Regime-aware quant risk and market stability monitoring framework.
(formerly THERMOQUANT – research name)
RegimeGuard is a regime-aware quant risk and market stability monitoring framework.
It is designed to answer one core question:
When are markets structurally safe or unsafe for models, strategies, and assumptions?
RegimeGuard does not predict prices, generate trading signals, or execute trades.
Instead, it operates as a meta-risk layer that evaluates market regimes, structural instability, and model reliability.
Think of it as weather radar for markets — not to forecast prices, but to warn when conditions become dangerous.
Most financial losses do not occur because models are incorrect.
They occur because models are used during the wrong market regime.
Traditional risk tools (volatility, correlation, VaR):
- assume stable distributions
- react late to regime shifts
- fail during systemic transitions
- ignore irreversibility and coordination failure
RegimeGuard addresses this by modeling markets as non-equilibrium systems and tracking structural instability, not price direction.
- A market regime and stability monitor
- A quant risk & model-reliability system
- A non-predictive decision-support framework
- A research-grade implementation with a working API
- ❌ Not a trading system
- ❌ Not an alpha generator
- ❌ Not a signal engine
- ❌ Not a backtesting platform
- ❌ Not investment advice
RegimeGuard focuses on risk governance, not performance optimization.
RegimeGuard is built in three conceptual stages.
Models market-wide instability using physics-inspired state variables:
- Entropy – distributional uncertainty
- Entropy rate – instability acceleration
- Free energy – available exploitable structure
- Instability probability
This layer detects latent fragility, even when volatility appears low.
Models endogenous instability caused by agent interaction:
- payoff dispersion
- coordination breakdown
- structural stress index
This explains why instability emerges internally, not only from external shocks.
A non-predictive observer that classifies market regimes:
- ORDERED
- TRANSITION
- DISORDERED
Machine learning is used only to detect regime structure —
not to predict prices or returns.
- Non-predictive by design
- Regime-first, not return-first
- Physics-inspired constraints, not curve fitting
- Explainable outputs, not black boxes
- Works without historical price dependence (no-data validation possible)
RegimeGuard exposes a minimal FastAPI interface.
-
GET /regime_state
Returns the current market regime and confidence -
GET /stability
Returns structural stability metrics and instability probability -
GET /structural_stress
Returns agent coordination stress indicators -
GET /explain
Provides human-readable explanations for current risk conditions
The root endpoint (/) redirects to /regime_state.
{
"regime": "DISORDERED",
"confidence": 0.87
}Interpretation
When RegimeGuard reports an unstable regime, it implies:
Market structure is unstable
Quant assumptions may be unreliable
Aggressive strategies should be disabled
This output is intended to guide risk posture, not trade direction.
How RegimeGuard Is Used
RegimeGuard is designed to be used alongside existing models, not instead of them.
Typical usage includes:
Enabling or disabling strategies based on the detected regime
Dynamically adjusting leverage and exposure
Detecting when diversification assumptions are likely to fail
Providing defensible, structural explanations to risk committees
RegimeGuard governs when models should be trusted, not what trades to place.
Typical Users
RegimeGuard is intended for institutional and research-focused users, including:
Hedge funds (quantitative and discretionary)
Asset management firms
Family offices
Risk and model-governance teams
Macro and systemic-risk researchers
## Repository Structure
regimeguard/
├── api/ # FastAPI application
├── core/ # Entropy, energy, instability logic
├── agents/ # Agent stress & coordination models
├── observer/ # Regime classification logic
├── validation/ # No-data and structural tests
├── examples/ # Demo runs
├── run.py # API runner
└── README.md
Installation & Running pip install -r requirements.txt python run.py
Open in your browser:
http://127.0.0.1:8000/
API documentation (Swagger UI):
http://127.0.0.1:8000/docs
Disclaimer
This project is provided for research and educational purposes only.
It does not provide investment advice, trading signals, portfolio recommendations, or execution logic.
Use at your own risk.
License
This project is licensed under the Apache License 2.0.
You are free to use, modify, and extend it, including for commercial purposes, subject to the license terms.
Project Status
MVP complete
Core logic implemented
API operational
Ready for research use, pilots, and further development
Contributions & Discussion
This repository represents an ongoing research and engineering effort.
Contributions, critiques, and discussions are welcome — especially around:
market regimes
systemic risk
non-equilibrium finance
model-risk governance
One-Line Summary
RegimeGuard is a market regime and risk monitoring system that determines when quant models and assumptions are structurally safe or unsafe — without predicting prices.
- Fully complete
- GitHub-ready
- Institution-safe
- Research-credible
- Product-aligned
- Zero hype