Model methodsSystemic riskactive

Long-Run Marginal Expected Shortfall

An institution's expected fractional equity decline conditional on the fixed 40% multi-period system decline used by the engine.

Registry slug:: lrmes
Visible surface:: /systemic

Data and implementation

Vintage

45,617 observations for 28 entities from 2020-01-03 through 2026-07-06, measured 2026-08-30

Data tables

data/parquet/systemic_series.parquet

Engine

engine/finweave_engine/layers/systemic/srisk.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 and the KBE system proxy, the producer fits GARCH(1,1) conditional volatility paths and a DCC(1,1) conditional-correlation path over the full aligned return history. It evaluates the engine's LRMES approximation at each filtered date with C equal to -0.40.

LRMES is a fractional conditional equity-loss estimate and is published. It is not the separate dollar capital-shortfall result often called SRISK.

The dollar result is not produced. Consolidated holding-company book debt and shares outstanding exist, but price levels do not, and the owner's price-source and licence decision remains unresolved.

Methodology evidence

Path and lines	Evidence
engine/finweave_engine/layers/systemic/srisk.py:151-185	Defines the stored LRMES approximation and the distinct dollar capital-shortfall formula.
docs/systemic_methodology.md:148-194	Documents the GARCH-DCC filtered path, LRMES interpretation, and why the dollar output is not produced.
docs/CODEBASE_LEDGER.md:1683-1709	Audits the remaining input gap: book debt and shares exist, while price levels and the source-licence decision remain unresolved.

Equations

Conditional betabeta = rho_t * sigma_bank_t / sigma_system_t

LRMESLRMES = 1 - exp(log(1 + C) * beta)

Dollar capital shortfall, not producedSRISK = max(0, k * (book_debt + market_cap) - (1 - LRMES) * market_cap)

Validation

The parquet contains 45,617 LRMES observations for 28 entities from 2020-01-03 through 2026-07-06, and stores them with method garch_dcc_filter.
The code audit confirms that the published series exercises the GARCH-DCC path. The methodology reports its narrow time-series range and its economically ordered cross-section rather than claiming a crisis-level spike.

Limitations

LRMES conditions on a fixed 40% system decline. It is a conditional vulnerability estimate, not the current probability of that event and not a failure probability.
The filter uses full-sample GARCH-DCC parameters, so historical conditional paths are descriptive and not point-in-time forecasts.
No dollar capital-shortfall parquet exists. Price levels are absent, and publication awaits the owner's price-source and licence decision; book debt and shares are not the missing inputs.
The underlying Yahoo-derived equity-return input carries an internal-use, display-aggregates-only licence posture and is not a redistributable raw dataset.

References

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/lrmes