Silicon Valley Bank 2023 collapse — case study

How RiskScan would have surfaced the warning signs by September 2022


TL;DR

Month Public signal RiskScan flag
Mar 2022 HTM ("held-to-maturity") securities pass $90bn (≈ 57% of total assets). HTMConcentration (⚠️ Medium)
Jun 2022 Unrealised HTM losses > $1.3bn. UnrealisedLosses (⚠️ Medium)
Sep 2022 Unrealised losses ~ $15bn vs equity ~ $16bn (paper equity nearly wiped). Composite high: UnrealisedLosses + HTMConcentration
Dec 2022 Moody’s warns of downgrade. (informational)
Mar 2023 48‑hour deposit run; FDIC takeover.

RiskScan’s composite rules would have escalated SVB to High by Sep 2022, giving PMs ~six months to de‑risk.


1) Signals over time

timeline title SVB public signals (2022→2023) Mar 2022 : HTM > $90bn (~57% assets) Jun 2022 : Unrealised HTM losses > $1.3bn Sep 2022 : Unrealised losses ~ $15bn; equity ~ $16bn (≈94% eroded on paper) Dec 2022 : Moody's places SVB on review for downgrade Mar 2023 : 48h deposit run; FDIC takeover

Source: SVB 10‑Q filings (2022 Q1–Q3), rating‑agency releases, FDIC.


2) Why the balance‑sheet structure was fragile

graph TD A[Rate hikes 2022] --> B[AFS/HTM bond prices fall] B --> C["Large unrealised losses (HTM)"] C --> D[Equity coverage thins] D --> E[Depositors grow nervous] E --> F[Outflows accelerate] F --> G[Forced AFS sales -> realize losses] G --> H[Solvency & liquidity spiral] H --> I[FDIC receivership]

The key issue was the mismatch: long‑duration HTM fixed‑income vs. flight‑prone deposits. Unrealised HTM losses don’t hit P&L, but they do hit economic value and equity buffers.


3) Risk rules (current implementation)

These checks are implemented today for regional banks:

Severity model: if ≥ 2 flags fire, severity = high; otherwise medium.

Cross‑sector text signal: GoingConcern is raised by a detector over filing text (regex first, LLM fallback), wired in ETL.

We do not emit an EquityWipeoutRisk flag today; the composite of UnrealisedLosses + HTMConcentration ( and often UninsuredDeposits) escalates severity to High and captures the same outcome.


4) End‑to‑end flow (daily)

sequenceDiagram autonumber participant ETL as "ETL (EDGAR, ratings, market)" participant Engine as "Risk Engine" participant DB as "Postgres" participant API as "API" participant UI as "Dashboard" ETL->>DB: Load 10-Q tables (AFS/HTM), ratings, metadata Engine->>DB: Compute flags per issuer (rules above) API->>UI: /flags?severity=High UI->>User: Highlight issuers with High severity (e.g., SVB in Oct 2022)

5) Composite flag snapshot

Flag Fired on Severity
HTMConcentration 2022‑03‑31 ⚠️ Medium
UnrealisedLosses 2022‑06‑30 ⚠️ Medium
GoingConcern 2022‑09‑30 🚨 High

References

  1. SVB Financial Group Forms 10‑Q (Q1–Q3 2022), esp. Note 3 (AFS/HTM).
  2. FDIC press release, Mar 10, 2023.
  3. Moody’s Investor Service, “SVB on Review for Downgrade”, 21 Dec 2022.

Numbers are rounded and indicative; exact figures vary by filing line‑item.