Credit risk assessment is largely a job for credit score algorithms, mass market models and systems that are relatively new in terms of the modern credit industry’s lifespan.
Before computers and data science were fully adopted by the credit sector, lending decisions were made by credit managers employed by banks or department stores, who then collected information about each individual borrower.
“Part of the job of the credit manager was to try to make inferences about the person’s personality, whether they seem to be organized and mature and responsible,” said Josh Lauer, a professor of media studies at the University of New Hampshire and the author of the book “Creditworthy.”
In other words, borrowers who were judged to be disorganized, immature and irresponsible were usually turned away. A credit manager could reject an applicant if they suspected they were an alcoholic, or for more blatantly discriminatory reasons related to someone’s race, sex or, for women borrowers, their marital status.
Borrowers had to to prove their creditworthiness under these conditions, which is why educational films like the 1960s short, “The Wise Use of Credit,” came to exist.
Transition to algorithmic credit scoring
A surge in demand for credit during the second half of the 20th century helped motivate lenders to adopt credit scoring algorithms. For one thing, algorithms were more efficient.
“It just took too long to have each of these credit applications vetted by an individual in real time,” said Lauer.
Credit bureaus started to computerize their massive consumer records in the 1960s and 70s. But computers had limited memory back then. Bureaus kept data like how many credit cards someone had, while more nuanced variables, like how responsible a borrower seemed, were dropped from credit records.
Regulation in the form of the Equal Credit Opportunity Act of 1974 made it illegal to deny credit based on factors like race, sex, marital status or religion.
In some ways, it was only a matter of time before lenders would move toward a data-science approach to credit scoring and lending.
“It replaced a lot of the human underwriting with an algorithm that allowed them to make decisions consistently and to also monitor how much better their decisions were,” said Sally Taylor, vice president of FICO Scores.
Introduction and adoption of “universal” credit scores
Credit-scoring algorithms existed as early as the 1950s. FICO, since its founding in 1956 by William Fair and Earl Isaac, designed credit score models for lenders.
But these algorithms were specifically designed for individual businesses and their unique customer base. You couldn’t apply a bank’s credit scoring model to a department store because they served statistically distinct customers.
In 1989, FICO released its first “universal” credit score that lenders could just buy and use instead of commissioning a custom-designed score for customers. But it took time to convince lenders to adopt it.
“I did a lot of mortgage broker, mortgage originator, you know, lunches, at that time, explaining the concepts of scoring,” said Taylor, who was then a FICO product manager.
The watershed moment for FICO and the mass market approach to credit scores came in 1995, when mortgage giants Fannie Mae and Freddie Mac decided that every mortgage application would need a borrower’s FICO score. That effectively cemented the credit score as one of the basic metrics of credit risk today.
Related Links: More insight from Kimberly Adams
The history of credit scoring in the U.S. is also related to the history of consumer surveillance, according to Josh Lauer. He says that credit scoring algorithms and computerized databases were built on credit bureaus’ vast consumer data infrastructure, dating back to the late 19th century.
The 1970 Fair Credit Reporting Act is another important piece of regulation in the history of credit score algorithms, which requires that credit reporting agencies make sure their data is fair, private and accurate.
What comes after credit score algorithms? FICO has been exploring the potential use of machine learning. VantageScore has already applied some level of machine learning in its latest 4.0 scoring model. Is artificial intelligence next?