ILLUSTRATION: KATMANDU JOURNAL
THE MEDIAN American entrepreneur is 42 years old. He—and it usually is a he—has spent two decades building a credit history, probably owns a home and has a FICO score north of 720 (a three-digit score from 300 to 850 that measures how reliably someone repays debt, with higher numbers signalling lower risk). He looks like a safe bet. A 28-year-old with six months of rent payments, a thin credit file and a FICO score stuck at 660 looks like trouble. Yet when both start a small business, their default rates are indistinguishable. One of them is priced wrongly.
A new working paper from the National Bureau of Economic Research by Christopher Hair, Sabrina Howell, Mark Johnson and Siena Matsumoto documents this distortion and shows how to fix it. The culprit is more than the fact that young people have had less time to repay debt. The FICO formula explicitly rewards length of history. That is a feature rather than a bug. But it turns into a bug when the same score is used to underwrite business loans. The applicant with an extensive record gets a halo; the applicant with strong recent revenues but a short file gets a thumb on the scale.
The authors analyse 1.1m loan applications and 74,000 originated loans from three fintech lenders. They exploit a neat natural experiment: one platform forwards the same application packet to 111 different lenders, some of whom rely heavily on cash-flow data from bank statements and others who stick closer to FICO. Because the same applicant goes to multiple lenders at the same time, the researchers can isolate the effect of underwriting style from the effect of borrower quality.
Younger owners—defined as under 40—are 2.4 percentage points more likely to be approved by a cash-flow-intensive lender than by a traditional one. That equals a 12% boost at the mean approval rate. The effect is concentrated among applicants with middling FICO scores, between 670 and 740. These are subprime borrowers who might reasonably be denied credit. They are people whose scores are depressed largely because they are young. Cash-flow data, which captures recent inflows, balances and overdrafts, shows that many of them can service debt just fine.
The sceptic’s question writes itself: perhaps these lenders are just taking more risk. The data suggest otherwise. Among a hand-collected subset of 46,000 applicants, survival rates are no lower for young borrowers approved by cash-flow-intensive lenders. Loan officers at one of the lenders are randomly assigned applications; those who rely more on cash-flow data approve more young owners without seeing higher default rates. This is regulatory arbitrage. It is a better measurement.
The paper’s methodological contribution is a tool the authors call TACO—Tail Analysis for Comparative Outcomes. It asks this question: if a lender switched from a FICO-only model to a cash-flow-augmented model, which borrowers would see the biggest change in their predicted default probability? The top decile, those whose predicted risk rises most, are disproportionately older. The bottom decile, those whose predicted risk falls most, are disproportionately under 35. Among applicants with FICO scores below 740, the under-40 group is twice as likely to be in the benefiting tail as in the harmed tail.
That pattern matters for policy. The Consumer Financial Protection Bureau is currently writing rules for open banking, which would let customers share their transaction data with competing lenders. Incumbent banks warn that this will weaken underwriting standards. The evidence suggests the opposite: standards become more accurate. The problem is that FICO is a bad score. It is that a good score calls for a lengthy run-up, and a lengthy run-up is something young people, by definition, do have.
One irony lingers. The Equal Credit Opportunity Act permits lenders to favour applicants aged 62 or older. It forbids them to discriminate against the young but it does require them to correct for inputs that are themselves age-correlated. A system that uses FICO exclusively is therefore compliant and unfair. Cash-flow underwriting is an act of charity. It is just less biased.
The median founder may still be 42 ten years from now. But the entrepreneur who is 32, has $100,000 in monthly inflows and a clean record of overdrafts will look like a phantom to the person holding the credit file. The data are already there, sitting in the bank. Lenders just need permission to look. ■
