eminshall/spectral-Fed2024-severely-adverse-5000

Spectral Fed2024 Severely Adverse — 5000 Synthetic Market Paths

Overview

5,000 stochastic realizations of a severely adverse macroeconomic scenario, calibrated to the Federal Reserve's 2024 DFAST stress test scenario. Each path includes correlated daily asset price trajectories (252 trading days, 1 year) and quarterly macro variable paths, generated by the Spectral Data Engine.

This is not a forecast. It is a synthetic scenario dataset for use in:

• Portfolio stress testing and risk model validation
• Reinforcement learning environment training (compatible with SpectralTradingEnv)
• Backtesting strategy robustness under crisis conditions
• Model training where real crisis data is scarce or unavailable

---

Calibration Source

Federal Reserve Board, 2024 Supervisory Stress Test Scenarios (Feb. 2024) https://www.federalreserve.gov/publications/files/2024-stress-test-scenarios-20240215.pdf

The severely adverse scenario is characterized by a severe global recession with heightened stress in commercial real estate and corporate debt markets. Key published targets used for calibration:

Variable	Jump-off	Peak/Trough	Quarter
Equity price index	0%	-54%	Q3
VIX	~23%	72%	Q2
Unemployment rate	~4.5%	10.0%	Q7
3-month Treasury bill	~5%	~0.1%	Q2
Real GDP growth (cumulative)	0%	-4.8%	Q6

These targets were used to calibrate the skeleton path of each variable. Generated path distributions converge to these values across the full batch — verified in the accompanying validation notebook.

---

What's in this Dataset

Price paths

• Shape: (5000, 253, 5) — 5,000 paths × 253 timesteps × 5 assets
• Format: Parquet (price_paths.parquet)
• Assets: 5 synthetic multi-asset price series with correlated GBM, GARCH volatility clustering, and jump diffusion
• Timestep: Daily (1 trading year = 252 steps)
• Starting price: 100.0 (normalized)

Macro paths

• Shape: Long format — (path_id, quarter, variable, value)
• Format: Parquet (macro_paths.parquet)
• Variables: equity_price_index, vix, unemployment_rate, three_month_tbill, real_gdp_growth
• Timestep: Quarterly (13 quarters, ~3 years)
• Horizon: Matches the Fed's 9-quarter stress scenario horizon

Validation notebook

• validation_notebook.ipynb — reproduces all calibration checks including convergence against Fed targets, distributional sanity, co-movement correlation structure, and price path conditioning

---

Key Properties

Correlated macro variables. Variables move together within a path via a shared path-level severity factor plus a Cholesky-correlated per-quarter shock process. A "bad path" is bad across all variables simultaneously — equity crashes coincide with VIX spikes and unemployment rises. This is validated in the notebook (§5a, §5b, §5c).

Macro-conditioned price paths. Asset price paths are conditioned on the macro regime: drift and volatility modifiers are derived from the quarterly macro paths and interpolated to daily granularity. Stressed paths show materially negative mean returns and ~3.5x higher annualized volatility compared to unconditioned baseline paths.

Interpretable and auditable. This is a parametric, not a GAN-based generator. Every parameter is traceable to either the Fed's published scenario or an explicit documented assumption. The full generation code and validation notebook are included.

Realistic stylized facts. Asset returns exhibit volatility clustering (GARCH), fat tails (jump diffusion), and cross-asset correlation, not just geometric Brownian motion. These properties are validated against known stylized facts of financial return distributions.

---

v1 Scope and Limitations

This dataset covers 5 core macro variables from the Fed's 28-variable scenario. The following are intentionally deferred to a future release: BBB corporate spread, house price index, CRE price index, 10-year Treasury, mortgage rate, CPI inflation, international variables.

This scope is appropriate for market simulation and trading model training. If you need the full credit/real estate variable set for bank-style loss modeling, watch this space for v1.1.

The trough distribution of generated paths has an expected overshoot of approximately 10-12% relative to the Fed's published point targets, due to the order-statistics behavior of stochastic path minima. This is documented and quantified in the validation notebook. The mean path across all 5,000 realizations converges to within 1% of the Fed's published targets.

---

Usage

import pandas as pd
import numpy as np

# Load price paths
price_paths = pd.read_parquet("price_paths.parquet")
# or as numpy array if stored in wide format
# shape: (5000, 253, 5)

# Load macro paths
macro_paths = pd.read_parquet("macro_paths.parquet")

# Example: get equity index macro path for path 0
path0_equity = macro_paths[
    (macro_paths["path_id"] == 0) &
    (macro_paths["variable"] == "equity_price_index")
].sort_values("quarter")

# Example: compute mean return across all paths for asset 0
# (assuming price_paths has columns: path_id, timestep, asset_0..asset_4)

---

Generation

Generated by Spectral Data Engine — a parametric synthetic financial data engine built by Spectral Forge Labs.

The engine uses:

• Regime-switching stochastic processes for macro variables
• Skeleton-plus-noise architecture calibrated to Fed scenario targets
• Shared path-level severity factor for cross-variable co-movement
• Cholesky-correlated per-quarter shocks for within-quarter co-movement
• GARCH(1,1) volatility clustering for asset price paths
• Jump diffusion (Merton model) for fat-tailed return distributions
• Correlated GBM for cross-asset correlation structure

---

License

MIT. Free to use for research, commercial, and educational purposes. Attribution appreciated but not required.

---

Citation

If you use this dataset in research, please cite:

@dataset{spectral_fed2024_severely_adverse,
  author       = {Spectral Forge Labs},
  title        = {Spectral Fed2024 Severely Adverse — 5000 Synthetic Market Paths},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/eminshall/spectral-Fed2024-severely-adverse-5000},
  note         = {Calibrated to Federal Reserve 2024 DFAST Severely Adverse Scenario}
}

---

Related

• SpectralTradingEnv — RL trading environment compatible with these paths
• Spectral Forge Labs — synthetic financial data for research and simulation
• More scenario datasets (baseline, 2022 severely adverse, stagflation) coming soon

---

Disclaimer

This dataset is for research and simulation purposes only. It is not financial advice, not a market forecast, and not suitable for direct use in live trading systems without further validation. Past simulated performance does not indicate future real-world results. The Federal Reserve scenarios used for calibration are hypothetical stress scenarios, not forecasts.