@Anonymous wrote:
@Anonymous wrote:Hello Inverse! Interesting that you get some benefit from going from 7% to a few percent lower.
Anytime one is testing one very narrow specific scoring factor, even a very important one like total CC utilization, it's important to make sure you are, as best you can, isolating that particular factor. For example, as you go from 7.5% to 4.5% to 1.5% to 0.5%, it's crucial to make sure that the number of credit cards that were reporting zero each time are documented as well. Ideally the number of zero-balance cards needs to be held constant. (TT raises this point and I agree with him 100%.) Likewise, because the shift can sometimes take 30 days to materialize, it's important to see if there is anything else happening around that time that could induce a score shift (even subtle things like an inquiry or a CC turning one year old or one's AAoA crossing into a new integer value).
As far as testing low and ultra low numbers, I would make a distinction between ultra-ULTRA low numbers (like $1.50) and slightly higher numbers like $6 or more. I say that because I believe some people on the forum have (rarely) reported getting their total U rendered as 0% when the non-zero amount is preposterously low. (Less than the price of a Big Mac, say.) I know that BOA, for example, is doing some stuff where they sometimes treat ultra-ultra-small balances as zero.
So for me, I almost think of the following as different test scenarios:
approx 0.4% of one's total U (and over $20)
roughly $6
roughly 80 cents
In theory they are all supposed to be rounded up to 1%, but if any one of them causes a 0% Util I would be almost sure it would be the last one, never the first two.
Note to BCH (our original poster):
As you can see, a few of us are using your thread as a jumping off place to discuss some general theories about where that fabled sweet spot is for overall credit card utilization goes. You seemed very focused on the ultra practical question of what you should do. I think all three of us (TomThumb, Inverse, myself) would tell you that as long as it doesn't pose you any financial hardship, you should be fine with paying your total credit card debt down to maybe $20 or so. That will give you a total U of 1% after rounding. Likewise you would be probably just as fine with 2%. But it is very important to get all your CC's reporting at $0 except one, since you need every available extra point on your FICO score.
I'm a trained mathematician and while not completely rigorous, the contstraints were good on my testing. I have a very stable profile due to my baddies and my installment balances were not hovering around any hypothetical steps. You can find the thread here: http://ficoforums.myfico.com/t5/General-Credit-Topics/Simulator-pay-down-revolving-test/m-p/4146154#M213009
On that other forum, the magical number for many is $2 to avoid it being dropped to $0. And is well documented if you care to read it.
I am with TT, in particular when it comes to mortgage scores, review of my simulator testing shows that the mortgage scores do not respond to anything less than 7%. There is some indication on that other forum that EQ 04 and to a much lesser extent EX 98 may respond to reporting $2 on a single card. I would even go further to say, if the algorithm is using a series of partial differential equations to solve the credit score, the boundary condition of outstanding revolving debt = $0 is an issue with the model and therefore revolving debt > $0 just makes it work better.
From a computational intensity perspective, I would suggest there's virtually no chance they're using a series of partial differential equations for the algorithm 
.
I am pretty certain that the all $0's being a negative is simply an artifact from the time when only current balance was reported... and the fact the VAST majority of people are not FICO strategists pre-paying their CC's.
Ergo, it was a reasonable statistical analysis to equate $0 with not using credit. I don't think it's really anything more complex than that TBH, as they could've just inserted a default value in there (much like AAOA reportedly does) for the calculation if it were simply a boundary condition or divide by zero type issue.
@Revelate wrote:From a computational intensity perspective, I would suggest there's virtually no chance they're using a series of partial differential equations for the algorithm
I am pretty certain that the all $0's being a negative is simply an artifact from the time when only current balance was reported... and the fact the VAST majority of people are not FICO strategists pre-paying their CC's.
Ergo, it was a reasonable statistical analysis to equate $0 with not using credit. I don't think it's really anything more complex than that TBH, as they could've just inserted a default value in there (much like AAOA reportedly does) for the calculation if it were simply a boundary condition or divide by zero type issue.
Agree completely - these models do not behave like they are calculus based. On the other hand... I suspect the models do include some iterative subroutines to converge on a profile specific score.