Overview
The Multi-Asset Scenario Stress Lab is a human-in-the-loop operational research system for 20-trading-day multi-asset stress analysis. It generates conditional joint return paths for eight liquid ETFs, organizes adverse scenarios into transparent stress archetypes, compares those scenarios with historical block-bootstrap resampling, and revalues portfolios on a common scenario cloud.
Evidence Position
The primary out-of-sample evaluation used 348 fixed forecast origins from 2019 through 2025. B1 EWMA-t remained the reference model. Flow F1 remained competitive and faster than the higher-step Flow reference, but the evidence did not establish predictive superiority for a modern generative model over B1.
System Contract
B1 EWMA-t
Uses eligible return history through the forecast origin to estimate a mean and exponentially weighted covariance matrix, then generates heavy-tailed multivariate Student-t innovations.
B0 block bootstrap
Resamples contiguous 20-day historical blocks without conditioning on the current state. It remains visible as the transparent reference.
Flow F1
Retained for research comparison only. No public operational warning, switch, blend, or confidence badge depends on Flow.
Selected Evidence
The retained Stage-D use asks whether the adverse scenario cloud contains a stable and externally relevant representation of the joint stress pattern that subsequently occurred. On 56 fixed large-drawdown origins, B1 placed more tail mass on the realized archetype and produced closer representative stress geometry than B0.
| Method | D1 share | D2 top-2 | D3 distance | Stability TVD | Status |
|---|---|---|---|---|---|
| B0 block bootstrap | 0.3661 | 0.9821 | 2.9171 | 0.0537 | Comparator |
| B1 EWMA-t | 0.6106 | 0.9821 | 1.8847 | 0.0468 | PASS |
| Flow F1 | 0.3420 | 0.9821 | 1.9148 | 0.0498 | Does not qualify |
Operating Model
Public Boundaries
- Generated stress-family shares within the B1 modeled tail
- B1 conditional stress structure vs B0 historical resampling
- Representative joint stress paths and geometry
- Model-implied tail/path diagnostics
- Human portfolio what-if comparisons on a fixed scenario cloud
- Calibrated crisis or archetype probabilities
- Market timing or crash prediction
- Causal macro/regime labels from archetype names
- Automated BIL de-risking or optimized portfolio weights
- Model-switching, blending, or predictive confidence traffic lights
Methods used in this system
These companion articles explain methods that are load-bearing for the system contract, validation evidence, and scenario-based decision support.
Out-of-Sample Forecast Evaluation
Explains the information boundary behind the paper's repeated pseudo-out-of-sample evaluation.
Look-Ahead Bias and Data Leakage
Explains how timing, data construction, and model-development choices can let future information enter a historical forecasting exercise.
Block Bootstrap
Explains dependence-aware resampling for time-series inference when IID resampling would break serial structure.
Multivariate Probabilistic Forecast Evaluation
Explains how multivariate probabilistic forecasts are evaluated with complementary proper scores, calibration, tail, dependence, and inference diagnostics rather than a single winner metric.
EWMA-t Scenario Simulation
Explains exponentially weighted covariance estimation, multivariate Student-t innovations, heavy-tailed joint path simulation, and the limits of a fixed conditional state over the scenario horizon.
Scenario-Based Portfolio Stress Testing
Explains portfolio revaluation on a common forward scenario cloud, path-dependent stress measures, VaR and Expected Shortfall, and same-cloud what-if comparisons without turning the exercise into optimization.
Stress Archetypes and Representative Scenario Geometry
Explains how adverse scenario tails can be organized into interpretable stress archetypes, represented by feasible scenarios, and assessed for stability, taxonomy adequacy, and external relevance.
Documentation
A public Stress Lab dashboard will be linked here when its public-safe interface is released. Until then, the white paper is the primary public document for the system architecture, validation evidence, and operating boundaries.