07-06-2013 09:39 PM - edited 07-06-2013 09:46 PM
Only reason I use r^2 is because I was getting better results... Do you have a better approach to this? I checked with a old college buddy who is at Wolfram and he said I was going about it all wrong, like you are suggesting. Now I am thinking of using another approach, but my Stats knowledge is limited. Any suggestions?
Wolfram makes some very good software for both analytic work and symbolic math. MathWorks also has some very good s/w focused more on engineering analytic work than Wolfram. I have both of these. You might want to check out SAS which focuses on statistics. I don't have that but it has a good rep.
For general background on Credit Scoring check out "The Credit Scoring Toolkit" by Raymond Anderson. There are lots of references for further investigation there. I'm sure some universities have programs that cover this. I'd focus on the ones Wall Street quants come from. There are lots of similarities.
If you had a 100,000 credit reports and FICO scores you likely could get a farily accurate estimate of what the scorecard groupings are but any regression analysis will, on the basis of fundamental statistical principles, produce algorithms further off than FICO scores themselves. Again. you will not be able to get this data. It's highly protected intellectual property and those that get such large datasets, like the CFPB, have to sign agreements not to redistribute or reverse engineer the data. You probably can get the agreements the CFPB entered into through the FOI act though you will not be able to get the raw data.
07-07-2013 10:09 AM
I hate math!!!!!