topic Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator? in Understanding FICO® Scoring

Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 00:22:24 GMT

In all the searching and research I've done I can't believe this hasn't been tried yet. Maybe there's something I'm missing.

The FICO is not a magic number right? It must be based on a hard mathematical formula albeit one with a crapload of variables...

For you football fans out there, the QB rating at first glance seems like an arbitrary decimal number pulled out of thin air but it's actually:

a = (((Comp/Att) * 100) -30) / 20
b = ((TDs/Att) * 100) / 5
c = (9.5 - ((Int/Att) * 100)) / 4
d = ((Yards/Att) - 3) / 4

a, b, c and d can not be greater than 2.375 or less than zero.

QB Rating = (a + b + c + d) / .06

The Fico formula will be exponentially more complex but there's no reason we shouldn't be able to derive it with enough data.

Now I've noticed that many of you will actually post your scores from several reports in your signatures and you aren't shy about them... with as many subscribers as there are to these forums, that is a TON of available data out there.

I propose we compile a database with scores as well as credit history, debt amounts, # and dates of credit inqueries etc. All anonymously of course. It could include several reports from the same people taken at different times... this way we would be able to calculate our own FICO scores based on the information we get for free in our annual credit reports and know EXACTLY how our score will change based on different credit activity.

Who's with me?

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

haulingthescoreup — Fri, 22 Feb 2008 02:21:19 GMT

Hi, ChuckG, ummm... that's pretty much what these forums are, especially this board!

Check out Credit Scoring 101, stickied at the top of the boards. And RobertEG has a formula, although his numbers don't match mine that often. I think it's because we're in different score buckets.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 03:18:11 GMT

I read all 13 pages or so of the Credit Scoring 101 post and there was nary a mention of a FICO formula or collecting data to derive one.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 03:56:08 GMT

All great breakthroughs are accompanied by naysayers who claim that such a thing could never be accomplished.

Please allow me to provide that service for you.

In the words of our venerable member MV:

"Understanding FICO scoring is an oxymoron! "

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 04:23:25 GMT

ChuckG wrote:
The FICO is not a magic number right? It must be based on a hard mathematical formula albeit one with a crapload of variables...

The "crapload of variables" will be the hardest part. It's not a trivial problem.

As the number of variables to account for on the input increases arithmetically, the variation in potential outputs increases exponentially. To top it off, you would have to account for absolutely every variable on a report, since you wouldn't really know which of them are relevant. You would also have to account for how each variable is scored in the presence or absence of other variables.

The variables on a report number in the thousands.

Are you suggesting an attempt to reverse engineer the process inside the "black box" down to the letter with access to nothing more than said "crapload of variables" going in and a single data point coming out?

All I have to say is, "GOOD LUCK!!!"

If you didn't already know the formula to determine a QB rating, do you think you could have reverse engineered it without even knowing which specific variables are even relevant to input, let alone how they are to be manipulated in order to arrive at the final result?

Edited to add: Are you aware there are also 10 different scoring "buckets," as we like to call them around here? That means 10 slightly different algorithms. Each CRA also uses a slightly different version of the algorithms, too. So now we're up to 30 algorithms. Which bucket does a given report belong in? Whoops, that's something you'll have to reverse engineer before you can even begin reverse engineering the actual scoring algorithm for that particular bucket.

Then once you've done the impossible and figured out an algorithm for one particular bucket on one particular CRA, you'll have to repeat the impossible 29 more times.

BTW, I'm not coming down on you. Just having some fun more than anything.

Message Edited by cheddar on 02-21-2008 08:23 PM

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 05:43:51 GMT

I didn't say it would be easy and I certainly don't believe that I'd be able to do it myself, I only took one year of calculus. But I'd be willing to give it a shot, and I'd bet there are a lot of other smart people out there who would be too if the data were available.

As for the QB rating, if You gave me 100 QB ratings and the stats which we knew were used then yes I could derive that formula.

The key ingredient being that this assumption is true: Everything in our credit report affects our FICO score to some degree... and nothing that is not reported has any affect. (with the exception of known constants like the passage of time, etc.)... It would be an exercise in futility if that assumption were not correct.

I had no idea about the different "buckets", that indeed tells us that there are score motivating factors NOT listed in the report. So we are left to play a game with our lives in which we are not given the rules. Lovely.

Oh well, it was an idea... I get ambitious sometimes, especially when it comes to fighting the system

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 06:46:54 GMT

ChuckG wrote:

The key ingredient being that this assumption is true: Everything in our credit report affects our FICO score to some degree... and nothing that is not reported has any affect. (with the exception of known constants like the passage of time, etc.)... It would be an exercise in futility if that assumption were not correct.

