The mysteries of utilization and rounding

[Gunnar419]

Common FICO wisdom seems to be that, for scoring purposes, utilization always gets rounded up. As MarineVietVet wrote in a 2010 thread, "Utilization is always rounded up for scoring so for example anything over 9% (9.1, 9.5, 9.8) is scored as 10%." Only one poster disagreed with that.

Here's that thread: http://ficoforums.myfico.com/t5/Understanding-FICO-Scoring/what-does-less-than-10-really-mean/m-p/79...

My question: I just gained 3 points on my EQ ScoreWatch FICO when a card reported a relatively small balance decrease. I didn't get any bump when my only other balance reporting card had a similar decrease last week. I got to wondering what might have triggered the bump.

My util was a 9.9 overall before the two balance decreases.

It was at 9.6 overall after last week's decrease (no score bump)

It's at 9.3 after today's report (3 point score bump)

Nothing else has changed, no AAoA milestones, just those two decreases.

I know its semi-useless to second guess FICO, but does anybody know what the rounding protocol is for the EQ algorithm? I assume 9.9 and 9.6 were both scored as 10%, but could 9.3 be getting scored as 9.5 or 9.0?

[user5387]

I don't think anyone knows for sure, beyond the various statements you can find on this forum.

It seems that utilization of less than 0.5% does not round down, but beyond that, it's not clear.

Rounding to the nearest integer is more accurate than always rounding up.

As far as a 3-point score change goes, there are various things going on day to day, including utilization changes, account age changes, and rebucketing.

[Gunnar419]

Thanks, user5387. I'm going to carefully watch what happens when my util drops below 9%. If EQ gave me those points because it was rounding to 9.5%, I might get another few points for going 1/2 percentage lower. If not, I'll try another theory.

[user5387]

If we assume that the various FICO scoring models are developed by means of heavy statistical crunching on historical data (consumer reports), then it seems likely that the raw percentages and thresholds and weights that come out of this process will not be nicely rounded numbers.

For example, instead of a utilization cutoff like 9%, what if the actual value is 8.47% for people in one bucket, and 11.56% for people in another bucket?

The numbers can certainly be tweaked by humans to make them look polished, but without this intervention, it's not necessarily all neat and tidy.

It's already known that the hit for a fresh derogatory is greater for those with high scores, and this is an illustration of the interdependencies in scoring.

[Gunnar419]

Great points and good thinking, user5387.

[HiLine]

---

@user5387 wrote:

If we assume that the various FICO scoring models are developed by means of heavy statistical crunching on historical data (consumer reports), then it seems likely that the raw percentages and thresholds and weights that come out of this process will not be nicely rounded numbers.

For example, instead of a utilization cutoff like 9%, what if the actual value is 8.47% for people in one bucket, and 11.56% for people in another bucket?

The numbers can certainly be tweaked by humans to make them look polished, but without this intervention, it's not necessarily all neat and tidy.

It's already known that the hit for a fresh derogatory is greater for those with high scores, and this is an illustration of the interdependencies in scoring.

---

Great post! I think the thresholds that people talk about on these forums are all estimates and should be taken as a general guidance, not as a set-in-stone rule. And in reality, the exact thresholds don't really matter; a 1% change in utilization generally doesn't effect a score change of more than 1 point. Not to mention there are 2 types of utilization: individual utilization and aggregate utilization. A change in one most likely leads to a change in another, and both will have an effect on FICO scores.