DeleverageSentinel

Institutional-grade early warning system for systemic deleveraging events.

DeleverageSentinel monitors credit spreads, tail stress, macro indicators, and cross-asset correlation clustering to detect early signs of synchronised market deleveraging — before it becomes front-page news.

Why This Exists

The 2008 financial crisis, the 2020 COVID crash, and every major deleveraging event share a common signature: assets that appeared uncorrelated suddenly move together. Diversification fails precisely when it's needed most.

This tool operationalises two key results from the Applied Probabilistic Systems research series:

• Axiom IV — Emergent Coupling: Correlation intensifies under stress. Independence is a calm-weather illusion.
• The Homogeneity Threshold: When cross-asset correlations spike, behavioural diversity collapses and systemic failure becomes likely.

The HOMOGENEITY_SPIKE signal — which detects correlation clustering across equities, volatility, credit, and the dollar — is the system's core innovation and carries the highest weight (32%).

Signals

Signal	Weight	What It Detects
HY_OAS_JUMP	18%	High-yield spreads widening fast (30 d)
TAIL_DIVERGENCE	18%	CCC−HY spread widening — worst credits deteriorating faster
CCC_OAS_HIGH	12%	Absolute CCC spread stress level
UK_APPROVALS_DOWN	12%	UK mortgage approvals declining (3 m)
HOMOGENEITY_SPIKE	32%	Cross-asset correlation clustering — diversity collapse
GBP_WEAK	8%	GBP/USD weakness as UK macro proxy

Regimes

Score	Regime	Meaning
0–24	🟢 GREEN	Normal conditions
25–49	🟡 AMBER	Elevated stress — monitor closely
50–74	🔴 RED	Significant stress — alerts fire
75–100	⚫ BLACK	Systemic deleveraging likely underway

Quick Start

# 1. Install
pip install requests pandas numpy pyyaml python-dateutil

# 2. Get free FRED API key: https://fred.stlouisfed.org/docs/api/api_key.html
export FRED_API_KEY="your_key_here"

# 3. Run
python deleverage_sentinel.py run --config config.yaml

# 4. Open out/index.html in your browser

Usage

# Live scoring
python deleverage_sentinel.py run --config config.yaml

# Backtest (2018–present)
python deleverage_sentinel.py backtest --config config.yaml \
    --start 2018-01-01 --end 2026-02-01

# Validate config without API calls
python deleverage_sentinel.py validate --config config.yaml

# Offline mode (local CSVs from data/ directory)
python deleverage_sentinel.py run --config config.yaml --offline

# Verbose logging
python deleverage_sentinel.py -v run --config config.yaml

Outputs

out/
├── latest.json    # Machine-readable snapshot (schema-versioned)
├── index.html     # Self-contained dashboard
└── history.csv    # Backtest time series

Host the out/ folder on Netlify, Vercel, or any static host for a live dashboard.

Architecture

Config (YAML)
    │
    ▼
Fetch ─────────────────────────────────────────┐
    │  FRED: HY OAS, CCC OAS, SP500, VIX, DXY │
    │  FRED: GBP/USD (replaces exchangerate.host)
    │  BoE:  UK mortgage approvals              │
    │  (or local CSVs in offline mode)          │
    ▼                                           │
Compute ◄──────────────────────────────────────┘
    │  6 independent signals
    │  Correlation clustering (homogeneity proxy)
    │  Severity scoring (0–1 per signal)
    ▼
Score
    │  Weighted composite (0–100)
    │  Regime classification (GREEN → BLACK)
    ▼
Output
    │  JSON + HTML dashboard + CSV
    │  Telegram / Email alerts (on RED/BLACK)
    ▼

Data Sources

All free. Only a FRED API key is required.

Source	Series	Frequency
FRED	ICE BofA HY OAS (BAMLH0A0HYM2)	Daily
FRED	ICE BofA CCC OAS (BAMLH0A3HYC)	Daily
FRED	S&P 500 (SP500)	Daily
FRED	VIX (VIXCLS)	Daily
FRED	Trade-Weighted USD (DTWEXBGS)	Weekly
FRED	GBP/USD (DEXUSUK)	Daily
BoE	UK Mortgage Approvals (LPMVYVA)	Monthly

Production Features

• Retry with exponential backoff — network blips don't kill the run
• Graceful degradation — non-critical feeds fail without crashing
• Data staleness detection — warns when data is > 5 days old
• Config validation — catches typos before API calls
• Custom exceptions — ConfigError vs FetchError for clear debugging
• Structured logging — to stderr, cron-safe, debug-level available
• Schema-versioned JSON — machine consumers detect format changes
• Backtest engine — historical evaluation with no lookahead bias
• O(log n) lookups — binary search replaces O(n) slicing in backtest

Alerts

alerting:
  alert_on_regimes: [RED, BLACK]
  telegram:
    enabled: true
    token_env: TELEGRAM_BOT_TOKEN
    chat_id_env: TELEGRAM_CHAT_ID

Offline Mode

For air-gapped environments, provide CSVs in data/:

date,value
2025-01-02,3.45
2025-01-03,3.52

Cron

0 22 * * 1-5 cd /path/to/sentinel && python deleverage_sentinel.py run --config config.yaml >> /var/log/sentinel.log 2>&1

Theoretical Foundation

Paper	Title	Key Insight
I	The Reflexive Stagnation Trap	AI epistemic convergence undermines innovation
II	The Homogeneity Threshold	Critical diversity level below which systems become fragile
III	The First Principles of Uncertainty	Five axioms + seven theorems for the science of risk

License

MIT

Author

Jason Gething — Founder, FishIntel Global

Not financial advice.