That assumption is not correct. There are lots of things on a report that have zero effect on FICO. Employment data. Addresses. INQs more than one year old. And there are lots of things on a report that have a certain effect on FICO only when certain other conditions are met, yet have a different effect on FICO when still other conditions are met.

This is where the exponentially increasing complexity comes from.

ChuckG wrote:
I had no idea about the different "buckets", that indeed tells us that there are score motivating factors NOT listed in the report. So we are left to play a game with our lives in which we are not given the rules. Lovely.

Yep. Search the forums and you'll find LOTS of complaints about the secrecy of the algorithm and its effect on our financial lives.

Now, having said that, that's precisely what this forum is for, as HTSU said a few posts ago. While we don't know the specific algorithm, we do have at our disposal some very general rules and guidelines, which frankly, is really all you need in order to maximize your own FICO score. It's really not necessary to know the absolute specifics of the algorithm. There is quite a lot of information out there already if one doesn't mind doing a little digging.

You said you noticed that people here are not shy about posting their scores. In many cases, it's because they have brought their scores up pretty quickly, and they're proud of what they've accomplished. Those many success stories go to show that as long as you manage your credit based on the advice you can get on this forum, your score will do just fine in the long run. Knowing the exact algorithm won't really help that much.

In any case, it has been pretty well established that above 800 or so, the only thing that can help you is more age and additional clean history. Even if you knew the algorithm, you couldn't magically turn the clock forward to add the necessary ingredient of time.

ChuckG wrote:
Oh well, it was an idea... I get ambitious sometimes, especially when it comes to fighting the system

Yeah, it was a thought. Sorry to burst your bubble there.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

haulingthescoreup — Fri, 22 Feb 2008 11:21:22 GMT

It's still fun to try, and there are some things that we've established. But the more members we have who contribute to this, the more obvious it becomes that it's nearly impossible to separate out all the variables.

For instance, it's bad enough trying to isolate the cause of a single score movement on your own account (hmmm, picked up an inq, but one account went over 6 months old and a 60-day late fell off), but then two different members will post identical actions with completely different score consequences. (Score buckets! Score buckets!)

After reading some of the original posts from back when these forums got started a year ago, I can tell that we've figured out quite a lot. Still a long way away from getting it into mathematical form, though.

Oh, if you haven't stumbled across this post, this was fun:

http://ficoforums.myfico.com/fico/board/message?board.id=generalcredit&thread.id=64959

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

RobertEG — Fri, 22 Feb 2008 13:47:41 GMT

As I have previously posted, I have attempted to develop a FICO model that I have on an Excel spreadsheet that attempts to model each of the FICO categories, which I would be glad to share with anyone, provided there is some way I can get it to you. It admittedly makes many mathematical assumptions, and the idea of having some kind of database of the experience of others is an excellent idea, for it would allow for tweaking of the model by comparing it against a broader base of actual actual results . I would be very interest in such a database, if it could be created. Of primary interest would be data on the acual results of prior lates, both minot and major, as they age, for this is a category that really hits hard. At least I have a start, and for my bucket, it gives me a pretty accurate projection. My real problem with the current FICO simulators that you can get online is that they are, for the most part, one-dimensional, and thus do you permit you to see the result of an action that can affect multiple categories, such as openiing a new TL that simultaneously involves a hard inq, a change in avg. crediit length, a change in CL, a change in %util, and a change in credit mix.

There is a very good reason why the actual FICO models are secret.... they are proprietary commercial products that make the vendors money! But like most things, they can be reverse engineered once the data and results are known. The real value I have received from my attempted model is not whether it provides a perfectly accurate projection, but rather that it allows me to see the general interaction of the various categories in more than just a purely speculative way.

Message Edited by RobertEG on 02-22-2008 05:47 AM

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 16:26:20 GMT

Hey join in.

http://ficoforums.myfico.com/fico/board/message?board.id=ficoscoring&thread.id=6034

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

BallBounces — Fri, 22 Feb 2008 17:01:25 GMT

I think this is a great idea. Let me know when we are done.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 22 Feb 2008 17:37:07 GMT

@RobertEG wrote:
As I have previously posted, I have attempted to develop a FICO model that I have on an Excel spreadsheet that attempts to model each of the FICO categories, which I would be glad to share with anyone, provided there is some way I can get it to you.

Message Edited by RobertEG on 02-22-2008 05:47 AM

Hey Robert, I would love to take a look at what you have in your Excel sheet so far. Just PM me back and I'll give you my email address.

