I think you're assuming here that the HTM accounting means the bonds don't actually lose value if they're held to maturity? That's not the case; it's just arbitrary accounting treatment, and the regulatory decision to permit such accounting is a big part of why the SVB blew up. Accounting rules are supposed to reflect economic reality to some extent, but they obviously don't do so exactly. For example, under FIFO inventory accounting, two identical boxes in the warehouse might be on the books at different values; but I assume you'd agree they're still actually worth the same amount, since they're identical.

All bonds get held to maturity by someone (unless they default, but that's not the problem here). The FMV of the bond is determined by the value of the bond's remaining cash flows to that person; so if the FMV went down, then that should be a clue that value was fundamentally lost, regardless of who's holding the bond.

Say you buy a theoretical 10 year zero-coupon bond with a 5% yield and a face value of $1,000. You should pay about $614 for it. You intend to hold it to maturity.

Interest rates take a random walk from now until maturity.

Under fair-value accounting, the balance sheet value starts at fair-value (obviously), then gyrates, but tends towards face value, and reaches it at maturity, due to time decay of bond premium. As you said yourself, every bond eventually matures in the absence of credit risks and fair-value can't indefinitely diverge from face value.

Under amortized cost basis accounting, the balance sheet value starts at fair-value but then increases every year until maturity, at which point it is also face value.

Surely you acknowledge that these are the same? They both describe the exact same cash flows.

Assuming I buy the bond today, the two accounting treatments agree (correctly) that in 2033, the bond is worth $1000 2033-dollars. The problem is that when interest rates increase, a 2033-dollar becomes worth relatively less than a 2023-dollar. So that agreement in future says nothing about the economic value of the bond today.

That's the loss that took down the SVB, and it's a real economic loss. If the SVB's depositors behaved like the simplest textbook model, then as soon as short-term Treasury rates increased, they'd insist on correspondingly increased interest on their deposits. The cost of that extra interest would be the SVB's loss--the cash flows in from the bonds stay the same, but the cash flows out to the depositors would need to increase.

In reality, I understand that depositors are generally "stickier" than that, leaving their funds at their existing banks even when bank interest rates increase slower than Treasury rates. That gave the SVB some hope that they could earn their way out of the hole, passing the loss slowly to their depositors over time by paying below-market interest rates. That's the behavior that HTM accounting roughly models. The SVB's depositors had no economically rational incentive to accept that though, and they didn't.

>The cost of that extra interest would be the SVB's loss--the cash flows in from the bonds stay the same, but the cash flows out to the depositors would need to increase.

... but SVB expected to have positive net-interest-income with increasing rates. Look here:

https://www.sec.gov/Archives/edgar/data/719739/0000719739230...

A 100bp increase in interest rates would've increased net-interest-income by 1.8%; a 200bp increase by 3.5%. This is typical for banks, because lending rates go up more than deposit rates as the base rate increases, increasing interest margin.

So the idea that this is about NPV of future cash flows is not right.

Why should economically rational depositors leave more money than they need for short-term working capital at the SVB at ~0% when a Treasury bill pays ~5% and carries less risk? Back when rates for everything were ~0%, a startup might lazily just put everything in the bank; but when interest rates increased, the reward for moving and the risk of staying both increased. I agree (and already noted above) that depositors are often irrationally sticky in practice; but nothing obligates them to be, and these ones weren't.

Of course most of the money left in a self-sustaining bank run, not directly for the reason above. There's a quite rational reason for someone to start that run, though. That reason wouldn't exist if the bank were well-capitalized on a MTM or NPV basis.

The SVB's expectation in that 10-K is a model of depositor behavior, predicting that depositors will leave their funds at a MTM-insolvent bank earning below-market interest for long enough for the SVB to earn its way out of the hole. That prediction obviously didn't come true. So doesn't that just mean their model was wrong, and shouldn't be relied upon elsewhere either?

Paying out from the FDIC insurance fund is inflationary, at the very least, which means everyone takes a bath.