9 Comments

Can we please stop writing "SVB bankruptcy," when 1) SVB is not in bankruptcy and did not file for bankruptcy; 2) no bank has fled for bankruptcy; because 3) banks are statutorily ineligible for bankruptcy relief.

Expand full comment

Generative AI (Augmented Inference, if you please) can fill all comm channels with effluent.

Although I won't predict, I do see either salvation where generative AI is so undelightful that it prompts us to disallow it completely or at least punish it with gusto when discovered

OR

humanity decides to let the algorithms rule even if they're sloppy - much less thinking required.

Expand full comment

I prefer "bailout" for cases where those that caused a problem are rewarded for it [Jaime Diamon being to poster geezer for this phenomenon]. In SVB the stockholders and executives lost everything, "pour encourager les autres."

Expand full comment

Imagine you run Bank of America just before Covid. You own half a trillion $ of residential mortgages and MBS. Mortgage rates plummet to 3% and say, half of your mortgage asset refinance, so you have a quarter trillion $ to reinvest. What do you buy? T-bills yielding 0.1%? There isn't enough demand for all other loan types combined to suck up your cash. And there aren't any adjustable rate mortgages being originated either.

So you buy 30-yr MBS, T-notes, and T-bonds yielding 2%. And you are screwed 3 years later because the Fed went huge with QE far longer than needed, and then went big with rate hikes while being completely oblivious to the damage they were doing to the banking system. Meanwhile ... last month average weekly wages fell slightly, and CPI was 0.2% MoM if you exclude the bat-shit crazy shelter component.

Banks are not going to be lending the rest of the year. They are going to buy T-Bills yielding 5% because that's what investors and regulators want. That's a recipe for recession and deflation, not inflation. But sure, let's raise Fed Funds next week.

Expand full comment

Pace Davies, it is not obvious to me that SVB took egregious risks in its asset book, in Keynes' "prudent banker" sense. According to the FDIC (https://www.fdic.gov/news/speeches/2023/spfeb2823.html), the American banking system was sitting at about $620 billion of unrealized securities losses at the end of Q4 2022, down a bit from $690 the quarter before but still around a quarter of the total capital of the system. Most of these losses are IR losses, not credit: the same source shows that the duration of assets held by American banks has increased steadily since 2008. SVB itself thought it was still a going concern as of last Wednesday, and I don't think they were wrong. Sure, their asset book looks worse than average, but for every bank with better-than-average risk management, there must be equal and opposite worse-than-average risk management somewhere else.

And I would say that it is not obvious to the Fed or the Treasury either? If SVB were sui generis, or even extremely unusual, then these prophylactic measures wouldn't be necessary, would they?

Expand full comment

> And now we are on to number four: AI. AI is, in my view, four things:

I wouldn't discount

5. "An accelerator for hypothesis crafting and testing in systems too complex for bunches of apes communicating through journal papers and conferences"

AI-assisted drug discovery in its current form is a scam, but there are true advances in basic chemistry, and biology in general is so high-dimensional that there's a possibility we are in a relative quagmire of nearly blind trial and error because language/images/tables are bad at that type of complexity, and apes' ability to do active learning (i.e. optimal experiment design) in large dimensions sucks.

I mean, most "AI" sold _is_ high-dimensional regression and classification, but there's much going on in large-scale causal discovery and so on, and that has different sorts of implications. It's not quite a [potential] change on the scale of the development of the Royal Society, but getting to somewhere Moore's Law-like in, say, biotech would be a non-trivial to say the least.

Expand full comment