Model methodsSystemic riskactive

CoVaR and Delta-CoVaR

The fitted system tail return conditional on an institution's tail return, and its change from the institution's median-return state.

Registry slug:: covar
Visible surface:: /systemic

Data and implementation

Vintage

CoVaR and Delta-CoVaR each contain 38,561 observations for 28 entities from 2021-01-04 through 2026-07-06, measured 2026-08-30

Data tables

data/parquet/systemic_series.parquet

Engine

engine/finweave_engine/layers/systemic/covar.py

Producer

scripts/build_systemic.py

Outputs

data/parquet/systemic_series.parquet

Source: Yahoo Finance equity returns via the argus collector, internal-use input

Method

For each institution, the engine runs a 5% quantile regression of system returns on institution returns over a trailing 252-trading-day window. It evaluates the fitted tail regression when the institution is at its own 5% return quantile to obtain CoVaR.

The corrected engine evaluates that same tail regression at the institution's median return. Delta-CoVaR is the difference between the two conditional tail estimates, following equation 10 in the cited paper.

The 2026-09-05 replacement uses the same-tail baseline, a tighter iterative stopping criterion, and a certified linear-programming fallback for non-convergent fits. The page and analyst reject archived or mixed method generations.

Methodology evidence

Path and lines	Evidence
engine/finweave_engine/layers/systemic/covar.py:38-95	Evaluates the same tail quantile regression at distressed and median institution returns.
docs/systemic_methodology.md:117-153	Defines the stored measures, rolling window, tail quantile, method label, span, and sign interpretation.
docs/systemic_methodology.md:292-382	Records historical crisis anchors, independent MES/absorption recomputation, and the validated Delta-CoVaR correction.

Equations

CoVaRcovar = params[0] + params[1] * bank_var_q

Median-state CoVaRcovar_median = params[0] + params[1] * bank_median

Delta-CoVaRdelta_covar = covar - covar_median

Validation

The parquet contains 38,561 CoVaR observations and 38,561 Delta-CoVaR observations, each covering 28 entities from 2021-01-04 through 2026-07-06.
The corrected generation preserved all row keys and other measures. Independent checks covered 110 windows, including all 54 certified fallback fits and the first/latest observation for every bank, with no failures at the prespecified numerical tolerances.

Limitations

The historical median-regression baseline was wrong. Its archived generation is retained for audit, while the displayed replacement uses the corrected same-tail definition. Numerical validation does not establish predictive skill.
CoVaR is a conditional return-tail estimate from one market proxy and one institution return series. It is not a failure probability or a causal estimate.
The 252-day rolling window means the series begins only after a full trading-year history is available.
The underlying Yahoo-derived equity-return input carries an internal-use, display-aggregates-only licence posture and is not a redistributable raw dataset.

References

Adrian and Brunnermeier, CoVaR, Federal Reserve Bank of New York Staff Report 348, revised September 2014, equation 10.

Metadata endpoint

Path	Method	Returns	Example
/api/methods/[slug]	GET	Registry metadata, implementation paths, measured vintage, methodology evidence, validation, and limitations. No model observations or parquet contents.	/api/methods/covar