Didn't know what I didn't know until today. I'm developing matched pairs of originated loans supported by a regression analysis. The universe is first mortgage loans on my LAR. My prohibited basis factor or concern is age. I'm making the following assumptions of items that MUST remain constant between control group and target group: 1)Loan Purpose same in terms of refi to refi and purchase money to purchase money; 2) location - since we are a national lender and there tend to be pricing adjustments based on locality I need to compare a NY loan to a NY loan and FLA loan to FLA loan; 3) CLTV; 4) term; 5) loan type; 6) Property type; 7) Credit Score. I also assume that I can't use the same loan in more than 1 comparison such as using Control Group loan #1 in a comparison with Target Group #1 and comparison with Target Group borrower #9. Correct? Variables such as race, sex, ethnicity should have no factor on underwriting or pricing so I don't have to control these variables. Any thoughts would be appreciated. Thanks.

If you are a national lender with significant volume, I would suggest that you engage a consultant to assist in the development of any type of regression analysis. Regulators tend to be very tough on the topic of model development and will scrutinize all aspects of what you put in your model and what you exclude.

Generally speaking, if you utilize a factor for pricing, you would expect it to be within your regression analysis. If you do not use a factor to determine pricing, regulators would not expect to see it within your regression or matched pair determination. So for instance, credit score, if you price at a single price (which is common within mortgage lending), a regulator may not expect you to group only those with a 780 credit score with other 780 customers. In their mind, a 780 would probably be similar to an 800 or 760. Much of this depends on the regulator, your volume, the product, etc. That is why some consultants are very beneficial in this area. Plus they typically will utilize a PHD statistician that can (sometimes) go toe to toe with the regulator PHD statistician.

Other prohibited factors (race, gender, ethnicity) should all be tested separately.