Banking stress (March 2023)

Definition

In March 2023 Silicon Valley Bank — the 16th largest US bank — failed within roughly 48 hours of announcing a securities-loss recapitalization, as ~$42B of deposits attempted to leave in one day; Signature Bank followed that weekend, and First Republic within weeks. The episode is the KB’s case study in the DIGITAL-ERA BANK RUN (deposit flight at smartphone speed, coordinated on social media) and in 2022’s rates shock arriving on bank balance sheets: the “risk-free” bonds banks held were credit-safe and price-devastated.

How it works / structure

• The balance-sheet mechanics (Barr review): SVB held long-duration Treasuries and agency MBS bought at 2020-21 yields; 2022’s rate rise (episode-rates-shock-2022) created tens of billions in unrealized losses, parked in held-to-maturity accounting where they didn’t hit capital ratios — until deposit outflows forced sales that REALIZED them (fa-sector-banks — the held-to-maturity/available-for-sale distinction became the sector’s most-watched line overnight).
• The run mechanics: a concentrated, networked, largely-uninsured depositor base (venture-backed companies, >90% above the insurance cap) moved as one coordinated crowd — the Barr review documents the outflow speed as historically unprecedented (sent-news-social attention mechanics applied to deposits; bias-herding where the run is individually rational by construction).
• The policy response: systemic-risk exceptions guaranteed uninsured deposits at the failed banks; the Bank Term Funding Program lent against securities AT PAR (directly neutralizing the mark-to-market hole) — containment inside two weeks, no broad credit crunch.
• Market fingerprints (engine-relevant): regional-bank index −25%+ in days while mega-banks gained deposits (dispersion INSIDE a sector — single-name analysis over sector beta); Treasury yields fell at record speed as rate-hike expectations unwound (event-fomc path repricing); options-implied moves on regional names hit crisis levels while the index barely repriced.

When it applies

Cited for bank-equity analysis (deposit composition, uninsured share, HTM marks — the post-2023 checklist in fa-sector-banks); for run-speed calibration (liability confidence evaporates in hours, not quarters); for duration-risk propagation (2022’s losses landing on leveraged holders in 2023); for sector-internal dispersion as the tradeable structure of a contained crisis.

Risk profile & failure modes

• The central lesson: interest-rate risk parked in accounting categories is still risk; the equity of a leveraged holder of underwater “safe” assets is a confidence instrument.
• Concentration as fragility: SVB’s depositor homogeneity (one industry, one network) made the run a single decision repeated — diversification logic applies to liability bases too.
• Speed recalibration: 2008’s runs took weeks; 2023’s took hours — monitoring cadences and position-exit assumptions inherited from prior crises are stale.
• Misuse: reading contained sector stress as systemic (the 2008 pattern-match failed — policy tools and the containment differed); and reading containment as certainty (First Republic still failed weeks later).

Evidence & limits

The Fed’s Barr review is the primary record (balance-sheet anatomy, run timeline, supervisory findings); FDIC and Treasury actions are public record. The counterfactual without BTFP is unknowable. Social-media amplification of the run is documented descriptively, its marginal contribution unmeasured.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Bank X’s uninsured-deposit share and HTM-loss-to-capital ratio both sit below the failed-2023 cohort’s medians (fragility screen)” — falsified by the filed numbers.
• “In the next regional-bank stress week, mega-bank deposits rise (flight-to-size thesis)” — falsified by the H.8/filing data.

Cross-references

• The sector toolkit it rewrote: fa-sector-banks
• The cause upstream: episode-rates-shock-2022, regime-rate-environments
• The systemic ancestor: episode-gfc-2008
• The run mechanics: sent-news-social, bias-herding