction. The US has one of the more restricted limitations in the use of age. It may be used but for people that are over 50 it can only be used if it produces more favorable credit scores.
@Anonymous wrote:
@cashnocredit wrote:
@Anonymous wrote:Hi CnC. Sounds like there are four issues, which are probably worth distinguishing. They are logically distinct and also temporally distinct: i.e. they are stages in time.
- The CRAs begin having rich enough data for exploratory studies to be conducted. (Month-by-month balance and payment data for CCs.)
- Exploratory studies begin to be conducted to assess how predictive these data are. (E.g. How often does a consumer carry a CC balance and by how much? How does that alter his risk?)
- Owners of credit scoring algorithms (e.g. FICO, Vantage, etc.) decide how they want to change their models (based on #2). Developers begin programming the changes in the new upcoming model.
- The company releases a new model with changes in it.
We hit #1 a few years ago, though I think that Revelate may have observed a week ago that not all CRAs and not all CCCs were supplying the data.
We have also crossed over #2 (though not necessarily finished it) Here is an interesting piece by John U where TransUnion describes some of its own exploratory studies on the data.
https://blog.mint.com/credit/guess-what-else-is-on-your-credit-report-now-0514/
It's not at all clear, however, that we have even started on #3. That is, we don't know yet that FICO or Vantage have decided exactlly how they'd want their models changed, based on the exploratory studies they have begun doing in #2. And assuming that they have finalized how they want their next model changed, #3 still involves many months (or years) of programming and testing.
Then finally they would need to release that model (#4). That takes time for PR and other issues as well. (Then finally there is a 5th stage would be the actual adoption by lenders of the model.)
Since VantageScore 3.0 was released (phase 4) in 2013 and FICO 9.0 was released in 2014, I am doubtful purely in terms of timeline whether they could have finished with stages 3 and 4. I am not sure they are even finished with stage 2 yet, though there's no question that some exploratory work has been done.
As I mentioned yesterday, however, there's another reason to be doubtful, and that is that there have been scores of articles published on ways that FICO 9 differs from FICO 8, or ways that Vantage 3.0 differs from FICO 8, and I haven't seen any of these articles mention that the newer model is assessing whether a consumer is a revolver vs. a transactor. Not one. This is such an important development that it would be very high on somebody's list to discuss. So that's another reason I am confident that the earliest we'll see this is in FICO 10 or in VS 4.
I am curious to hear you explain your thoughts more about the regulatory concerns you think might be involved. You mention that it's important that any factor FICO uses needs to be shown that it can't act (for example) as a proxy for age. The thing is, FICO and VS already employ factors that can act as a proxy for age. The most notable one is "age of oldest account" which is highly correlated with a person's chronological age. I.e., if a person has that factor as < 2 years, there is a strong likelihood that he is in his 20s (though not a certainty). Likewise if that factor is (say) 30 years, then it is almost certain he is in his 50s and likely to be in his 60s.
John's article in Mint refers to a study from TU published in 2013. It stresses the high predictive value of the payments made data. Payments made have been reported since 2010 by all CRAs. As of 2011 most issuers were reporting this data. At about the same time as John's article FICO's banking blog has a similar entry. There has been silence AFAIK since then. However, the report from 2013 only evaluated a 6 month time frame and while this information was highly predictive for that frame. General credit scores would have to be tested over a 2 year frame. The dichotomy between transactors and revolvers itself is highly valuable. It can be used to more accurately make income estimates and credit capacity estimates and for this purpose the time frame is not relevant. I suspect FICO and the CRAs are selling these specilized metrics and probably have been for over a year.
As to PR, I don't think either banks, CRAs, or FICO would discuss this aspect or at least want to be first out of the gate to talk about it without a NDA. However, the overall quality of a score is a major selling point and VS3 has been pretty vocal about its improved score quality. They have also been the most explicit in revealing specific factos. For instance VS3 factors in whether a card is secured or not and whether the CLs are "high." In the case of CLs scores can be impacted negtively by having lower CLs as can be seen in a few of their "reason codes" while nothing in the reason codes relates to whether a card is secured or not in spite of their specific statement that secured status is factored into scoring. There is also nothing in their reason codes that relates to payments on revolving accounts.
As for age, there are considerable numbers of retired people aged 65 and up that have long, excellent credit but because CRAs only go back so long and most people by far with long histories are still working I don't believe scoring algoritms use credit age as a proxy and they certainly cannot use legally use it to produce more negative scores. It's going to become more of an issue in 10 years or so.
In any case regulators have to make sure that any algorithm change does not disparately affect a protected class. This is a complicated regulatory area and is an issue in most countries though each country has their own restrictions. For instance many countries allow age to be used without restriction. The US has one of the more restricted limitations in the use of age. It may be used but for people that are over 50 it can only be used if it produces more favorable credit scores.
It sounds like, as it touches its inclusion in FICO 9 (a general credit score), you now feel that this is unlikely because they would have needed to have been testing it for two years, whereas you feel like they might have developed some highly specialized narrowly focused metric (not a general credit score) that they might be selling to specific customers. Is that right?
You mention that distinguishing between revolvers and transactors is highly valuable. Absolutely. No doubt in my mind. And indeed, I personally would love to see that distinction become a substantial part of the FICO and Vantage Score model as swiftly as possible. I am just doubtful that has in fact become part of either F9 or VS3.
I have altered somewhat my belief that VS3 and FICO 9 include payment data but not so much from the limited 5 years of reporting now but I realized there is another, dominating, factor. It could seriously impact issuers in adverse ways even if payment data is, as claimed, highly predictive.
The problem is that FICO, as historically used for granting credit doesn't distinguish between revolvers and transactors. Issuers make more money from revolvers and 60% of the population revolves. If suddenly revolvers had lower scores and transactors had higher scores issuers may start turning down profitable accounts they would otherwise approve and approve transactors they will lose money on. Score cutoffs should vary between transactors and revolvers with a lower cutoff for the more profitable revolvers. If most places only use a general risk score for approval an improved, in statistical terms, risk score could produce unintended consequences. So I'm starting to think that issuers will either not want a risk score that includes payment info or they will want such a risk score together with a flag that identifies whther the consumer is a transactor or revolver so they can apply a score cutoff for each based on their business needs.