Underwriting

Anyone applying for a loan or seeking to take out insurance doesn’t automatically get approved. First, the company checks how likely it is to lose money. This exact check is called underwriting. In the end, there’s a decision: approval, rejection, or approval at a higher price. The term comes from the insurance business in London: there, financiers signed their names underneath a ship’s risk that they wanted to take on. This signing underneath became the English word underwriting.

Why every loan has a price for its risk

Banks and insurers don’t make money from everything going well. They make money by correctly assessing risks. If a bank grants a thousand loans, some portion of them won’t be repaid. These defaults have to be covered by the interest from the other loans. If the bank underestimates the risk, it makes a loss.

The opposite mistake is just as costly. Anyone who checks too cautiously rejects customers who would have repaid without any problems. These customers go to the competition. Good underwriting therefore doesn’t mean the strictest possible check, but rather the most accurate one.

How important this is was shown by the financial crisis starting in 2007. In the USA, mortgage loans were granted to people whose ability to pay had barely been seriously checked. When many of these defaulted simultaneously, banks worldwide got into trouble. Since then, regulators have prescribed much more precisely how checks must be carried out.

From checklist to scoring model

In the past, underwriting was purely manual work. An experienced employee, the underwriter, looked at payslips, bank statements, and contracts. A judgment emerged from experience and fixed rules. Such rules still exist today, for instance an upper limit on what proportion of income may go toward loan installments.

Nowadays, software mostly handles the preliminary work. It calculates a score, i.e. a number expressing the estimated probability of default. This is based on data from hundreds of thousands of previous customers whose loan outcomes are known. The program searches this data for patterns: which combination of age, income, occupation, and payment history frequently went wrong? It then applies these patterns to new applications.

A common misconception is that such a model knows the future of a single individual. It cannot. It only states that, statistically, about three out of a hundred similar applicants will default. For the individual case, this remains a probability. That’s why in Europe, particularly consequential decisions may not be left to a machine alone. Customers have the right to demand a justification and to involve a human.

From mobile phone contracts to IPOs

You encounter underwriting more often than you think. When shopping online on invoice, a model checks in the background within a fraction of a second whether you’re likely to pay the invoice. The same thing happens with mobile phone contracts, installment purchases, and car insurance. That young, novice drivers pay especially high rates there is a result of underwriting: their accident rate is statistically higher.

The term has a second meaning that appears in business news. When a company goes public, an investment bank takes on the underwriting of the new shares. It guarantees the company a certain proceeds and bears the risk if the shares sell worse than hoped. So here too, it’s about taking on a risk in exchange for payment.

In reports about financial start-ups, so-called fintechs, underwriting is almost always the selling point. These companies claim that with better models and more data, they can serve customers whom traditional banks reject. Whether that’s true only becomes clear after years, based on actual default rates. Until then, it’s a claim, not a track record.