Skip to content

QUICK_START.md

Latest commit

 

History

History
236 lines (170 loc) · 6.54 KB

QUICK_START.md

File metadata and controls

236 lines (170 loc) · 6.54 KB

Quick Start Guide

Rebonato-Denev Bayesian Networks for Black Swan Events

What You Have

A complete implementation of Riccardo Rebonato and Alexander Denev's Bayesian network methodology for modeling "black swan" financial events - specifically, the Eurozone breakup scenario.

Files at a Glance

File Purpose Size
rebonato_denev_eurozone_crisis.py Main implementation 22 KB
rebonato_denev_tutorial.ipynb Interactive notebook 11 KB
README.md Comprehensive documentation 8 KB
PROJECT_SUMMARY.md Executive summary 8 KB
eurozone_crisis_network.png Network structure 334 KB
scenario_comparison.png Results visualization 383 KB
causal_flow.png Probability flow diagram 303 KB
requirements.txt Dependencies <1 KB

Run It Now (3 Options)

Option 1: Quick Demo (2 minutes)

python rebonato_denev_eurozone_crisis.py

This runs the complete analysis and shows:

  • 4 stress scenarios
  • Portfolio implications
  • Monte Carlo simulation results

Option 2: Interactive Exploration (10 minutes)

jupyter notebook rebonato_denev_tutorial.ipynb

Step-by-step tutorial with:

  • Custom scenario analysis
  • Interactive visualizations
  • Portfolio allocation examples

Option 3: Just See the Pictures

Open these files:

  • eurozone_crisis_network.png - The causal structure
  • scenario_comparison.png - All results in one view
  • causal_flow.png - How probabilities propagate

Key Insights (TL;DR)

The Problem with Traditional Approaches

❌ Historical correlations break down during crises
❌ Can't model events that haven't happened
❌ No understanding of causality

The Rebonato-Denev Solution

✅ Explicit causal modeling
✅ Forward-looking scenarios
✅ Expert judgment + data
✅ Bayesian updating

What the Model Shows

Normal Times → 7% breakup risk, 22% equity decline
Black Swan (High politics + econ crisis) → 76% breakup risk, 62% equity decline
If Breakup Occurs → 88% flight to quality, 76% equity decline
Market Signals (Spreads widening + flight) → 96% inferred breakup risk

Portfolio Implications

Scenario Recommendation
Normal 60% equities, 30% corp bonds, 10% gov bonds
Elevated Risk 40% equities, 20% corp bonds, 40% gov bonds
High Stress 20% equities, 10% corp bonds, 70% gov bonds

The "Black Swan" Connection

What Makes It a Black Swan?

  1. Rare: Eurozone breakup unprecedented in modern finance
  2. High Impact: Affects ~20% of global GDP, €20 trillion contracts
  3. Retrospectively Predictable: We can reason about causality

Why This Matters

Traditional models say: "It hasn't happened, so we can't model it"

Rebonato-Denev says: "We understand the causal mechanisms, so we CAN model it"

This is the difference between correlation and causation.

Understanding the Network

The Causal Chain

ROOT CAUSES
  Political_Instability (EU tensions)
  Economic_Weakness (GDP, unemployment)
        ↓
CRISIS EVENT  
  Eurozone_Breakup (THE BLACK SWAN)
        ↓
MARKET MECHANISMS
  Credit_Spreads (widen during stress)
  Flight_to_Quality (rush to safety)
        ↓
ASSET IMPACTS
  Corporate_Bonds (credit sensitive)
  Government_Bonds (safe haven)
  Equities (most vulnerable)

Why This Structure?

  • Political_Instability → Increases breakup risk (sovereignty concerns)
  • Economic_Weakness → Increases breakup risk (costs of membership)
  • Eurozone_Breakup → Causes market disruption
  • Credit_Spreads → Corporate bonds become risky
  • Flight_to_Quality → Investors seek safety
  • Asset Classes → React based on their characteristics

Practical Usage

For Portfolio Managers

  1. Monitor leading indicators (politics, economics)
  2. Calculate scenario probabilities using the network
  3. Adjust allocations dynamically
  4. Stress test portfolios under extreme scenarios

For Risk Managers

  1. Model tail risks beyond VaR
  2. Early warning system via probability updates
  3. Scenario planning for crisis response
  4. Regulatory compliance (Basel III stress tests)

For Researchers

  1. Extend to other black swans (pandemic, cyber attack)
  2. Add more variables (central bank policy, contagion)
  3. Learn from data (estimate probabilities empirically)
  4. Integrate with ML (hybrid models)

Next Steps

Immediate

  1. Run the demo script
  2. Look at the visualizations
  3. Read the README for details

Short-term

  1. Modify probabilities to reflect your views
  2. Add new variables to the network
  3. Try different scenarios
  4. Compare with your current portfolio

Long-term

  1. Apply to other crisis scenarios
  2. Integrate with real market data
  3. Build production-grade system
  4. Validate with historical crises

Common Questions

Q: Why not just use correlations?
A: Correlations break down during crises. Causal relationships are more stable.

Q: Aren't the probabilities subjective?
A: Yes, that's the point. Expert judgment is necessary for unprecedented events. But the logic is rigorous.

Q: Can this predict the future?
A: No model predicts the future. But it helps you reason systematically about possibilities.

Q: What about other black swans?
A: The methodology applies to any scenario you can model causally (pandemic, climate, cyber).

Q: Is this better than traditional VaR?
A: It's complementary. VaR handles normal times. This handles tail events.

Resources

Essential Reading

  • Rebonato & Denev (2014) - The original book
  • Pearl (2009) "Causality" - The foundations
  • Taleb (2007) "The Black Swan" - The problem this solves

Technical Background

  • Bayesian networks: Pearl, Koller & Friedman
  • Portfolio theory: Markowitz, Black-Litterman
  • Stress testing: Basel Committee guidelines

Code Examples

All in this package - it's complete and self-contained!

Support

This is an educational implementation. For production use:

  • Consult the original book for full methodology
  • Validate probabilities with domain experts
  • Consider established libraries (pgmpy, PyMC)
  • Add comprehensive testing
  • Integrate real market data

Final Thought

"The core of the problem is that past data is not necessarily indicative of future performance, especially during periods of stress."
— Riccardo Rebonato

This implementation shows how to move beyond that limitation by modeling causality, not just correlation.

Enjoy exploring!