I thought the majority of SVB's cash was tied up in 10-year MBSs, not treasury bonds?

Pretty sure that is the case

The kind of mortgage-backed securities they held are still US Government securities! They are “mortgage-backed” more-or-less in the sense that the Government is willing to offer such-and-such amount of investment opportunity at so-and-so rate of return because they are holding a bunch of mortgages that they expect will be able to pay for it. They are different to Treasury bonds in important ways so it’s imprecise for me to refer to all of it as Treasury bonds, I agree.

In practical terms, the reason SVB held MBSes was the same reason they held Treasury bonds (low risk because of Government guarantee), Treasury bonds and MBSes react to Federal Reserve rate increases in much the same way, and the ways that MBSes and Treasury bonds differ are largely walled off by the Government guarantee and thus immaterial to SVB (if the housing market does some wild stuff then MBSes might pay out differently to Treasury bonds, but this didn’t happen here).

you are doing a pretty good job trying to explain away the officers of this bank running it into the ground. you havent even started to try to excuse their lending practices. what are you trying to accomplish? their CRO committed career-ending blunders and even with only short term securities on their books this bank could run into serious trouble simply by dint of their customer base all running into trouble at the same time, as is happening. why do you need to posture here man-splaining this shit incorrectly to people who dont know any better

Because the officers of the bank are guilty of maybe 10% more garden variety bank greed than the average bank, while the real criminals here are the Silicon Valley CEOs who conspired to start a bank run and within 24 hours were demanding taxpayers fork over billions to clean up all their mess.

> …while the real criminals here are the Silicon Valley CEOs who conspired to start a bank run…

Yeah, way to victim blame.

Only an irrational person would “take one for the team” and leave their assets in an obviously failing bank.

I mean, if only so-and-so had kept their mouth shut nobody would have noticed that some cryptodudes and dudettes had blown a few measly billion of their customers’ funds.

I mean, I’m much more concerned about the “immediately demanding taxpayers make them whole” bit.

I think fwlr is doing a good job bringing actual data & detail to the conversation. They don't seem to be attempting to excuse or exonerate the bank officers, indeed they do call this behavior "dirty laundry". But they are fairly questioning whether the regulation would have actually prevented this scenario. If there is a subtext intended, it seems to be that the Fed didn't adhere to its previous assertions about how fast interest rates would rise, which is something that would undermine the stress tests.

> If there is a subtext intended, it seems to be that the Fed didn't adhere to its previous assertions about how fast interest rates would rise

This is extremely perceptive! I do feel that way, and you know what? I didn’t fully realize that I did until you pointed it out. Yeah, two major activities of the Federal Reserve are “saying what it will do with interest rates in the next few years” and “changing interest rates”, it’s really bizarre that they can’t get these two to even remotely line up. I know they have to respond to fluctuating macroeconomic situations, but is that really enough to almost completely disconnect “their predictions of their actions” from “their actions”? It feels like a bit like a fig leaf, it kinda doesn’t pass a smell check.

Thanks for your contributions. I've been pondering the same thing. You've saved me a lot of effort.

> Yeah, two major activities of the Federal Reserve are “saying what it will do with interest rates in the next few years” and “changing interest rates”, it’s really bizarre that they can’t get these two to even remotely line up.

True, but it wasn't just the Federal Reserve. I live in Australia. It was exactly the same story here: https://www.afr.com/markets/equity-markets/why-the-rba-and-e...

What I didn't realise is the Reserve Banks also operate in "hard mode" when it comes to economic predictions, just as the VC's did in this bank run:

https://www.abc.net.au/news/2023-03-12/imf-how-we-missed-the...

So you can put this down to SVB operated on the best available advice, butt the best available advice was wrong. Well that, and their customer base was a bunch of lemmings.

Dynamics of the trade will be the same though.

I think your explanation is missing one big thing: duration

It isn’t that treasury bonds are inherently “risky”. The risk is that when you hold longer duration bonds at low interest rates like SVB did, even modest rate increases will have a large effect on value.

Combine that with the bonehead investment in MBS which allegedly only yielded 1.5% at 10 year duration and you have massive losses.

But is that covered by Dodd-Frank? I think what the parent is getting at is that interest rate risk isn't factored in at all, only asset type (ie. whether it's "safe" stuff like treasury or it's "risky" stuff like derivatives). I skimmed https://en.wikipedia.org/wiki/Provisions_of_the_Dodd%E2%80%9... and can't find anything relating to interest rate risk.

* * *

The is the most "smoking gun" explanation that I've seen of how everyday corruption caused this. However, I'd love to see a follow that confirms the law resulted in a change in bond purchases by SVB.

