Subscribe to DSC Newsletter

I am assigned a project of building a credit score model for my company which currenly pays up to nearly a million dollars annually to DnB for their credit model. Could anyone share with me the kinds of techniques available to build a credit score model? We currently have SAS/STAT in house and I would start by using Logistic Regression to build a model. What other data mining techniques are available out there that compete with logistic regression? Thanks a lot.

Views: 5962

Reply to This

Replies to This Discussion

Check out this book

Credit Scoring & Its Applications (SIAM Monographs on Mathematical Modeling and Computation) (Paperback)
It's less a matter of what technique to use and more a matter of what data you have available to build the model. What historical data do you have on your customers (i.e. -credit/default history, demographic, financial and credit data).
I agree with Mike L that the choice of model type is less important than the quality of data, but to answer your question, there is a huge number of alternatives available, including General Discriminant Analysis, C&RT, CHAID, Boosted Trees, Support Vector Machines, Multi-Layer Perceptrons, Radial Basis Functions and many others - all with their own particular strengths and weaknesses. You really need to be able to compare the performance of these alternatives empirically and you need a software environment that allows you to do this easily. You will find this very difficult to do using just SAS/STAT.

You also need a way of deploying your models into a live scoring environment. Take a look at this:
check my blog here on Analytic Bridge, metholdological references!
hi, have you build your model sucessfuly?
is Mike right when chaims that data are more important than method & especially if you know LR ? :))
I haven't built my model yet. Any suggestion? I am looking at a SAS training course on this topic,
great book about credit risk in SAS with code examples is "Olivia Parr Rud: Data Mining Cookbook".
that's all you need. and she uses LR :)
Hi, Jozo:
Thanks for the recommendation. I took a look at the contents and I did see a chapter at credit scoring application. However, does it also tell you how to assign credit limit in dollars after you successfully calculate the credit score? I didn't see that in the contents of the book and hence I assume this book does not cover this topic. Do you have suggestion on this topic? Thanks. By the way, which IBM office are you working at?
Credit limit is a different topic.
Basicaly there are two pillars for you decision:
1. Risk score - if risk is too high, reject applicaiton - there is guide 'how to choose right cut-off' in the book. it mostly depends on your own strategy.
2. Net income of customer - higher net income, higher limit.
Exact function limit=fn(risk, income) depends on what credit products you sell, how good is your collection process, etc. Think twice. If not sure, hire consultant and learn from him.
Good luck.
Thanks. Does IBM offer this kind of consulting service to retail industry? Where can I get this kind of consulting service and learn from them? What I found so far from different companies that I talked to was that they all wanted to sell you their products but won't tell/teach you how they do it. They all want you to keep going back to them for more and keep charging you.

Also, does this process of assigning credit limit involves some kind of optimization? Something like you want to optimize the revenue while containing the exposure to default to certain dollar amount?
You don't other software than SAS/STAT :)

Of course - ultimate goal is to have optimized revenue, but it is long run approach, require lot of difficult decisions. And you always have to start with data gathering, continue with first models and climb higher step by step.

I guess there are many suitable software-independent consulting companies around the world. IBM has experience, other companies also have. Try to find one close to you and get consultants on your site.

Or write me an email with contact/details and I can try to find someone in IBM to help you.
Agree with the many posters here that data is more critical than technique in developing a model. In credit scoring, there are several other considerations, including model stability (will it validate over time?), model interpretability (can the users understand it?), model palatability (does it produce reasonable adverse action reasons? do the variables pass ECOA muster?) and ease of deployment (does your implementation platform support the computation of score and its corresponding adverse action reasons?)

For some additional insights, please check out our web site at Our Xeno software has a proven track record with many of the largest US banks.


On Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service