PhD research investigating whether Large Language Models can detect structural constraints in financial markets through genuine reasoning rather than training data memorization.

Core Innovation: Temporal obfuscation testing — stripping all dates, ticker symbols, and contextual markers from financial data, forcing LLMs to reason from numerical structure alone.

Application Domain: Options dealer gamma exposure (GEX) — the aggregate hedging constraints that market makers face due to their options inventory positions.

Finding: LLMs reconstruct dealer positioning from raw strike-level data, outperforming pre-calculated metrics — empirical evidence that scalar GEX aggregation discards structural signal.

Finding: Detection tracks 0DTE options adoption — 3.7% (2021) → 100% (2024) — with GEX magnitude growing 360%, revealing a structural market reorganization.

Stable detection (68–74% quarterly) persists while economic profitability collapses (Sharpe 1.8 → 0.1), confirming detected patterns are structural mechanics, not exploitable anomalies.

Title: Inferring Latent Market Forces: Evaluating LLM Detection of Gamma Exposure Patterns via Obfuscation Testing

• arXiv: 2512.17923
• Venue: IEEE LLM-Finance 2025 (IEEE BigData, Macau)
• LaTeX: docs/papers/paper1/
• Status: Published, journal revision with 5 MC defenses complete

Title: Validating LLM Structural Reasoning: Detecting Persistent Market Regimes Through Temporal Obfuscation

• Core finding: 2023→2024 structural shift driven by 0DTE adoption
• Formats: AIAI/Springer LNCS, JFDS/Elsevier, JFQA/Cambridge
• LaTeX: docs/papers/paper2/
• Status: Ready for submission, multi-venue formats prepared

Title: Validating LLM Structural Reasoning in Options Markets: From Single-Day Detection to Persistent Regime Identification

• Format: IEEE Access journal (single-column, ~20 pages)
• LaTeX: docs/papers/ieee_access/
• Status: 18 pages, sensitivity analysis complete, expanding

Title: Validating LLM Structural Reasoning: Detecting Persistent Market Regimes Through Temporal Obfuscation

• Format: Springer svjour3, author-year citations (6,000–10,000 words)
• LaTeX: docs/papers/digital_finance/
• Status: 24 pages (~7,000 words), includes Hayek/Kirzner theoretical interpretation

• Focus: Intraday OI-to-volume pattern validation
• Status: Monitor running since Jan 2026, collecting data (target: 60+ trading days)
• LaTeX: docs/papers/paper3/

Raw:        SPY, 2024-03-15: Net GEX: -$32.9B, Flip: $485.00
Obfuscated: Day T+0, INDEX_1: Net GEX: -$32.9B, Flip: $485.00

Remove dates, tickers, events → preserve only quantitative structure → force structural reasoning.

Every detection must specify:

• WHO: The constrained actor (e.g., dealers with negative gamma)
• WHOM: The affected parties (e.g., directional traders)
• WHAT: The forced mechanism (e.g., pro-cyclical hedging amplifying volatility)

• PostgreSQL 18.1: 81.8M contracts, 50 symbols, 2020–2025 (20.58 GB)
• Intraday snapshots: Yearly-partitioned table, 21 snapshots/day

• Alpha Vantage Premium: Historical options chains (1000 calls/min)
• Polygon.io: Stock price data (free tier)

• OpenAI o4-mini: Reasoning model via Batch API
• Cost: $11.07 for all 2,221 evaluations

gex-llm-patterns/
├── src/
│   ├── agents/              # LLM market mechanics agent
│   ├── analysis/            # Pattern library (15 patterns)
│   ├── gex/                 # GEX calculator (Black-Scholes)
│   ├── llm/                 # LLM integration
│   ├── validation/          # Obfuscation & regime classification
│   └── data_sources/        # Alpha Vantage, Polygon clients
├── scripts/
│   ├── validation/          # Paper 1 & 2 validation pipelines
│   ├── analysis/            # Sensitivity analysis, figures
│   └── data_collection/     # Intraday OI monitor
├── docs/
│   ├── papers/
│   │   ├── paper1/          # Single-day obfuscation (IEEE BigData)
│   │   ├── paper2/          # Regime detection (multi-venue)
│   │   ├── ieee_access/     # Combined journal paper
│   │   ├── digital_finance/ # Springer journal paper
│   │   └── paper3/          # Intraday infrastructure
│   └── presentations/       # PhD symposium, fundamentals
├── reports/                  # Validation results (YAML)
└── config_defaults/          # Configuration templates

git clone https://github.com/iAmGiG/gex-llm-patterns.git
cd gex-llm-patterns
pip install -r requirements.txt

# Configure API keys
export OPENAI_API_KEY="your_key"
export POLYGON_API_KEY="your_key"

# Verify
python -c "from src.analysis.pattern_library import PatternLibrary; print('OK')"

• Academic research only — not trading advice
• Public data only — Alpha Vantage options chains
• Open source — methodology and code fully available
• Not financial advice — detection rates do not imply profitable strategies

@inproceedings{regan2025obfuscation,
  author = {Regan, Christopher and Xie, Ying},
  title = {Inferring Latent Market Forces: Evaluating {LLM} Detection of
           Gamma Exposure Patterns via Obfuscation Testing},
  booktitle = {2nd IEEE International Workshop on Large Language Models
               for Finance, IEEE International Conference on Big Data},
  year = {2025},
  publisher = {IEEE},
  url = {https://arxiv.org/abs/2512.17923}
}

GNU Affero General Public License v3.0 — see LICENSE.

Last Updated: March 2026 Contact: Christopher Regan (cregan1@kennesaw.edu) · Ying Xie (yxie2@kennesaw.edu) Institution: Kennesaw State University, College of Computing and Software Engineering