I think that’s fair. I hope we do see that and other details of how this all went sideways come out over the next months and years.

I assume though, that like any black swan there were many things that went wrong together. Lately at work, I have been telling people that one in a million happens eight times per day at just 100 requests per second! Shit is going wrong in small ways all the time. So you design for that — but eventually in a big system enough of those minor failures will line up to cause something you notice. You probably remember that huge facebook outage in 2021? Lots of things went wrong and then they were locked out of their own conference rooms. The financial system is different parts at a different scale but the same theory of failure applies.

Not smoking guns, more like a bad bet here, and a stupid decision there, and one dumbass VC influencer gets jumpy and runs his dumbass mouth in slack and bobs your uncle bye bye SVB.

With all due respect to Dean and his fantastic work on the sounding the alarm before the GFC, I think he’s wrong here. The stress test doesn’t really consider sharp changes in Federal Reserve interest rates, it uses one predicted rate for the baseline scenario and a second predicted rate for the adverse scenario. With those rates it calculates the other variables, including bond rates. The highest treasury rate in either scenario in the last three years is like 2.5%, while the current rate right now is almost 4%, so I would argue the interest rate rises considered in the stress test are not quite capturing the pressure SVB was actually under. (There’s room to disagree; the baseline scenario for 2019 does predict 3.6% in 2022, so at least some years might have tested the right amount of stress.)

More fundamentally though, the stress test is testing capital, not liquidity. What actually did in SVB was a bank run during a liquidity crisis. Due to interest rate rises, their liquidity got so bad they had to sell long-term assets at a loss, and when they announced they were raising capital to cover that loss, they triggered a bank run: the ultimate and most cruel test of liquidity there is. The stress test they lobbied their way out of would have indicated poor-but-manageable health, it’s not useless, and they should have been required to do it. But the stress test does not involve simulating a bank run, it would not have predicted this collapse.

I agree that this is not black and white. However as an example, the Europeans are basically saying this situation cannot happen under Basel3 which is exactly what the D-F relaxing allowed SVB to avoid. One thing that will be telling is if another bank collapses under similar circumstances.

I don’t think the stress test is meant to model a collapse. As has been mentioned elsewhere, no bank survives outflows of 40-50% of deposits. I understand Baker’s comment to imply that stress tests are not simply “are you over this minimum bar” but also designed to probe for any weakness in lots of different scenarios so the weaknesses are discovered and can be addressed. I imagine it like the regulator is the parent telling the kid to eat their veggies — they might not force them into your mouth but they’re acting as a third party check on your worst impulses. If that’s the case, the stress test should have prompted a discussion about this risk about nine months ago. That this fell apart so fast implies that risk management failed, that those discussions never happened, and so there will presumably be motion to fix that weakness for next time. So you’re right that the run would still have killed them, but better long term risk management would have prevented the blood in the water that spooked the VCs that caused the run.

PS: I feel like that one meme from it’s always sunny — everything is connected to everything. Have to look past the first-order effects.

Yes, I like this “eating your veggies” metaphor.

For what it’s worth, the most plausible scenario I can see where SVB doesn’t collapse is something like “Their lobbying isn’t successful, they are required to undergo rigorous Dodd-Frank testing, because of this they are forced to get a risk officer, the risk officer armed with the kinda-worrying stress test results convinces SVB management to at least ratify a liquidity strategy even if they don’t really want to act on it, when they go to sell bonds and raise capital they present it as part of this liquidity strategy, the VCs see the strategy and aren’t as spooked, so the wave of withdrawals fizzles out instead of becoming a bank run - but inside SVB it gets way closer to a full-blown run than anyone outside realizes, and this scares management into executing on the liquidity strategy”. That’s an absurdly long chain of events and it’s very easy to step off that path at any point (or just get unlucky) so I still think it’s unlikely, but if I had to give an account of how SVB hypothetically survived, this is what I’d give.

While we can easily see that long-duration low-yield bonds are risky if interest rates rise, I would point out 2 things.

1. Financial analysts were not devoting much time to analyzing the risks from rapid rate increases, as rapid rate increases were deemed very unlikely (in part because the Federal Reserve itself had offered guidance that it was unlikely!). Furthermore, the time that was spent on analysis of this low-probability event was spread very thin (interest rate increases affect pretty much everything), and the perception of “zero risk” in T bonds would have bumped them to the end of the list for analysis, meaning even less analysis time for this outcome. Just now on Twitter I’ve seen two financial analyses praised for their prescience in this event - one from the FDIC, one from Byrne Hobart. Both of them only picked this up after the rate increases came into effect.

