@folks19 wrote:
@Anonymous If you say so, I believe you, and you're probably right. I just didn't realize that was the case. I can understand that each bureau has a different algorithm for mortgage scores since they are listed as different versions of FICO score, but I wouldn't have thought that the FICO 8 score, which is used for all 3 bureaus, actually has 3 different versions. Thanks for the information.
@folks19 from my knowledge and opinion, I'll try to explain it as best I can:
The first thing you should know is the mortgage scores are not really "mortgage" Scores. Two of them are just a simple base model from 2004 and the other is the simple base model from 98.
Keep in mind the core algorithms that create the scorecards for each version are the same. The Segmentation factors for each version are the same across CRAs.
It's kind of complicated to explain and I don't know everything about it. @Anonymous could probably fill in more of the blanks about how the scorecards are created via logistic regression, but...
basically they have the core algorithms that they use to evaluate the datasets from the bureaus, and that core algorithm is going to be unique for each version, but the same core algorithms create the scorecards for that version at each CRA. Hence the segmentation thresholds are the same across CRAs for a particular version.
that core algorithm evaluates the CRA dataset and produces the respective scorecards, each with ~40 most predictive characteristics (out of ~500) and the most predictive weightings for those characteristics. (Characteristics and Scoring Factors could be considered almost synonymous.)
anyway, for example, this was done for score 8 at Experian and it produced the 12 scorecards that are used for version 8 at Experian. (It also produced the scorecards for the industry options of the version, but I'm keeping this simple.)
this was also done at the other two CRAs. However, due to the difference in their datasets, different characteristics were chosen as the most predictive and the weightings may be different as well, based on predictivity.
so there you have it, that's why you have differences among the bureaus even with identical data for the same version, imho.
now, you mentioned the mortgage scores. I want to make clear that version 4 and 5 are the same base 2004 version; the differences between them are based on the differences that were in the datasets evaluated that produced the scorecards for that version at those bureaus, as described above. (Score 3 was also created for Experian, but it's hardly used.)
Just like there are differences in score 8, there are differences in the 2004 version which is called 4 at TransUnion, 5 at Equifax and 3 at Experian.
The Experian "mortgage" Score 2, is a totally different version based on a 1998 model. Equifax and TransUnion had corresponding models, but they're just not really in use anymore.
so basically you had the 1998 version (Score 2), the 2004 version (Scores 3, 4, 5), the 2008 version (Score 8), then I guess the 2014 version (Score 9), and then the 2021 version (Score 10).
