TRADER: Real-time Arbitrage Detection via Negative Cycles on Dynamic Graphs

Real-time arbitrage detection in decentralized finance (DeFi) is challenging, as opportunities, which are often represented by negative cycles in token graphs, appear and disappear within milliseconds. High update rates and algorithmic complexity make real-time detection difficult, while existing approaches either operate only on static snapshots, which quickly become outdated, or recompute all affected cycles after each update, which is prohibitively expensive. Both approaches fail to meet the latency and throughput requirements of modern trading systems. To address the challenge, we present TRADER, an incremental solution for high-performance real-Time aRbitrAge DEtection via negative cycles on dynamic gRaphs. TRADER uses a dynamic programming-based algorithm with incremental state updates and localized propagation to detect and maintain arbitrage opportunities within milliseconds on fast-changing crypto token graphs. It supports edge updates in batch and employs an edge grouping strategy to filter out non-impactful updates, reducing latency and boosting throughput. On realworld DeFi datasets, TRADER achieves up to $5.8 \times$ speedup over baselines while maintaining high detection accuracy.