EconSSI

Physics-Informed Systemic Stress Intelligence for Financial Markets

EconSSI is an open-source, research-oriented framework for systemic risk monitoring and forecasting
that integrates econophysics, network theory, critical phenomena, and deep learning to model market-wide stress regimes.

Unlike traditional approaches that focus on short-horizon event prediction only, EconSSI emphasizes regime-level stress dynamics and early-warning signals.

---

Project Status

• ✅ Full research pipeline implemented in: SystemicStressLab_EconSSI.ipynb
• ✅ Dashboard prototype notebook prepared in: Html_Dashboard_EconSSI.ipynb
• ✅ Static HTML research report generated: econssi_report.html
• ⏳ Streamlit deployment (public link) planned as a next step

---

System Architecture

EconSSI follows a modular, physics-informed pipeline designed to track the evolution of systemic fragility rather than predicting single crash dates.

flowchart TD

A[Raw Market Data - Prices and Returns]
B[Market Mode Extraction - Rolling Returns and Correlations]

C1[Network Contagion - Correlation Networks and Lambda1]
C2[Critical Slowing Down - AR1 Variance Autocorrelation]
C3[Multifractality - MFDFA and DeltaH]
C4[Herding Proxies - Ising Spin Glass]

D[Systemic Stress Index SSI]

E1[LSTM Forecasting - Future SSI and Crisis Probability]
E2[Agent Based Modeling - Liquidity Shock Simulation]

F[Monitoring and Decision Support - Dashboard and Early Warning]

A --> B

B --> C1
B --> C2
B --> C3
B --> C4

C1 --> D
C2 --> D
C3 --> D
C4 --> D

D --> E1
D --> E2

E1 --> F
E2 --> F

Loading

This architecture prioritizes regime detection and stress accumulation over short-horizon crash timing.

---

Core Methodology

EconSSI combines multiple physics-inspired layers:

• Network Contagion

  • Rolling correlation networks and eigenvalue concentration (λ₁ ratio)
  • Graph topology metrics (clustering, centralization)

• Critical Slowing Down (CSD)

  • Rolling AR(1), variance, and autocorrelation on market mode proxies

• Multifractality (MFDFA)

  • Periodic multifractal analysis (ΔH) for regime characterization

• Ising / Spin-Glass Proxies

  • Magnetization and susceptibility-style herding indicators

• Systemic Stress Index (SSI)

  • Interpretable composite index built from standardized physics-based features

• Agent-Based Modeling (ABM)

  • Liquidity shock simulations and early-warning proxy signals

• Deep Learning (LSTM)

  • Multi-task forecasting:
    • Regression: forecast future systemic stress level (SSI(t+H))
    • Classification: forecast future crisis probability

---

Key Message

Physics-informed features improve regime-level stress forecasting,
while event-level crisis classification remains intrinsically difficult
under extreme class imbalance.

This framework should be interpreted as a monitoring and early-warning system,
not a deterministic crash-date oracle.

---

Why Brier Score instead of PR-AUC?

Crisis prediction in financial markets is an extreme rare-event problem with strong class imbalance.

In such settings:

• PR-AUC primarily measures ranking performance,
• but does not sufficiently capture the quality of probability calibration.

Brier Score, on the other hand:

• directly measures the squared error between predicted probabilities and realized outcomes,
• evaluates how well-calibrated the risk estimates are,
• is more informative for early-warning and risk monitoring systems.

In EconSSI, crisis probabilities are not interpreted as alarms,
but as gradual indicators of risk accumulation.

A model can therefore exhibit:

• low PR-AUC (poor event timing),
• but a strong Brier Score (accurate risk calibration at the regime level).

This is consistent with the project’s goal of systemic fragility monitoring, not exact crash prediction.

---

How to Interpret Results (Non-Technical)

1) SSI = “Stress Level” (Regime Indicator)

• Think of SSI as a thermometer for overall market fragility.
• High SSI means the system is more vulnerable, not that a crash must happen immediately.

2) Crisis Probability = “Event Risk” (Rare Outcomes)

• Crises are rare; probabilities may appear small even when risk is building.
• Calibration metrics (e.g., Brier Score) are often more meaningful than classification accuracy.

3) Why a model can be “right” with low PR-AUC

• Crisis timing is dominated by:
  • regime shifts,
  • exogenous shocks,
  • market microstructure effects.
• The model’s strength lies in risk buildup detection, not event timing precision.

---

Data, Privacy, and Security

• Data sources consist of publicly available financial time series (e.g., ETF prices and returns).
• This repository does not include:
  • personal data,
  • sensitive information,
  • proprietary datasets.
• If exporting derived artifacts (e.g., df_test.csv) for dashboards, ensure they contain no private or sensitive information.

---

Quickstart

Option A — Run the full research pipeline

Open:

• SystemicStressLab_EconSSI.ipynb

Recommended environment:

• Python 3.10+
• GPU optional (A100 supported)

Option B — Run the dashboard prototype

Open:

• Html_Dashboard_EconSSI.ipynb

Option C — View static research report

Open:

• econssi_report.html

---

License

You may use, modify, and distribute this work with attribution.

---

Citation

If you use EconSSI in academic work, please cite:

EconSSI: Physics-Informed Systemic Stress Intelligence for Financial Markets
Author: Meriç Özcan

---

Author

Meriç Özcan
Statistics Student & Quantitative Risk Researcher