SVB Didn't Die in 36 Hours. It Died Over Three Years of Exams Nobody Acted On.

On the afternoon of March 9, 2023, depositors asked Silicon Valley Bank to give them back $42 billion.² That is a quarter of the bank's money, gone in the hours between lunch and a late dinner, mostly by phone and login from venture capitalists telling their portfolio companies to wire out now. The next morning the bank was seized, before another $100 billion that had already queued up could leave.⁷ It is told as a freak event — a 36-hour run, a panic that came from nowhere, a digital stampede no one could have seen. But the run was the last 18 hours of a story the Federal Reserve had been writing in its own examination files for three straight years.

The official story is a liquidity shock: a sudden, unforeseeable bank run sank a healthy bank. The real story is a solvency problem dressed as a liquidity problem, on a balance sheet the regulator had flagged in 2020, in 2021, and again in 2022 — and never once forced to fix.⁵ The run was the proximate cause. The structural cause was a duration mismatch left to grow in plain sight.

How a bank for startups bet the house on long bonds

Here is the trap, and it is older than banking software. SVB took in an enormous flood of deposits during the 2021 startup boom and had to put the money somewhere. It chose bonds — and not short, nimble ones. Its securities book ballooned from $23 billion in 2018 to $125 billion in 2021, with the held-to-maturity portion alone growing from $15 billion to $98 billion, a more than 500% jump.⁴ By early 2022 that HTM book was roughly 46% of the bank's entire balance sheet.⁴ And it was long: by the end of 2022 the HTM portfolio carried a weighted-average duration of 6.2 years, stuffed mostly with agency mortgage bonds maturing in ten years or more.¹

Now hold that against the other side of the ledger. About 94% of SVB's deposits were uninsured.¹ That is the number that should have kept someone awake. An uninsured deposit is not loyal — it is a flight risk with a keyboard. SVB had funded a decade-long bet with money that could vanish in a morning. The bank borrowed short and lent long, which is what every bank does — but it did it with the most fickle funding in the country against the longest paper on the market, and it did it just before the Fed raised rates faster than at any time in forty years.

Rising rates don't actually hurt a held-to-maturity bond — on paper. You can carry it at cost and wait for it to mature. The loss is unrealized, parked in a footnote, invisible to net income and to regulatory capital. That is exactly why the duration mismatch felt safe right up until it wasn't. The losses become real only the instant you are forced to sell before maturity. And the one thing that forces a bank to sell is the one thing 94% uninsured deposits guarantee: a run.

The capital raise that lit the match

On March 8, 2023, SVB did the thing that turns an unrealized loss into a realized one and a quiet problem into a public one. It announced it had sold over $21 billion of securities — substantially all of its available-for-sale book — crystallizing a $1.8 billion loss, and that it would raise about $2.25 billion in fresh capital, including $500 million from General Atlantic.²³ In the bank's own telling this was prudent housekeeping: take the loss, reinvest into shorter, asset-sensitive paper, repair the balance sheet.³ To a depositor it read as a confession. A bank that has to sell its bonds at a loss and pass a hat for capital in the same press release has just told the most networked, group-chatted customer base in finance that it is short of money.

And then it moved at the speed of the people it banked. By percentage, this was the most severe bank run in recent history: a 25% deposit loss in one day, with another 62% scheduled to flow out the next — roughly 81% of $175 billion in deposits attempting to leave in 48 hours.⁸ A run that took weeks in 1929 took an afternoon in 2023, because every depositor was a fund manager, every fund manager had a group chat, and money now moves at the speed of a wire confirmation. But notice the sequence. The run did not invent the loss. The loss was already on the books. The run simply forced SVB to admit it all at once.

The regulator that saw it coming and watched it happen

This is where the comforting version of events falls apart. The Fed was not blindsided. Its own April 2023 review documents that examiners identified interest-rate-risk deficiencies in the 2020, 2021, and 2022 exams — yet supervisory findings weren't even issued until November 2022, and the planned downgrade of SVB's interest-rate-risk rating was never finalized before the bank failed.⁵ In mid-February 2023, weeks before the collapse, the Board of Governors was briefed that SVB posed 'significant safety and soundness risks.'⁶ The information was not missing. The escalation was. Three years of warnings produced not a single formal enforcement action before the doors closed.

The bank's own conduct made the supervisory failure cheaper to commit. Rather than hedge the risk of rising rates, SVB removed hedges; rather than reduce the duration mismatch, it changed the assumptions in its own models so the risk would measure smaller.⁶ A firm that manages the thermometer instead of the fever is a firm that needs a regulator with a spine. It had one with a clipboard.

Wasn't this just an unforeseeable run?

The honest counter is that the speed was genuinely unprecedented. No risk model in 2021 assumed a quarter of your deposit base could request withdrawal in a single afternoon; the digital, concentrated, herd-prone nature of SVB's clientele was a new kind of accelerant, and even a well-supervised bank might have struggled to survive 81% of deposits heading for the exit in two days. That is fair, and it is why 'liquidity shock' is not a lie — it is just the last chapter. But speed is not the same as surprise. A bank with insured, diversified, sticky funding does not face that run, because there is nothing to flee from. A bank with a short-duration book does not face that run, because the capital raise is never necessary. SVB's vulnerability to a fast run was not bad luck; it was the designed-in consequence of a 6.2-year asset book funded by 94% uninsured deposits — a structure the regulator had named, in writing, three years running. The run was fast. The fragility was old.

SVB became the second-largest bank failure in U.S. history — behind Washington Mutual's $307 billion in 2008, not the largest, whatever the early headlines said. But the size was never the interesting part. The interesting part is that nothing about its death required new information. The duration mismatch was in the filings. The 94% uninsured figure was in the filings. The interest-rate risk was in three years of exam reports sitting inside the Federal Reserve. The 36 hours everyone remembers were just the moment all of it stopped being a footnote and started being a fact. A bank doesn't fail when the depositors run. It fails when it builds a balance sheet that gives them a reason to.