Why Banks Make Bone-Headed Decisions

 |  Includes: KBE, XLF
by: Felix Salmon

Stephen Dubner passes on what he calls the “bizarre story” of a man whose bank is unwilling to give him a $50,000 loan, even if it’s fully collateralized in cash.

Dubner seems unsure that the story’s entirely true — but it rings absolutely true to me. So why does Dubner find it so hard to believe?

The answer is that Dubner’s looking at the bank in functional terms — as a place which makes profits by lending out money at some spread over its cost of funds. If such an institution were perfectly rational, then it would almost certainly accept this customer’s offer. There’s no real opportunity cost to extending this loan, because if the bank turns down the customer, then it’s also turning down the funding source for the loan. (The offer, of course, is essentially self-funding: the customer is offering to put $50,000 on deposit and then use that as collateral against a $50,000 loan.)

But of course there’s no such thing as a perfectly rational institution. And as banks grow, they become hyper-aware of the number of places where errors in judgment can cause losses. And their reaction is always the same: they reduce that number.

This does make a certain amount of sense — if you don’t let bank managers approve loans, then you won’t get a rogue bank manager throwing away millions of dollars of shareholders’ money. More importantly, if you approve all loans centrally rather than at the bank-branch level, then, at least in theory, you can see bank-wide risk exposures emerging which individual branch managers would never know existed.

Of course, centralizing loan approvals, and computerizing them so that they’re automated, has costs as well as benefits. For one thing, it means that errors of judgment at the loan-approval level can cost billions of dollars, rather than millions. And it also means that you’re building model risk into the system, and losing a lot of the natural diversification that you get from a heterogeneous set of individual bank managers making individual decisions to customers whom they personally know.

But the people making the decisions to centralize are also the people responsible for making the centralized lending decisions, and they tend to be very sure of themselves and their models. So bank managers get ever less freedom to do sensible and profitable things, while computers churn away making decisions which sometimes defy common sense.

I’m quite sure that there’s no one at the bank in question who thinks that its response in this case made sense. But that’s a known issue when you automate underwriting decisions: computers don’t have common sense. Some unknown proportion of sensible loans will end up not being made. But the bank will sign on to such a system anyway, because it’s cheaper than having humans make those decisions, and because it reckons that computers will make fewer errors than humans in aggregate.

After all, it’s not exactly every day that someone walks into a bank and essentially offers to lend himself money, while paying the bank a decent rate of interest at the same time. If it did happen every day, then I’m sure someone at the bank could program the computer to set attractive terms for such people. But it’s rare enough that it’s not worth the time and effort involved in doing so.

Now, from the point of view of the customer — and, indeed, of the customer-facing loan officer at the branch level — all of this is extremely frustrating and Kafkaesque. Which is one reason why it makes a lot of sense to bank with a small community bank, or a credit union, rather than some enormous centralized franchise.

But yesterday I spoke to the woman who got turned down by her credit union for a personal loan, and who was forced to go to a much more expensive installment lender instead. (I’ll return to this story once I have a bit more information.) Her experience at her credit union was also frustrating, and constrained by computer-set rules. So even credit unions with only a handful of branches aren’t immune from this syndrome. But I feel safe in saying that there’s a direct correlation between the size of the institution, on the one hand, and the chance of running into this kind of frustrating stupid-computer situation, on the other.