Steven Kelly is associate director of research at the Yale Program on Financial Stability and publishes research notes at his Substack, Without Warning.

Silicon Valley Bank’s former CEO Greg Becker went before Congress last May to tell the SVB story. When he got to the section about the run on SVB, it was clear from the title where he laid blame:

From Gregory Becker written testimony to the US House Financial Services Committee

Yet, he appeared to have a rock-solid citation: the Federal Reserve’s postmortem on SVB. The Fed report on SVB mentions social media three times, citing nothing.

This is not to suggest the Fed needed reams of data and robust regressions. (The academic literature, while publishing some interesting things around the edges of this topic, also hasn’t demonstrated the “social-media-fuelled run” that is often stated as received wisdom.)*

But the Fed has a whole operation in New York designed to gather market colour, and its publications often cite this channel as having provided otherwise uncited market intelligence. Similarly, the Fed publishes its so-called Beige Book eight times a year; this report publishes informal market colour from all 12 Fed districts from “key business contacts, economists, market experts, and other sources.” Yet, the Fed gave no mention to even this less-structured type of intelligence gathering.

But the Fed isn’t alone: Virtually every regulator and inspector general report mentions social media, and how it conspired with modern-day banking apps and the prevalence of uninsured deposits, to manifest the fastest bank run in history. Banking officials’ speeches, when citing anything about social media, typically cite those reports — robust-sounding headlines, where the substance of the story was just that social media swirled at the same time the banks failed. 

But of course it did! These runs were a big deal! But just because we can see this social media communication, doesn’t mean Twitter is the cornerstone of modern bank runs.

The FDIC’s postmortem on First Republic lays down the increasingly establishment view matter-of-factly [my emphasis]: 

In retrospect, it does not appear that banks or banking regulators had sufficient appreciation for the risks that large concentrations of uninsured deposits could present in a social media-fuelled liquidity event.

We did have the fastest bank runs in history. Social media did swirl. Digital banking apps are more common than ever. It’s not clear there’s much overlap in this Venn diagram. As we look to reform banking regulation and supervision to be more resilient to fast bank runs (Acting Comptroller of the Currency Michael Hsu offered his opening salvo last month), it’s important we get their diagnosis right.

The ECB has told some lenders to monitor their social media sentiment for any signs of deterioration. To the extent this takes oxygen from focusing on the good ol’ fashioned balance sheet, it’s likely putting the Twitter cart before the actual-mechanics-of-banking horse.

Corporations are people, and people run on banks

Despite plenty of bank failures, there were no banking panics in the US between the Great Depression and 2007. Deposit insurance had killed the retail run — but was no match for the growth of institutional cash pools ahead of the Global Financial Crisis (GFC). Modern bank runs are institutional; think about the GFC (and inklings of a financial crisis at the outset of the pandemic): repo, prime brokerage, commercial paper, etc. This is institutional money. Uninsured deposits are deposit balances (cash only!) over $250,000 — Ie, institutional balances (and those of institution-affiliated principals).

Those on the list below, courtesy of Bloomberg News, were the biggest depositors of SVB:

FDIC document © Bloomberg

No sign of Mom and Pop. At end-2021, four depositors accounted for 14 per cent of Signature Bank’s assets. Forget social media; that’s a below-average-sized group text. 

Indeed, another story from Bloomberg News — which reveals the closest thing we have to a first-hand recounting of events from those who actually ran on SVB — quite explicitly says the genesis of the run on SVB was private communications among a networked group of sophisticated investors, not Twitter:

Channels like messaging platform WhatsApp, email chains, texts and other closed forums were full of chatter over the bank’s financial precarity well before those fears showed up Twitter. In tech, where executives’ networks can dictate whether companies have access to the best information, warnings about SVB had been simmering for a while when they boiled over into wider view online. […] 

By the time most people figured out that a bank run was a possibility on Thursday, March 9, it was already well underway.

If “other closed forums” included Bloomberg Terminal chats, this would be indistinguishable from a “normal” run on Wall Street. Everyone on Wall Street knows when a large hedge fund pulls funds or stops trading with a specific counterparty. Nothing new under the California sun.

Twitter and digital apps are retail phenomena. Corporate treasury departments already read the financial press, and they aren’t pulling hundreds of millions via digital app (which is often not even possible, nor is that likely even materially faster than calling their banker).

Moreover, the point of a bank run is to be first, not to have backup. Your incentive as a runner — whether retail or institutional — is not to broadcast it on Twitter. As the Wall Street Journal reported on SVB: “Some [VCs] debated if they should wait to warn startups to buy themselves more time to move their own, much bigger, balances.” Similarly, Bloomberg: “In many cases, investors moved to warn companies in private, perhaps seeking to both safeguard their investments and avoid a larger bank run.”

