Market Autopsy: how Silicon Valley Bank died in 44 hours

This series reconstructs failures in the order they happened, using what was known at each step. The temptation with any collapse is to narrate it as though the ending were obvious. It rarely was — and pretending otherwise teaches nothing.

Silicon Valley Bank was, until the week it closed, an unremarkable institution in the way that matters here: it was not committing fraud, it had not made reckless loans, and its regulators were not raising public alarms. It failed anyway, in under two days.

The setup

SVB banked technology start-ups and the venture funds that financed them. During the funding boom of 2020 and 2021, those clients raised enormous sums and deposited them. The bank's deposit base grew far faster than its lending opportunities.

So the bank did what banks do with excess deposits: it bought securities. Specifically, long-dated government-backed bonds — safe from a credit perspective, in the sense that they would be repaid in full at maturity.

Then interest rates rose at the fastest pace in four decades. The bank's bond portfolio fell in market value. At the same time, its depositors — start-ups no longer raising fresh capital — began spending their balances. The bank needed cash and its most liquid assets were worth less than it paid.

The sequence

The mistake that turned a problem into a failure

The securities sale on its own was survivable. Realising a $1.8bn loss to raise liquidity is painful, not fatal, for a bank of that size.

The failure was in the sequencing. The bank told the market about the loss and the need for new capital in the same announcement, without the capital raise already being secured. It handed depositors a complete argument — this bank has losses it cannot absorb and needs money it does not have — with nothing attached to resolve it.

What was knowable beforehand

This is the part where hindsight has to be handled carefully.

• Knowable: the composition of the securities portfolio, the unrealised losses on it, the deposit concentration and the uninsured deposit share. All disclosed in public filings.
• Knowable: that interest rates had risen sharply, and what that mechanically does to long-dated bond prices.
• Not knowable: that the bank would announce the sale and the capital raise together, in that order, on that day.
• Not knowable: that the withdrawal request would reach $42bn in a single session — a speed with no precedent to reason from.

Why it still matters

Duration mismatch is not a banking curiosity. It is the same structure that appears whenever a long-term commitment is funded with short-term money — in a business, a fund, or a household with a floating-rate obligation against a fixed income.

The question SVB leaves behind is not "was this bank badly run." It is: what do you own that is fine as long as you are not forced to sell it, and what would force you to sell it?