Also, If you want to give me a list of known relevant variables, I can put together a database with an HTML interface that we can ask users to start filling in.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

RobertEG — Sat, 23 Feb 2008 06:43:16 GMT

Please, my fellow FICO-holics, do not short-sell the benefits of developing our own FICO score simulation just because of the obvious fact that it cannot be replicated down to the decimal point of the actual FICO model and score.

If you have used the online score simulators provided by the major credit bureaus, including myFico, I think you will agree that they are woeful. "What-ifs" 24 months from now. No help at all. The real modeling benefit, if a comprehensive simulation is done, is not to replicate a FICO score, for all buckets, and all the other unknown variables that only the propriretary FICO model can know for sure, but rather to show the general impact of any single action upon each of the five major FICO categories NOW. That cannot be done without some kinda modeling. I know that most people hate math, but FICO is math! It is a number, generated from math equations, not anecdotes. How many know the total points for each category, and the general impact of points lost for each category for past sins? And the concurrent impact on other FICO categores? People post and dwell, for exsample, upon hard inquiries, but they are minor in the algorithm and drop at 12 months. Not an area of prime focus. People want to get a feel for numbers, and not anecdotes..

People ask again and again on here about how an action taken in only one FICO category will affect their score in the future. That is impossible to do without a comrehensive model of all categories of credit model. Any model, no matter its faults, thaqt considers the impact of an action on all cetegores of scoring, is kinda more meaningful than saying "my Aint Bessie paid off her Junipter Red Elite Platinum Visa down from 70% to 8% last month," and then asking what her score will be in 3 months. FICO does not rely upon Ouija boards, so neither should we. Math is FICO.

I have developed a statistical mathematical algorithm, covering all FICO categories, that works for me. I invite others to help in refining it. ChuckG has offered to set up a user database for user input that will give us a broader database of actual experience that will help in refinement of the model. In the meantime, if anyone would like a copy of my current Excel model, just private message me, and a copy will be sent to you.

Message Edited by RobertEG on 02-22-2008 10:24 PM

Message Edited by RobertEG on 02-22-2008 10:26 PM

Message Edited by RobertEG on 02-22-2008 10:34 PM

Message Edited by RobertEG on 02-22-2008 10:37 PM

Message Edited by RobertEG on 02-22-2008 10:43 PM

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

RobertEG — Sat, 23 Feb 2008 07:06:56 GMT

Hey Cheddar and Hauling, you are neat and knowledgeable responders! Pragmatists! Not naysayers, just doubters in the ability for us to reverse-engineer the FICO model to the extent that we can put some general numbers, per category, and in total FICO score, on what has heretofor been conjecture. I know we can do it! Give me the data, and I will give you the math!

We have such diverse talent and experience on here.

We can put numbers on the previously ephemeral!

It will answer, once compiled, a lot or the questions that flow day-to-day through this forum.

People want to see immediate impact.

I am an immediate imapct kindfa guy, if you handt noticed!

Message Edited by RobertEG on 02-22-2008 11:06 PM

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

RobertEG — Sat, 23 Feb 2008 07:16:51 GMT

Vista, the goal is not to be done. It is to begin.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

haulingthescoreup — Sat, 23 Feb 2008 15:57:48 GMT

@RobertEG wrote:

Hey Cheddar and Hauling, you are neat and knowledgeable responders! Pragmatists! Not naysayers, just doubters in the ability for us to reverse-engineer the FICO model to the extent that we can put some general numbers, per category, and in total FICO score, on what has heretofore been conjecture....

I'm not even so much a doubter in the ability to do this. Given enough time, data, caffeine, and days missed from work, anything can be cracked.

My concern is for when one set of numbers gets crunched and seems to work pretty well for one person, and they're posted. Those of us who've been analyzing our own score changes for several months know that there's a lot we don't know, but new members here are apt to take those predicted score values as applicable across the board, as plainly they are not. My own numbers are very, very different from what I would have expected from your formula.

I just hate seeing the new people here get even more confused (and upset!) than they already are. So I hope that you guys go for it, but preliminary is preliminary. Good research always regards early results with LOTS of skepticism!

And BTW, there is a thread on Credit in the News quoting Craig Watts of FICO as saying that FICO 08 will start up in May. But I'll bet a nickel that it's going to take a long, long time to be widely adopted. So we'll all have 6 FICO scores to keep track of, 7 when you included both the old and less-old versions of TU out there, and lets not think about the auto-enhanced and credit-card enhanced versions.

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

acadia11 — Fri, 14 Mar 2008 16:31:56 GMT

Some functions do not have inverses, meaning you can only go in one direction, you can't go backwards. The math is pretty simple to prove this one, but I don't want to bore you with the details. But you could keep taking the info in your credit report, and try and see if you can come up with the function that can give you your fico score. What you would need is a bunch of individuals credit info to see if your model accurately mimics fico.

