A deposit is core when it runs the client's business
After March 2023 I stopped classifying deposits by their age. The balances that stay through a frightening week are the ones a client's payroll, payables and collections depend on, and a bank should build its securities book, its pricing and its bankers' pay around that distinction.
I no longer believe that the length of time a deposit has sat with a bank says much about whether it will stay. For twenty-five years I accepted a version of that idea, and in March 2023 it failed in front of me. My position now is that a deposit is core only when the client's business runs through it: payroll goes out of it, receivables arrive in it, card sales settle into it. Everything else, however old, is a balance waiting for a reason to move, and a bank should invest, price and pay its bankers accordingly.
The distinction itself is not new. Liquidity rules for the largest banks have long treated operational deposits differently from other wholesale funding, and most asset-liability teams segment their deposits in some fashion. My argument is that the distinction belongs in strategy and not only in the liquidity model, and that most banks, mine included before 2023, used it in one place and ignored it everywhere else.
What tenure told us in March 2023
In the second week of March 2023, after the failure of several US banks whose uninsured depositors had withdrawn funds at a speed nobody had planned for, Varrowfield lost US$2.9 billion of deposits in eight business days, about 7% of the total. We met the outflow from cash and maturing securities. Afterwards my team examined every account that had shrunk by more than half.
The result was uncomfortable. About 60% of the money that left came from commercial and private-banking clients who had been with us for more than seven years. Our model had treated them as our most stable funding, because age was the principal input to the assumed life of a balance. The clients who stayed, including many younger relationships, shared a different feature. Their payroll or payables ran through our systems, or their own customers paid them through lockboxes and merchant accounts we operated. Moving those balances would have meant changing instructions to employees, suppliers and customers, and a corporate treasurer does not do that in a week on the strength of a rumour.
Age had been standing in for something else. In the decade of low rates before 2022 almost nothing gave a client a reason to move idle money, so old balances looked loyal. When rates rose and then confidence fell, the stand-in failed. The clients had not changed their minds about us. They had simply never been tied to us.
Where the deposit assumption hides in the securities book
The assumed life of a deposit is not an accounting curiosity. It decides how long a bank can afford to lend and invest. If a bank believes its checking and savings balances will stay for seven years on average, it can hold seven-year assets against them and call the book matched. In 2020 and 2021 that belief, applied to an extraordinary surge of deposits, encouraged many banks to buy long mortgage securities at the lowest yields in their history. When rates rose the securities lost value, and when deposits proved shorter than assumed, the losses stopped being theoretical.
We were partly fortunate and partly deliberate. In late 2021 my team compared each client's balance with its average before 2020 and found US$5.1 billion of the increase sitting in accounts holding more than twice their usual balance, with no new service attached. We invested US$2.5 billion of a proposed US$6 billion, in securities maturing within three years. That choice cost about US$34 million of 2022 income, and our treasurer, who argued against it, was right that our models had understated stability for years. In March 2023 it meant we had cash.
The lesson I draw is narrower than hindsight suggests. Nobody, me included, forecast the speed of the 2022 rate rises. What we did was decline to treat a surge as core because it happened to sit in old accounts. Our asset-liability model now gives each deposit segment its own life and rate sensitivity, set by the services attached rather than by tenure, and the securities portfolio is sized against the shortest credible life of the deposits that fund it. The board's risk committee sees that segmentation every quarter, beside the change in our economic value of equity under rate shocks.
Pay the bankers for the operating account
The hardest part was not the model. It was changing what relationship managers were paid to gather. Until mid-2023 our funds-transfer pricing credited a business line for every deposit dollar at a rate reflecting a long assumed life, whether or not the dollar did any work for the client. A banker who persuaded a property-management firm to park US$40 million with us on a generous earnings credit was rewarded more than one who moved a manufacturer's payroll and collections onto our systems, which takes months of implementation and produces smaller balances.
We changed the credit so that operational balances carry a higher transfer value and non-operational balances a lower one, and we put the classification into relationship managers' scorecards. We invested US$48 million over three years in treasury-management products and in the implementation teams that move a client's payments onto our platform. We also left US$2.2 billion of escrow and title deposits concentrated in twenty-seven firms whose balances were priced against market rates and ran through nothing else.
The cost is visible. Replacing those balances added about US$31 million of interest expense over 2023 and 2024. Our deposit growth has lagged peers that kept paying for balances. Two of our commercial teams spent months defending clients we had decided to leave, because I let the exit announcement run ahead of the scorecard change. Against that, uninsured deposits have fallen from 46% of the total to 33%, the share of commercial balances attached to a treasury-management service has risen from 59% to 78%, and the fees from those services have grown by about 40%.
There are fair objections. Some non-operational clients are valuable for other reasons, such as wealth management or lending, and a blanket rule would drive them away; we judge those case by case. Operational deposits can also leave, more slowly, if the bank's service fails or its reputation suffers badly enough, so the classification describes speed rather than permanence. And a smaller bank may lack the treasury-management capability to compete for operating accounts at all, in which case its strategy has to rest more heavily on insured household deposits.
Before approving its next strategic plan, I would ask any bank's board to look at three numbers together: the share of deposits above the insured limit, the share of commercial balances a client could move only by changing payment instructions, and the assumed life the interest-rate model gives each segment. If those three numbers do not tell the same story, the plan is relying on a loyalty nobody has tested.