@dcal wrote: This is a good subject, You want to know something crazy. I personally, think that Experian has the algorithmn and software to calibrate the Plus scores and other Fako scores into the Fico scores if they wanted to from the Experian site. If they wanted to. They just want to offer the worthless ones to consumers. But I have noticed something weird. By applying for a Amex and a new car (they pull Experian only) where I live and getting a mortgage approval. I noticed that on all occassions My Experian is only 8-10 points higher than my Equifax with pretty much the same information. My TU04 is only 6 points lower than my TU98 here and Credit Karma has consistently been closer to the TU04 (only 3 points difference all year). I know this may change, but this has been the case 2011 until this month. I know they are fake scores, but they have been dead on. It just make me wonder, if they have a algorithmn to calibrate back forth between the Fako's they crank out for us and the Fico's they let us purchase and the banks and creditors. The credit reporting agency's have had enough time to perfect what they do. Maybe they want us to think it is more complicated than it really is. I think they have the technology to do this, but for whatever reasons they want us to think they don't. They could offer the consumers the Fico's that lenders use, but they just don't want to. I always think this is the case, because I am a forward thinker. Dcal Most likely any large bank could estimate FICO scores pretty accurately by statistical data crunching. However, the Terms of Service under which a lender buys FICO scores may have legalese that specifically prohibits reverse engineering. If a scoring algorithm is both truly independent of FICO scoring and equally good at predicting a person's credit risk, that algorithm would probably be a rather poor predictor of their FICO scores. Suppose for example that for a certain group of consumers the FICO score has an r-squared (a standard measure in statistics of how well a multiple regression or similar model fits the testing data) of 80% for predicting their credit risk, meaning it accounts for 80% of the variation in their credit risk. Suppose a certain FAKO score also has an r-squared of 80% but was developed completely independently of FICO. Well, that FAKO score would have an r-squared for predicting a person's FICO score of about 64% (80% times 80%). My point is, from the lender's perspective such a FAKO score might have comparable statistical value for classifying applicants according to their credit risk. But from the perspective of the consumer, we don't really care how well that FAKO score works from the lender's point of view. We want to know how will lenders see us? Since most lenders use FICO scores, this means as consumers we need to know actual FICO scores. A big bank certainly could build a model for estimating the FICO scores of their customers, by purchasing actual FICO scores for a large sample of their customer base and using that as the dependent variable in a modeling project. Indeed, for a large bank it might be easier to build a FICO-score-predictor model than it would to build a default-predictor model: to build a default-predictor model they would have to follow their sample over time and see which customers defaulted. To build a FICO-score-predictor model they could just pull FICO scores and crunch data. But if the bank started using their FICO-score-predictor model to compete with FICO in any way, they would probably hear from FICO's lawyers. Much safer either to purchase FICO scores or to build their own internal scoring model. In any case, what incentive does somebody selling FAKO scores to the general public have to strive for high predictive accuracy anyway? If their "credit score" looks like a FICO score, how many of their customers will know the difference? How many people know about the totally free and anonymous FICO Estimator, which gives an estimate (probably better than most FAKO scores) of your FICO score from your answers to a few questions? I work with a number of purchased databases and analytical tools in my job (biopharma R&D). Many vendors have specific contract language limiting how we may use their data, and how we may not use their data. In particular, vendors often want to make sure we don't license their product for a short time so we can reverse-engineer it, then switch to our own implementation and stop paying them. I have never worked in the financial industry, I am in pharmaceutical research. My interest in FICO scoring began from intellectual curiousity about the algorithms used, because I use very similar algorithms in my work. Lenders want to know, which of these applicants are most likely to have difficulty paying their bills in the near future? My company wants to know, which of these people are most likely to have a hear attack or catch HIV or experience whichever medical event is being studied in the near future? FICO wants to know, of the various facts in a person's credit history, which of them are the best predictors of credit problems? My company wants to know, out of various medical tests we could do, which of them do the best job of predicting who will respond to our drugs? For an example of an important medical scoring algorithm, google Framingham risk score The Framingham risk score is an algorithm that estimates a person's risk of heart disease; it is frequently used by physicians to identify people who are at high risk and therefore might benefit from interventions that lower their risk (such as drugs to lower blood pressure or cholesterol).
... View more