The goal of credit underwriting is to make prescient decisions. All credit has an outcome – either it pays as agreed or it does not. The role of the underwriter is to best predict that outcome which is why it is critical to limit the amount of bias inherent in any decision. While we have looked at overt bias in credit underwriting in the past (HERE for example), in this article we look at a particular bias inherent in all bank’s processes and why it matters. In the second and final part in this short series, we will also look at ways on how banks can mitigate this bias.
It is called the Gambler’s or Monte Carlo Fallacy, and while the theory dates back to 1796, it came to reality one hot August night in 1913 in a casino on the shores of Monte Carlo. It was typical roulette game, and black had come up six times in a row. The table was abuzz as with each spin the patrons thought red was sure to come up next. By spin 13, the whole casino stopped and was gathered around the table as black was still coming up and almost all the money in the house was betting on the reversion to the mean and that red would be next. Millions of Francs were now at risk.
By the 17th spin, red was yet to show up, and there were now cries that the game was rigged. Play stopped as management assured everyone that it would make no sense to rig the game as gamblers were free to bet in any direction and that balls and personnel were consistently changed throughout the process. Even with this discussion, few players switched to black as everyone knew red had to come up sometime. Unfortunately for most, it wasn’t until the 27th spin did that ball drop into a red number.
Among fair games, where outcomes are independent of each other, such as roulette, dice or coin flipping, the odds for every event are identical. For American roulette, because of two green slots, the odds that the ball falls into any one of the 18 red slots is 47.4%. After 26 spins where the outcome was black, the odds of red coming up next is still 47.4%.
In life, there are many fair events where each outcome is independent. Credit is a good example. If you have five loans from unrelated borrowers and projects, the outcome of each are independent of each other.
Sequencing In Credit Decision Making
Unfortunately, our processes often inject bias into our credit decisions. This makes our decision making suboptimal and either over or under-allocates capital. This isn’t a result of any conscious effort, but a byproduct of human nature. Just like each gambler in Monte Carlo that night believed that red has to come up next, when you get a series of good credits you subconsciously start to look for a bad credit until you find it. This can have you turning down credit you should be approving and approving credit you should be turning down.
This assertion can be validated in a number of ways, but one is a small sample test we conducted when we assigned a group of loans to three different underwriters. Each underwriter was given a sequence as to look at the credit. The credit was scored on a one to 20 scale with the first ten grades being a pass credit (the lower score, the better). The underwriters did not know the auto store and were only asked to approve or decline the credit with no consequence.
In the table below, the credits are labeled A thru E and the sequence is detailed that was presented to each underwriter along with the pre-scored, objective value. The credit is green if the underwriter issued an approval and red if they issued a decline.
As can be seen, in all cases there was at least one pass credit that should have been approved was declined. In two of the three cases, two credits were declined. When credits were presented in order of credit quality as judged by the auto score, the last two credits were declined. However, when the order was reversed, only one of those credits was declined. When the credits were randomized, credit C was declined when it should have been approved.
A Larger Study
In a 2016 paper titled “Decision-Making Under the Gambler's Fallacy: Evidence from Asylum Judges, Loan Officers, and Baseball Umpires,” Professors Daniel Chen from Harvard Law School and the Toulouse School of Economics; Tobias Moskowitz from the University of Chicago and the National Bureau of Economic Research; and Kelly Shue also from the Booth School of Business at the University of Chicago and the National Bureau of Economic Research looked at the Gambler’s Fallacy as it applies to baseball, judges and loan officers. Similar to the above findings, Professors Chen, Moskowitz and Shue found bias in each area. When it came to loan officers, they also tracked the loan to see how the approved loan performed.
What the research found was that the loan officers got it wrong 8% of the time simply because of the sequence of how the loans were presented. This error coefficient was magnified for loan officers with under ten years of experience. For those less seasoned loan officers, they were 23% less likely to approve a loan if they approved the previous loan.
Further, the research found that the Gambler’s Fallacy is magnified when credit streaks are involved. After one negative credit, a loan officer may only have a 2% error rate when they were presented with a similar credit that they should have declined, but they approved. However, if the there are two negative credits in a row, then the error rate is closer to 8% and if three negative credits in a row, then the error rate is even greater.
What This Means
In part two of this post, we will look at how banks can take steps to mitigate this bias. However, until then, ponder the fact that right now, it is possible that you could have 8% more performing loans on your books or that 8% of your portfolio probably shouldn’t have been approved. For a bank with $500mm in loan footings, that is likely about $3.2mm of capital that is at risk that shouldn’t be if you mitigated this risk.
Submitted by Chris Nichols on November 28, 2016