2. Even if there was full awareness of this risk, and plenty of forewarning that the rapid rate increase was definitely coming, the regulations as written would still permit these investments. The bank itself would probably have made smarter decisions and wouldn’t have collapsed, but that would be orthogonal to the regulations under discussion in the OP. (In my initial comment I even sketched out the argument that the regulations would have been mildly opposed to the smarter decisions.)

You're making a lot of very specific assertions in this thread about what bank officers or regulators would/should have done, or what the effect of SIFI stress tests would have been, without any actual evidence to back this up. Do you actually have inside information or prior experience on these things?

I have no inside information on SVB or their depositors, and no prior experience with managing a bank run. (I also have no exposure to SVB, nor do I have any stock market positions that may be affected by it.)

If you want my motivation for discussing it all day: I’m curious, it’s an interesting and fast-moving puzzle that a lot of smart people are discussing. If you want my qualifications for offering so many specific assertions: I’ve read the primary sources and it just doesn’t seem like it would fix or prevent this the way other people assume it would. If in my enthusiasm I’ve oversold myself as an expert, I apologize unreservedly.

At least they're the only person who bothered to look up the interest rates used in the stress tests (although citing sources would have been nice)

https://news.ycombinator.com/item?id=35119007 Sources here for DFAST 2022, 2021, 2020. 2019 is in the parent comment to that. You’re looking for Appendix A, Supervisory Scenarios, Tables A3 and A5 (Domestic Variables). Note that the actual interest rate used is not explicitly specified anywhere I could see, there’s a footnote that it was calculated by Federal Reserve staff according to Lars E. O. Svensson (1995), “Estimating Forward Interest Rates with the Extended Nelson-Siegel Method”. The 10 year treasury bond return is derived largely from just the interest rate, though, and it’s also the variable we actually care about - if they did give us the interest rate we would just be using it to calculate the Treasury rate anyway.

Notice the risk free rate is based on short term bonds. Those are not as sensitive to rate changes as they regularly reach maturity and you roll them over.

If SVB had owner those they would not have had liquidity problems which lead to insolvency.

Why is n earth did they buy 10 year bonds at historically low rates? I don’t understand why that seemed like a good idea.

Because the rate in long term bonds was 1.x% compared to 0% for the short term bonds.

It was a risky move but one that could have panned out if the deposits had remained.

It could only have panned out if the interest rate have remained. Once the interest rate increased, the deposits will chase higher rate, which SVB can’t provide and depositors will naturally withdraw their funds.

True but that doesn't happen all at once. Rates have been rising for more than a year now (not in a straight line but overall of course). They could have unwound this carry trade in March 2022 and probably come out ahead. Had they waited until summer 2022 they probably would have had a small loss. As it turns out though they held on for bigger losses.

Banks often do this, trade short term risk for long term and pocket the difference.

It seems SVB didn't account for the fact that there was a strong correlation between the source of their money (tech propped up by low rates) and their investment risk (bonds/loans propped up by low rates).

I'm honestly unsure if we can ascribe this to to incompetence or greed.

This is a form of the carry trade.

You're borrowing at a lower yield to lend at the higher yield, and the risk is that you have to liquidate before the long one is mature.

Traders do this all the time, because every day you carry this position you're making the difference in yields.

The risk is always the same: the value of the things that give you the yield may change by enough to wipe out the interest difference.

buy side traders blowing up is a lot different than a bank blowing up so the risks are not the same

Because rates were near 0 at the time for short term bonds. Interest rates were so low that there was nowhere else to park the money. Usually this is fixed with loans but with startups having lots of VC money they didn’t need loans.

> rates were so low that there was nowhere else to park the money

There were plenty of places to park cash. Just none whose yield SVB liked.

> Notice the risk free rate is based on short term bonds. Those are not as sensitive to rate changes as they regularly reach maturity and you roll them over.

That is not why they are less sensitive to interest rates. They are less sensitive to interest rates because the value of a bond (assuming no coupon) is (whatever you get paid at maturity)^((1+rate)^(-time to maturity)). The sensitivity to rates increases exponentially with the time!

If you buy 30-year bonds, hold them to maturity then roll them over, and don’t mark to market, you are just taking the graph of your returns and heavily smoothing it. And the smoothing obscures a fascinating thing about bonds: they’re sensitive to rates both ways. When rates go up, existing bonds lose present value immediately, but they start to gain present value at the new higher rate forever, or at least until the next rate change or until you sell them or let them mature without rolling them over.

