Bank / Financial Stress: validation report
The full evidence profile for this index. For how each test works, see the methodology.
| Matched stress gauge | Correlation | vs equity VIX | Timing |
|---|---|---|---|
| St. Louis Fed Financial Stress Index | 0.30 | 0.03 | Coincident |
Lower than the other two because bank stress is episodic (it spikes in crises) while the benchmark is a continuous spread-based index. It is essentially independent of the equity VIX (0.03) - it captures bank-specific stress, not general market fear.
77% of major events caught, about 9 days after onset. The highest event recall of the three indexes - the purest crisis theme. The March 2023 cluster spiked hardest (z above 20).
| Date | Event | Flagged within |
|---|---|---|
| 2023-03-09 | Silicon Valley Bank run and collapse | 14 days |
| 2023-03-12 | Signature Bank seized by regulators | 11 days |
| 2023-03-19 | UBS state-brokered rescue of Credit Suisse | 4 days |
| 2023-04-24 | First Republic collapse and JPMorgan seizure | 11 days |
Examples of caught events; not the full 91-event calendar. Misses are events that never drew a large enough coverage surge, not events the index read wrong.
+0.9 SD above normal. Spike days score about 0.9 standard deviations above normal on text novelty - the index rises on new information, not repetition.
No, by design. It does not predict bank-stock prices, volatility, or returns - coincident by construction, confirmed across horizons and model classes.
Reads bank-sector conditions, does not lead them. The index tracks bank-sector conditions - deposits, bank credit, financial conditions - in real time at the regime level. But it does not lead them: an apparent stress-to-credit lead turned out to be an artifact of the 2023 banking crisis and disappears once that single episode is excluded. So it is a real-time read on where bank stress stands, not an early-warning signal for credit.
Descriptive validation - measured, not asserted; not investment advice. The shared method for every index is on the methodology page.