Message Edited by acadia11 on 03-14-2008 09:31 AM

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 14 Mar 2008 17:49:33 GMT

I'm up for it..if someone is smart enough to take all this on we should all be supporting them!

... I know just looking at my own fico report and my DH and trying to figure out what is going on makes me crazy.. !!

I think the fico machine has many flaws (change in status dates being one of them that effects scores)..if you can figure it out great as half the time I don't even think the fico folks know how it work's and can't explain it.

good luck..I'm rooting for you

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Fri, 14 Mar 2008 21:34:53 GMT

Check out the Web sites in my .sig...they contain much information about FICO scoring, including a CR breakdown and listing of the variables affecting your FICO score (utilization, average age of accounts, % of accounts delinquent, etc.).

Re: Wild idea to stick it to the man. Who's in for a reverse engineered FICO calculator?

Anonymous — Sat, 15 Mar 2008 02:16:19 GMT

Ok, Chuck G you are on to something. For all the other naysayer. Reverse engineering precisely to a point or so can be done with Fico. The question has anyone done it and does anyone care to? I’m not sure where the value is to do such a thing. The general guidelines allow you to boost your score. But perhaps I missed the point, if you could figure out your score and plan to strategize knowing how to use your personal analysis. Then what would this be worth to people? On a personal note, this is what I do for a living. For 14 years, I reverse engineer software solutions of competitors to understand how one has achieved a mathematical formula. The next factor is peer reviewing if the calculation is correct. Which is what everyone on this forum can help me do? So here is what I would and I am considering it if I get support from the forums and the data required validating the calculation. Perhaps this website moderator wishes to have this solution for their website. I personally have my fico score reports from Equifax for each month for 10 years now. All I would do is create an online database website that allows everyone to log all reports based on the variables found at http://www.myfico.com/CreditEducation/WhatsInYourScore.aspx There is a comprehensive guideline of how each score is affected. Now if I enter all my Equifax reports using the guidelines and calculator I create. I can make the software reverse engineer the mathematical formula used to get to the Fico score Equifax issued me. I suspect what Fico has done has created 5 different relationship databases. The common thing I believe they have done is created an adjustable minimum and maximum value of what a personal score can be. Virtually creating unique scores for each person. This is why each person is getting different scores with similar data. So if you have been reported for over 10 years, the Fico software would calculate a trend over the period of time. Since this is a private scoring system, they don't have to get rid of your info after 7 years. Sure they can publish it, but you don't know what they have collected on you over your lifetime electronically. From this baseline, I believe they attack it with 5 different relationship variables in the database to figure out your credit score quickly so you can buy that Plasma. So basically each personal has a baseline that all FICO variables are applied to dynamically based on one common thread. The Calendar Date. For example, this is how I believe they can make it flexible for each person, since each person is different. Now I should mentioned, I specifically reverse engineer much more difficult calculations to keep it short in the medical industries, all day long for years now. So for the FICO score to output a variable score is simply a matter of reverse engineering the values that led to the personal score. Every person is different and Fico has a flexible calculation that takes into account many variables. (this is way beyond excel guys) So what I would do with this online software. Is allowing all of you to input what ever data you have anonymously. The software would calculate how that individual credit score was calculated. For example if Length of Credit History is worth 15% of a score that can only max out at 850 points when combined with all the other variables. Then the variables are Time since accounts opened, time since accounts opened by specific type of accounts and time since account on activity. I believe entering the above data for that one database module; the software can calculate minimum and maximum values for each variable within a category like “Length of Credit”. Now if I get 1000 entries into the system, I can correlate the common values of responses for all the entries allowing me to identify how FICO calculates their minimum and maximum values for each person’s baseline. This could be dynamically displayed online for everyone to see. If I had to guess, the chances are the values are the nearest in range of scores over a period of time. So they will take the minimum and maximum values you scored in each area and analyze your past history to find the closest response in ranges. This would create peaks in graphs and would allow them to see missed payments, repos and so on. So for example, making payment on time and paying off at a certain ratio would create a minimum and maximum value in ranges to be measured. This creates a baseline that can be measured. From this I believe Fico calculates 5 Categories X 26 Variables which can have many types of options against that baseline. I think this is what confuses most of you. Here is the basic example of what I am trying to say. Take Length of Credit History. It has 3 Variables. So I need to know what “the type of accounts” this area would record. So if there are 3 types of accounts. I have a possibility of 9 variables within 1 question. This is how I believe FICO has compiled their data. On top of it, if this calculation was done in 1950, please we have advanced so much since then. Hackers hack the pentagon and you think FICO cannot be figured out. As I started out this post, does anyone care to?