PhD Research: Validating LLM Understanding of Market Microstructure Through Obfuscation Testing

This repository contains a novel obfuscation testing methodology that validates whether Large Language Models truly understand financial market constraints or simply memorize patterns from training data.

Core Innovation: Strip all temporal context (dates, tickers, events) and force LLMs to reason purely from market structure.

Test Domain: Options market dealer constraints (gamma exposure hedging)

Key Finding: LLMs detect structural dealer constraints with 71.5% detection rate and 91.2% predictive accuracy without any temporal context.

• Methodology - Obfuscation testing framework explained
• Pattern Taxonomy - Three validated dealer constraint patterns
• Key Results - Paper #1 findings (full 2024 validation)

• Getting Started - Installation and quick start guide
• API Reference - Key components and usage

• RoadMap - Multi-paper research trajectory (Papers #1-4+)
• Publications - Papers and presentations

Status: Submitted to IEEE LLM-Finance 2025 Workshop (Oct 26, 2025)

Critical Finding: Detection-Profitability Divergence

• LLM detection remains stable (84-100%) even as profitability declines to zero (Q1→Q4 2024)
• Proves LLM detects market structure, not profits
• Validates methodology rejects temporal context leakage

• Problem: How do we know LLMs understand vs memorize?
• Solution: Strip all temporal context, force reasoning from structure alone
• Validation: Compare obfuscated vs non-obfuscated detection rates

• Explicit causal identification required
• Not just "pattern exists" but "dealers are forced by regulation to hedge negative gamma"
• Mechanistic understanding, not statistical anomalies

• Tested 3 different narrative framings of same underlying constraint
• LLM correctly identifies identical mechanism across framings
• Proves detection is structural, not pattern-matching specific keywords

gex-llm-patterns/
├── src/ # Core system components
│ ├── agents/ # MarketMechanicsAgent (LLM orchestration)
│ ├── gex/ # GEXCalculator (gamma exposure metrics)
│ ├── validation/ # OutcomeCalculator, PatternTaxonomy
│ └── data_sources/ # Historical data fetching
├── scripts/ # Validation and experiment scripts
│ ├── validation/ # Pattern taxonomy validation
│ └── orchestrate_experiment.py # Main entry point
├── docs/ # Comprehensive documentation
│ ├── papers/ # Paper #1 content, research roadmap
│ ├── guides/ # Conceptual guides and tutorials
│ └── presentations/ # Symposium and conference materials
└── reports/ # Validation results (YAML)
 └── validation/
 └── pattern_taxonomy/ # Full 2024 results

# Clone repository
git clone https://github.com/iAmGiG/gex-llm-patterns.git
cd gex-llm-patterns
# Install dependencies
pip install -r requirements.txt
# Set up environment
export PYTHONPATH=$(pwd):$PYTHONPATH
export OPENAI_API_KEY="your-key-here"
# Run validation on single pattern
python scripts/validation/validate_pattern_taxonomy.py \
 --pattern gamma_positioning \
 --symbol SPY \
 --start-date 2024年01月02日 \
 --end-date 2024年03月29日

See Getting Started for detailed setup instructions.

Paper #1 (Submitted Oct 2025):

• "Validating Large Language Model Understanding of Market Microstructure Through Obfuscation Testing"
• IEEE LLM-Finance 2025 Workshop @ IEEE BigData 2025
• Full paper content

Presentations:

• PhD Symposium 2025 (October 2025)
• Research presentation at academic institution

This is an academic research project. For questions or collaboration inquiries:

• Open an issue
• Review active research directions

AGPL-3.0 - See LICENSE

Last Updated: October 25, 2025