I work for a company that employs our own internal credit scoring. We do not use any bureau scores (FICO, Beacon, Emperica, nor any others) - but rather create our own. In our modeling we do find that the models are predictive and give us a very good idea if a person is likely to default. We do find that not only your individual performance, but those in your neighborhood and zip code can also be very predictive with regard to performance. The models tend to predict groups as a whole and there are always exceptions. I'd be happy to try and answer any questions those on the board have with regard to credit scores as used by creditors, credit models and other statistical models. I will not be able to reveal my company nor any proprietary information.
So your model predicts the future performance of an individual based on the past performance of people who live a mile away? I guess you can use whatever you want at your company to grant credit. This sounds pretty shady to me. I've lived in high end zip codes and zip codes considered low end. Those at the high end that I knew were much more likely to get in credit trouble than those at the low end. Maybe you might get a few more 30 and 60 day lates at the low end, but you'd get a lot more bankruptcies at the high end...
Credit Scores Absent any other information about the consumer, having information and using it with regard to the aggregate credit history of the zip code we do find to be predictive. Agreed, its not nearly as good as individual data about the consumer, but it does generally work.
I'm curious, how does your company/model define default? Is this model an algorithm formula? Does your model use statistical formulas? How much weight is given to actual historical defaults (both individual and group/subset data points)? What is the verification method of the model, and degree of accuracy/margin of error? Is there a specialty with your model/formala? (i.e. is additional weight given to an item inyour industry-auto, mortgage, etc.)?
