Odds Are, You Have Never Seen One of These

By Michael Cochrum

We like to know our odds.  What are the odds that you will be struck by lightning?  In the span of one year, that could be one in 700,000, but in your lifetime, one in 3,000[i].  Yikes! That is a little too close for comfort.  How about the odds of dying from cancer?  There are a lot of factors here, but roughly your odds are about 0.5%, or 5 in 1,000[ii],  about the same as your chances of dying in a car accident.  Living in America, your odds of being a victim of murder is about 1 in 6,000.  According to a 2002 Department of Justice study, the chances of you being murdered by someone close to you as opposed to a stranger is 74%[iii].  So, based on the odds of getting struck by lightning and the odds of being killed by someone close to you, why on earth would one ever plan a family reunion at an outdoor venue.  Would you agree, knowing our odds helps us to make better decisions?

I find it interesting, however, that when I conduct training classes with credit union lenders, many of them say that they have never seen a credit bureau odds chart.  How then, when the credit score is the primary decision and pricing factor for many lenders, can a lender make a solid lending decision without having ever seen the odds value attached to a given score.  Many of the pricing matrices that I have seen in use are very similar, even though the default models supporting them may be significantly different.  For example, I’ve seen lenders attempting to align their loan pricing tiers with competitors in the market place, not knowing for sure what credit bureau model their competitor is using.  This is the epitome of an apples-to-oranges comparison.  It is meaningless, and dangerous.

I assume, based on feedback from clients, that many lenders believe that a borrower’s score is simply a relative number on a linear scale, i.e. a 700 score is ten points, or 1.45%,  better than a 690 score.  Instead, the scoring scale is more logarithmic.  In other words, instead of a 700-credit score being  1.45% better than a 690, it is 22.58% better.  So then, if you raise the average credit score in your portfolio from 690 to 700, you have decreased the probability of default by nearly 23%.

Many lenders view a credit score as a grade, much like one would grade beef you find in the super market. The trouble with that understanding is that a meat grade is a measurement of what is, and a credit score is a forecast of what might be.  For that reason, lenders are often disappointed when a borrower doesn’t perform as expected relative to their credit score on the day a loan was originated.  In other words, US Prime Beef will always be Prime, no matter how long it ages and no matter how it may be cooked, because the grade is based on immutable factors, such as where it came from and what its contents are.  A credit score is a forecast based on what has been observed in a pool of loans over time.  It is predicted, then, how a borrower will perform based on the performance of similar borrowers under similar conditions.  If you bite into a Prime steak and find it very lean with less marbling, then the meat was graded improperly.  But if a 740 credit score borrower defaults on a loan, its because the conditions for performance for that individual has changed, not because the model is faulty.

So then, should one conclude that using a credit score is just a shot in the dark with no way of knowing, for sure, how a borrower will perform.  I think we know better than that.  But, if we know the probability of performance, we can better measure our risk and appropriately mitigate those risks across the loan portfolio.  Allow me to use the example above to demonstrate how this might work in the real world.  For this example, we will assume our risk-free investment return is 2.0% and that our target ROAA is 1%.  Also, we will assume that both borrowers have the same repayment speed and have an average monthly loan balance through the life of their loan of $15,000.  Finally, we will assume that our average loss on a loan is $5,000 and that the loss occurs at the 24th month in the life of the loan for both borrowers.

Credit Score
Probability of Default
Loss Given Default
Probability of Loss
Annualized Loss Ratio

Based on our assumptions and the probability of default provided by the credit score model, we now know that we should add 13 bps to the rate of borrower B to accommodate for the additional risk.  Why is this important?  It’s important because we typically underwrite individual loans and when we make exceptions, such as bumping a score, we don’t always account for the impact that the change in score would make on the probabilities of loss.  Certainly, a complete pricing model would include other significant risk factors, such as LTV and DTI, but this example demonstrates how we should look at the difference in risk between borrowers with different credit scores.


  1. Get your hands on your credit bureau default odds chart and determine whether your pricing scheme aligns with the predicted risk.  If you have never seen this document and/or do not know where it is, you can contact your bureau rep and they will send you a copy right over.
  2. Evaluate whether your portfolio is actually performing the way it was predicted to perform with what is commonly referred to as a retrograde analysis.  This can be done with software applications such as CUBI.Pro Analytics.  If your portfolio is underperforming, then you may want to review other underwriting criteria, or adjust pricing for the increased risk.
  3. Review other significant factors of default, such as LTV and DTI.  Two people with the same credit scores but different LTV’s or DTI’s can perform much differently.
CUBI.Pro works with credit unions across the country to support a data-driven decision culture and provide meaningful business intelligence that allows CU leaders to make better and more timely business decisions.  If your credit union is looking to enhance its business intelligence for decision making, visit www.cubi.pro for more information or email michael.cochrum@cubi.pro.


Popular posts from this blog

How to Estimate the Impact of CECL on ALLL

Measurements Matter

Are You a Coach or a Referee?