Kevin Litvin -- Simon Fraser University
Published at the Canadian AI 2026 Graduate Student Symposium (GSS).
Cross-exchange cryptocurrency arbitrage enables low-risk profit from price discrepancies across exchanges, yet existing approaches employ negative cycle detection that targets opportunity identification rather than execution feasibility. We introduce an execution-aware pathfinding framework using A* search with domain-specific guidance heuristics, applied to stablecoins -- a unique asset class exceeding $300 billion in market capitalization. The problem is modelled as a weighted directed graph where nodes represent (exchange, stablecoin) pairs across 12 centralized exchanges and edges encode real-world costs including fees, slippage, gas, transfer delays, and exchange reliability. Three guidance heuristics and a multi-start strategy are evaluated over 7,200 search instances. Our slippage-aware heuristic h2 reduces node expansions by 29% relative to Dijkstra while matching its profit, demonstrating that domain-specific heuristics can meaningfully improve execution feasibility in real-time arbitrage planning.
Watch on YouTube: https://www.youtube.com/watch?v=Fuu5OI83uMY
.
├── README.md
├── requirements.txt
├── scripts/ # Core implementation
│ ├── ui.py # Streamlit interactive UI
│ ├── astar_vol.py # A* search with volatility
│ ├── graph.py # Graph construction from live data
│ ├── data.py # Exchange data fetching (CCXT)
│ ├── fees.py # Fee models
│ ├── h1_vol.py # h1: volume/liquidity heuristic
│ ├── h2_slippage.py # h2: order-book slippage heuristic
│ ├── h3_parallel.py # h3: multi-start parallel A*
│ ├── h4_chaincongestion_exchange_risk.py # h4: risk-weighted A*
│ ├── run_ui.sh # Launch Streamlit (background)
│ └── stop_ui.sh # Stop Streamlit
├── experiments/ # Reproducible experiment scripts
│ ├── monte_carlo_simulation.py
│ ├── compare_heuristics_live.py
│ ├── sensitivity_sweep.py
│ └── ...
├── results/ # Raw experiment outputs and logs
├── analysis/ # Figure generation for the paper
├── testing/ # Verification and debugging utilities
└── docs/
├── latex/ # LaTeX source for the papers
│ ├── StablecoinArbitrage_GSS2026/ # GSS paper (primary)
│ └── StablecoinArbitrage_CAIAC2026/ # Full-length paper
├── img/ # README images
├── videos/ # Demo recordings
└── internal/ # Development notes and analysis docs
- Python 3.12+
- pip
git clone https://github.com/kevinl03/Stablecoin-CrossExchange-Arbitrage.git
cd Stablecoin-CrossExchange-Arbitrage
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtstreamlit run scripts/ui.pyOpens at http://localhost:8501. Initial load takes ~1 minute while fetching live market data from all exchanges.
# Monte Carlo simulation (10 trials per heuristic)
python experiments/monte_carlo_simulation.py
# Live heuristic comparison
python experiments/compare_heuristics_live.py
# Sensitivity analysis
python experiments/sensitivity_sweep.py| Heuristic | Strategy | Strength |
|---|---|---|
| h1 (volume) | Prefers high-liquidity routes using 24h trading volume | Reliable execution, 100% success rate |
| h2 (slippage) | Walks the order book to estimate actual price impact | Best for large orders; 29% fewer expansions than Dijkstra |
| h3 (parallel) | Multi-start A* from random entry points | Discovers diverse opportunities |
| h4 (risk-weighted) | Penalises congested chains and unreliable exchanges | 10--20x faster; production-oriented |
See the paper for full experimental comparison across 7,200 search instances.
If you use this code or build on this work, please cite:
@inproceedings{litvin2026execution,
title = {Execution-Aware {A*} Search for Cross-Exchange Stablecoin Arbitrage},
author = {Litvin, Kevin},
booktitle = {Proceedings of the 39th Canadian Conference on Artificial Intelligence,
Graduate Student Symposium (Canadian AI GSS)},
year = {2026}
}This project is released under the MIT License.