If you want to do an honest HTM forecast of SVB’s position, you need to project everything out to the same future time, say 10 years. So you’re looking at SVB’s portfolio, with a reasonable expectation (with error bars or confidence intervals) of returns over 10 years, in 2033 dollars. In this model, the value of SVB’s 10-year bonds didn’t change, but the value of its liabilities exploded: it needs to pay a lot more interest on its interest-bearing deposits, and customers (I assume) are a lot more likely to demand interest when rates are high.

whoops, I can’t type. That’s (1+rate)*(-time to maturity) in the exponent. It’s exponential but not doubly exponential.

What about their exemption from Basel 3? https://www.ft.com/content/c95e7708-b903-405d-a017-963844eb3...

"Arguably, being exempt from Dodd-Frank may well have allowed SVB to be more resilient than otherwise."

Wow, that's wrong wrong wrong.

Dodd-Frank would have forced SVB to do and report tests including the vanilla interest rate curve sensitivity values.

Look for "10-year Treasury yield" in this document (it's about everywhere as a parameter):

https://www.federalreserve.gov/publications/files/2019-march...

Which would have alerted the authority that SVB management was heavily playing a dangerous game.

All fair and good, as a noob I ask what else should SVB have done if not invest in Treasury Bonds (which according to OP are lowest risk)?

> what else should SVB have done if not invest in Treasury Bonds

Bought Treasury bills. They’re more liquid and have less interest rate risk. They yield less, however, which is why SVB bought the riskier stuff.

That's the thing I don't get at all, they had what, something like 160b in deposits - they could've kept 150b in bills, ten billion in Treasury bonds, and still made millions to hundreds of millions per year ostensibly risk free - surely enough to keep a ledger balanced. The bet on mbs at historic lows seemed like a pennies in front of steam rollers play

It was. You're not missing anything.

Bob, responsible for risk management, goes to the boss and says "Hey here is the plan that makes us $X and keeps us at acceptable risk of interest rate changes." Jim, the boss says "Jeez $X isn't very much money, our shareholders expect returns! If we changed our portfolio like this we'd make $Y. You are being a debbie downer, bond prices won't collapse and even if they do we just hold them to maturity and we are fine. We simply can't leave this much money on the table."

Didn’t help the chief risk officer quit in April and a new person wasn’t hired until January. So there wasn’t even a Bob in this scenario.

SVB's Chief Administrative Officer, Joseph Gentile, was formerly the CFO for Lehman Brothers' Global Investment Bank.

Gee, it's almost like limiting the liability of these c-suite types let's them do whatever they want and get away with it. Weird.

DFAST is another part of Dodd-Frank, and one part that might have helped catch this issue, I agree - although, I do wonder what potential Federal Reserve interest rates they would have plugged into the calculations…

The 2021 Report https://www.federalreserve.gov/publications/files/2021-dfast... took the baseline scenario to 1.9% in 2024 for 10yr Treasury returns, while the adverse scenario went to 1.5%. Likewise for the previous year: in 2020, baseline took it to 2.7% in 2023 while the adverse scenario went to 2.2%. The 2022 Test variables are quoted at https://www.federalreserve.gov/publications/2022-Stress-Test... saying year 2025 would see 2-1/2 and 1-1/2, for baseline and adverse respectively. For comparison, the actual rate now is close to 4%. So, honestly, I’m not sure if DFAST would have caught it, since this is a full percentage point and a half above the highest values they used in the last three tests. 2019 had 3.6% in baseline for 2022, so maybe it would have, though.

But ignore that, let’s assume DFAST would have clearly highlighted this. If it did, it would be cautioning “you have the potential for significant unrealized losses, make sure you maintain a good position so you can carry those securities to maturity” - which SVB was doing right up until its dwindling liquidity was in reach of a big bank run. The stress test is simply not designed to detect all potential periods where a sudden bank run of significant scale might kill a bank - fundamentally, it’s a capital stress test, not a liquidity stress test.

At best I think we can say DFAST, in some years but not in others, could plausibly have given a moderate to strong indication that more of SVB’s assets would be tied up than usual. From that, you could certainly take a hint and look at liquidity.

But then you have to have enough paranoia to model “The Fed lights a rocket under the interest rate just like they said they wouldn’t” and “our depositors, who all walk around with their burn rates practically tattooed on their foreheads, decide to do a bank run” at the same time to see this coming. And then you have to tell other people this is coming, and grit your teeth listening to them quote the Fed’s forecast that interest rates will stay low. Ultimately they would calculate the cost of your proposal and ask you why they ought to give up that much potential profit to hedge against the chance of these two unlikely events happening together - “that’s like seeing a pair of black swans mating”, they would say to you.

