Several years ago, I began a conference presentation by making the statement: “To be successful at fair lending, you must learn how to properly discriminate.”
The statement was obviously meant to be provocative, and it must have worked because it was quite some time before I was invited back (just kidding). The statement, however, is accurate; and the point that is being made is absolutely critical to fair lending compliance. However, explanation is necessary.
When lenders make loan decisions (approval or denial), they must discriminate. Simply, they have to determine which customers will repay the loan and which will not. Why this is critical to the viability of the institution is obvious. Too many bad loans and they are soon out of business. To further be successful at fair lending requires proper discrimination. So what does this mean and how is it done?
One of my professors used to say that science is the art of simplification, so let’s start with a basic question: How does a lender know which applicants will repay and which will not? Clearly, there is no way to answer that question with absolute certainty; therefore, institutions must rely on the probability of which applicants are most likely to repay and which are less likely to repay.
This is where discriminating (and doing so correctly) becomes critical. Lenders must have a method of determining which applicants fit into which group.
It is impossible to know the future and, therefore, a perfect method for doing this does not exist. The lender, therefore, considers the attributes of the loan in an attempt to identify good and bad credit risks. Credit score, for example is simply an index of the likelihood of repayment. Lower scores are associated with a higher probability of default and higher scores a lower probability of default.
There are, of course, a number of factors that a lender must take into account in order to discriminate between good loans (high probability of repayment) and bad loans (low probability of repayment). Which attributes are considered, and how they are considered, then make up the institution’s policy. How effective this policy at separating between which loans fit in the approval group and the denial group is associated with the level of risk.
Let’s develop this a little further with a simple example. Assume that a lender uses (3) core criteria in the loan decision: credit score, DTI, and LTV with the criteria being a credit score of at least 620, and DTI and LTV maximums of 40% and 80%, respectively.
If the lender never deviates from these factors, in that all loans meeting all three criteria were approved and those that did not were denied, then it would be easy to examine a sample of loan applications and put them into groups based on these factors. In this case applications and denials would be clearly distinguishable based on the bank’s policy and consistency in loan decisions readily apparent. This is illustrated by the Figure 1, below.
We know, however, in the real world it is never this simple. For example, borrowers that meet the basic criteria may actually be bad credit risks, and vice versa. Therefore, approvals and denials may share some of the same characteristics after factors come into play that offset the positive or negative attributes.
Just like the three factor example above, however, additional factors considered in the loan decision must be identified and quantified in order to be able to fully evaluate the loan decisions. Again, a diagram will help illustrate.
As shown in Figure 2, there is overlap between the approvals and denials in the pool of loans on the left (i.e. some approvals and denials may share some attributes but the other attributes are identifiable which still makes approvals distinguishable from denials. For the pool of loans on the right, however, some of the approvals and denials look just alike, therefore, they are not as easily distinguishable.
Hopefully, the point being made is clear: In order to be successful at fair lending, lenders must discriminate, BUT do so based on relevant criteria. When loan decisions can be tied back solidly to loan attributes, the risk of fair lending issues is low. The blurrier the lines becomes, the more risk exists.
