What are the different buckets?

[pdxuser]

I read that there were about 10 different buckets. Apparently there is one for people with less than 2 years of history and one for people with 2-5 years of history. Does anyone know the others?

[Anonymous]

This is mostly speculation, but most of the ones below are general consensus. I am of the opinion that longest account age is not the sole factor for age defined buckets, just the main factor. I will put a * in front of those that are more solidly evidenced on this board.

2 Negative Buckets:

* Public Record (AKA bankruptcy)

* Collections and/or Major lates (60 under 2 years, all 90's and 120's)

8 Positive buckets:

* 0-2 Years

* 2-5

5-12 (some break this out as 5-8 and 8-12ish)

*12-19

*19+ (some people believe there is another around 25+)

That gives us 5-7 positive buckets.

The other speculated buckets are

"Thin file" aka few accounts

and Recently obtained lots of new credit

[MattH]

Well, according to a talk at a 2003 Harvard University symposium by Hollis Fishelson-Holstine, Vice President of Research and Development at Fair Isaac, at

Article

The FICO scoring system at each credit bureau contains ten separate scorecards. The ealth of information available from credit files allows development of a much finer
segmentation than could be achieved when developing a scorecard from a single lender's data. A ey step in the development of multiple-scorecard systems is aligning the scorecards so that the resulting scores correspond to the same odds of repayment, regardless of which scorecard was used.

This article does not say anything about how those scorecards are defined, though it does have some interesting information about their process for selecting which variables to use. Since many variables are correlated with each other, selecting too many variables can actually reduce accuracy of modeling, a phenomenon with which I am very familiar in my own work as a pharmaceutical researcher -- while I have never worked in the financial world, most of the statistical modeling techniques for data mining are similar in all fields.

I have no idea whether they still have 10 scorecards, or whether they've added any since December 2003 when this presentation was given.

Message Edited by MattH on 10-25-2008 05:12 PM

[haulingthescoreup]

I think that the two negative buckets are:

- either a public record/ judgment OR a major derog (collection, chargeoff, big lates, etc)

- both a public record/ judgment AND a major derog

This is from looking at some of the codes used by the CRA's, especially EQ.

Still just a guess, though. We crouch around the fire and examine chicken gizzards, trying to figure out what's what.

FICO 08, which we may or may not ever see, added 2 negative buckets, and from its emphasis upon history within the last two years, my guess is that its the above buckets, separated by whether the baddie occurred within the last 2 years or longer ago.

DIG's guesstimate of the positive buckets is probably as close as we're ever going to know. Lots of speculation about the last one or two positive ones.

[RobertEG]

If you are interested in a detailed understanding of how professional credit scoring algorithms divide consumers into different "buckets," I can offer you one of the few public disclosures of how this process is viewed.  Most who do credit scoring religiously keep such algorithnms as trade secrets.

However, FirstUSA applied for and was granted a patent on such an algorithm bucketing process back in 2001.  Grant of a patent requires sufficent disclosure to allow one to practice the invention, so it contains a virtual wealth of description of how their bucketing tree is done, at least in their scoring process.  It covers such bucketing concepts as "clean vs dirty" files, "thin vs thick" files, "%util", etc.

It is U.S. Patent 6,202,053. 

You can obtain and view it for free at the US Patent Office website, uspto.gov.

Go to patents,then to patent search.   Enter the patent number in the "search by patent number" field, and you have it.  You will need to install a  free "image" viewer (instructions provided on the site)  to see the graphical patent drawings that accompany the text of the specification, but I highly recommend this, for it shows the full flow charts of the process.

I dont think you will find a more detailed discussion anywhere.

Message Edited by RobertEG on 10-26-2008 02:11 AM

Message Edited by RobertEG on 10-26-2008 02:16 AM

[MattH]

---

@RobertEG wrote:

If you are interested in a detailed understanding of how professional credit scoring algorithms divide consumers into different "buckets," I can offer you one of the few public disclosures of how this process is viewed. Most who do credit scoring religiously keep such algorithnms as trade secrets.

However, FirstUSA applied for and was granted a patent on such an algorithm bucketing process back in 2001. Grant of a patent requires sufficent disclosure to allow one to practice the invention, so it contains a virtual wealth of description of how their bucketing tree is done, at least in their scoring process. It covers such bucketing concepts as "clean vs dirty" files, "thin vs thick" files, "%util", etc.

It is U.S. Patent 6,202,053.

You can obtain and view it for free at the US Patent Office website, uspto.gov.

Go to patents,then to patent search. Enter the patent number in the "search by patent number" field, and you have it. You will need to install a free "image" viewer (instructions provided on the site) to see the graphical patent drawings that accompany the text of the specification, but I highly recommend this, for it shows the full flow charts of the process.

I dont think you will find a more detailed discussion anywhere.

Message Edited by RobertEG on 10-26-2008 02:11 AM

Message Edited by RobertEG on 10-26-2008 02:16 AM

---

Fascinating level of detail indeed!

For those not interesting in viewing the entire patent, I provide the key diagram here:

Many aspects of their model seemed familiar to me and I think would be familiar to most regulars here, but the patent did include one concept I had not seen elsewhere: revolving balance accumumulation:

Revolving balance acceleration is a dollar figure calculated by taking the total current revolving balance shown in a credit history and dividing that total by the number of months the credit history has been on file with the credit bureau. The low revolving balance acceleration for the second segment 232 is an acceleration of $35 or less...

I'd realized lenders get nervous about (1) new accounts and (2) large balances, but the idea of taking the ratio to get a speed seems moderately clever.  In practical terms this would seem to suggest: avoid running up a large balance on a new card (and with a new card consider making prepayments before the statement date).

Message Edited by MattH on 10-26-2008 07:56 AM

[haulingthescoreup]

This is interesting, but I'd like to point out for anyone who missed it that this has nothing to do with FICO score buckets. It's from a 7-year-old patent by First USA, which was gobbled up by Chase, and who knows if anyone uses it any more, or if they ever actually did.

I love this stuff as much as anyone, but PLEASE, don't start taking things from this and adding it to later FICO score discussions! It's confusing enough trying to make sense of FICO feedback, without dragging unrelated stuff in as well.

If nothing else, have some pity on the newbies who come across this while they're still in the drinking-from-the-FICO-fire-hose stage.

At my age, I don't want to have to de-frag my internal carbon-based tissue hard drive (brain) and get all this junk back out again.

[marty56]

The talk of score buckets might be interesting buy IMHO a mute point.  We know what we need to do to improve and keep our scores.  If our score go up or down as a result of re-bucketing, there isnt realy anything we can do other then keep on the correct course.

The general question of why our scores change when there is no obvious reason might have been something we pondered along with why and how we are here back in the early 70's listening to Yes, Emeron Lake & Palmer, Pink floyd and the Moody Blues  

[pdxuser]

Well, it may be good to know that you want to apply for a mortgage or car loan either right before the 2- or 5-year mark or a while after, but not right after you have your 2nd or 5th birthday.

By the way, how long does it seem to take to recover from rebucketing? 

[GFer]

"By the way, how long does it seem to take to recover from rebucketing?"

Very good question!!!

Seeings how Marty is right, this is the what interests me most.

Message Edited by GFer on 10-26-2008 11:44 PM