Again, I want to stress I’m not saying Dodd-Frank caused this, I’m not saying Dodd-Frank exemptions for SVB were a good thing, etc. I’m saying that this collapse was mostly out of scope of Dodd-Frank, the parts that were in scope were affected in both directions, and all of this in service of the original point: the article bringing up their exemption lobbying is just airing dirty laundry, not elucidating the cause of the downfall.

If you're subjected to a regulation with periodic reporting, you're just not going to play and look bad on a report if you can avoid it. And you know parameters will be refreshed

SVB did lobby so it could play the game it wanted, it looks like it was a basic game from what I can read in the media.

In Europe the current test is page 5, 2023 10 year interest rate at 4.5% in the stress test, and you see that it's a moving target (nice historical graph):

https://www.eba.europa.eu/sites/default/documents/files/docu...

Incentives matter ... While regulation will not prevent failure, it helps. We already know what happens after deregulation of risk management in banking.

Disclaimer: worked for a decade in finance, including on software used for those report computations...

> (in fact, in researching this I discovered, horrifyingly, that it’s not uncommon for Treasury bonds to be described as literally risk-free)

Correct me if I'm wrong, they are practically risk-free if held to maturity no? It's only when you need them to be a liquid asset that you can sell for money in a pinch that it's not so risk free as the amount you can get for them depends on the market.

Risk free in the sense they’ll pay out.

But if you buy $1B in bonds at 1% and inflations runs 4-5% over the bond term, you’re running a -3-4% real return.

Well what do you mean by cash?

Keeping cash in a vault is of course highly impractical, and also doesn't produce any money for you or the bank. If that's what you want, I think you could find it but it would be at a security company (security as in guards), not a bank.

The electronic equivalent would be depositing the money with the federal reserve. (What they do with the physical bills I don't know! But most deposits would be coming in electronically these days anyways.) However, up until the passage of the Emergency Economic Stabilization Act of 2008, the fed didn't even pay interest on reserves. [1] Now they do, but until 2022 this rate was only 0.15%. [2]

The first problem with this would have been it didn't provide enough money to run the bank. SVB's non-interest expenses were over $2 billion annually for the last 3 years, so with $200 billion of deposits you need at least to earn 1%, just to keep the lights on. Also your deposits will go up and down but you'll still have to pay your rent, so better be more than 1%. (This is how they ended up invested in longer duration treasuries, to earn more money by taking on greater interest rate risk.)

But the more fundamental problem with just keeing the money at the fed is that I don't think they would let you. It is not, traditionally, the social-economic purpose of a bank to just collect deposits and do nothing with them. Instead, the bank is supposed to take in deposits and then provide loans to the community (credit cards, business loans, mortgages, etc.). Probably this purpose has gotten a little fuzzy over time, since banks are not holding on to the loans, but it is the basic idea.

There is a company calling itself "The Narrow Bank" that actually wants to do exactly this, deposit all client money at the fed. [3] Tellingly, the fed has not granted them approval yet to do so. Maybe they will! But notice that this proposed bank is bare bones: no physical branches, no tellers, no ATMs, no FDIC insurance, just a place to deposit vast amounts of money from institutional clients and earn money from the interest rate spread from the fed vs what you pay out. I do not think such a bank would be very popular with many "regular" people or businesses, who have needs beyond just stashing money under an electronic mattress.

[1] https://www.stlouisfed.org/open-vault/2018/april/why-fed-pay...

[2] https://fred.stlouisfed.org/series/IORB

[3] https://www.tnbusa.com/2022/04/tnb-seeks-to-become-states-ne...

Inflation. Keeping it in cash has a negative return.

There's one thing I don't understand.

If I'm buying low yield bonds for 100B, even if interest rates go much higher, shouldn't I still be able to sell the bonds for 100B?

There's still an obligation by the government to give me 100B back at the end of the loan.

Not necessarily. They’re worth 100B still, but they require the purchaser to “lock up” their capital. If you need liquid cash now and potential buyers are only interested in locking up their capital for higher gains/rewards, you might consider selling at a loss.

Yes, and if you hold it for the 10 year term, you do get the money back, plus a small amount of interest. The problem is that now, you can buy bonds with a shorter term that provide higher interest rates. Thus to sell the old bonds now, you have to sell them for less than you paid for them so that the return on them becomes competitive with buying a new bond.

Okay that is clear, thank you.

No, if you have to sell your bonds with, say, 9 years remaining, you won't get 100B for them. Think about it, you paid 100B to get 10-yr treasury bonds paying 2%/$4 billion, annually. Now someone can go to a treasury auction and pay 100B and get 4%/$8 billion annually. Why would they still pay you $100 B for yours? They are obviously worth less.