Broadcasting your running was unavoidable in the olden days: you were either very obviously in line at the bank, or you weren’t. Yes, someone who’s already “out” or who has nothing at risk can go on Twitter and shout to run, but famous short sellers do this kind of shouting all the time and often fail. If this “works,” it likely means the run on the bank is already in progress. And again, tweeting likely offers much less on the institutional side — where modern runs happen — than would a note to clients or a quote on the business wires.

Occam’s Razor

There’s no denying that the final stage of the bank runs we witnessed in 2023 were the fastest in history. But a preponderance of real-time tweets about a historical event is not preponderance of evidence of Twitter causality. The “the internet caused the bank run” narrative lacks material evidence and structural consistency with the actual dynamics of the bank runs. Absent new evidence, we risk skipping over the simpler explanation for the unprecedented speed of the run on SVB, which is just that, as far as banking crises go, the vulnerabilities of the banking system at the time were fairly straightforward

The market was keenly aware — from public filings, the financial press, and uhh social media — that SVB’s unrealised losses ate through all the bank’s capital. The market wasn’t ignoring banks’ unrealised losses, but rather was discounting them. Remember, unrealised losses are calculated by assuming the assets are funded at Treasury rates. But banks fund via deposits, which price much more cheaply. SVB did not have negative mark-to-market equity if its assets were marked to the deposit yield curve. There are thus two ways for a bank to realise its unrealised losses: 1) losing deposits, or 2) suddenly having to pay market rates on funding (this at least terms out the losses over time; this is what the Fed’s Bank Term Funding Program was doing if used to replace lost deposits).

So, on March 8, when SVB said its deposits were being drained faster than expected as its VC depositors burnt through cash — there was little left to figure out. Meanwhile, other banks with same unrealised loss vulnerability have been left standing by the market; they’ve lacked (so far) what SVB revealed in that March 8 emergency 8K filing. 

This wasn’t even a typical crisis where some assets had gone bad but retained some unknown value. SVB had told the world for months that its equity was negative on a mark-to-market basis. And, come March, the tech/VC economy’s downturn was forcing that mark-to-market into reality.

No one would argue that the banks that failed in 2023 were random or “wrong”. It’s not as if social media hearsay turned into hysteria that took down JPMorgan.

Screen time

The FDIC has said it’s looking at new tools to monitor social media, and that’s all fine. (And maybe they could offer the SEC some pointers.) While there was nothing to see on social that wasn’t already well-covered in the press until the run was already underway, surely there’s a benefit in this post-GameStop world to improving regulators’ internet presence. Contagion is important, and there’s no reason it shouldn’t happen in various ways on social media as it does elsewhere.

But before we assign the bank supervisor army to spend their time doomscrolling, and before we throw out the post-GFC rulebook and make banks be able to fully liquidate in 30 minutes or less, let’s get the first-order macro and balance sheet analyses right. Just because we can watch modern bank crises on Twitter, doesn’t mean that’s where they happen.

Further reading
— The Fed gets ratioed, bank capital edition (FTAV)

* The academic paper most frequently making its way into these conversations is one showing that Twitter activity didn’t just follow SVB’s stock price in its final day, but actually contributed to its plunge. Moreover, it shows that for other banks, pre-existing social media pervasiveness meant a steeper stock dip once the SVB run began. This is important, no doubt! And is similar to the lesson we learned with GameStop. But the spark for the run was SVB’s emergency 8K filing on March 8. After that, the stock price nosedived and took deposits with it. That other banks relatively pervasive on social media also saw more of a selloff is interesting to a degree, but this was already a run on the banks of the (economically wounded) innovation sector — so the greater social media exposure might be expected. Sure, we have the benefit of hindsight, and there’s a counterfactual story to tell (down goes Schwab!) — but again, it’s not as if Twitter spread the SVB contagion to Wall Street.

There’s another interesting paper on the slower moving “bank walks” in response to changing interest rates and digital apps’ role. This is a wholly separate issue from the issue of bank runs: The ease of a financial super-app may be what prompts you to chase down an extra 50 basis in points, but the clunkiness or lack of an app is unlikely to be that much of a deterrent when you perceive your bank and deposits to be at risk. Moreover, as emphasised throughout the note, the relative role of retail depositors in bank runs is small versus institutional ones. Also worth noting: the convenience of bank apps is also a force for keeping down deposit betas; preferred app habitats can keep you from switching banks at every basis point of rate hikes.



Leave a Reply

Your email address will not be published. Required fields are marked *