-
Peer-to-peer lender Upstart is aiming to identify young but high-quality borrowers by integrating applicants' grades, alma mater and even their SAT scores into its underwriting decisions. Banks could learn from this model, even though it may have shortcomings.
July 8 -
Once known as peer-to-peer, online marketplace lending grew phenomenally and evolved rapidly in 2014, as capital flooded in from institutional investors and traditional bankers sat up and took notice.
December 17 -
Some banks try to justify tight underwriting standards by arguing that borrowers with lower credit scores and those who can only afford lower down payments are more likely to default. But the research supporting this argument is based on outdated analysis of high-cost, risky loans.
October 22 -
RevolutionCredit is looking to partner with creditors on its behavior data software for underwriting and prospecting.
December 4
The success of online alternative lenders like Lending Club and Prosper has given rise to renewed interest in underwriting models that
Subjective data is difficult to interpret without a loan officer. But lenders might wonder whether loan officers are worth the investment when high-tech options are readily available. After all, these employees can be costly to hire. And an excessively cautious loan officer or one who is simply bad at his or her job can make decisions that cause problems down the line.
My
I came to this conclusion by studying loan and repayment data during the years 2010-13 for approximately 32,000 borrowers at a Chinese lender. This large lender specializes in making unsecured cash loans to households and small businesses. Loan officers view the borrower's entire file including financial statements, references, notes, credit scores, and even photographs before choosing an approved loan amount.
My paper calculates the value of these employees by creating an algorithm that makes underwriting decisions based only on the borrowers' hard information, such as their financials and credit scores. I then compare that information to the performance of the loans that the loan officers actually approved.
It is important to note that this approach does not measure the value of credit scores. Lenders have made extensive use of risk-based pricing since at least the early 1990s, and the effectiveness of this method is not in doubt. This paper simply compares the efficacy of using both loan officers and credit scores to evaluate borrowers against using credit scores in isolation.
My paper found consistent differences in how loan officers evaluate applications. One person, for example, approved a higher loan amount than his or her peers at every level of credit quality. Unsurprisingly, people empowered to subjectively analyze data will make different decisions based on their individual risk attitudes, ability and level of confidence.
Nonetheless, the average loan officer contributes three times his or her annual pay in additional profits each year, according to my calculations. Some loan officers are not profitable compared to the algorithm, but most help their employer rake in far more revenue than the lender would have been able to achieve on its own.
One might wonder how much the effectiveness of this algorithm influences the result. Could a better algorithm outperform loan officers?
The answer is likely no, since the algorithm is calibrated using the borrower's actual repayment data. This approach is analogous to traveling back in time to pick an investment portfolio in 2013 after observing data in 2014. Algorithms developed at the time of a loan's origination would be likely to produce poorer results.
Therefore it is safe to conclude that loan officers are a valuable resource for lenders despite their biases. Lenders can maximize their profits by using loan officers in conjunction with automated lending models that have access to extensive amounts of hard information and repayment data. While computers may have bested humans in chess and mutual fund management, underwriting is one area where men can still beat machines.
James Wang is a PhD student in economics at the University of Michigan. His research studies asymmetric information and financial risks in lending markets. This post is adapted from his job market paper "
