Abstract
The proliferation of High-Frequency Trading (HFT) has fundamentally altered the microstructure of modern financial markets, introducing liquidity benefits alongside complex vulnerabilities. Among these vulnerabilities, manipulative strategies such as spoofing and layering have become increasingly sophisticated, often evading detection by traditional rule-based surveillance systems and aggregate statistical methods. This paper proposes a novel framework for detecting such anomalies by modeling the Limit Order Book (LOB) not as a time series or an image, but as a dynamic temporal graph. We introduce the concept of Order-Book Temporal Motifs—recurring, directed subgraphs that encode the causal intent of market participants across micro-timestamps. By mining these motifs, we construct a feature space that captures the intricate interplay between order placement, modification, and cancellation. We subsequently employ a Motif-Aware Graph Neural Network (MA-GNN) to classify these patterns in real-time. Our experimental results, conducted on high-resolution LOBSTER data, demonstrate that this graph-centric approach significantly outperforms state-of-the-art Deep Learning baselines in identifying manipulative intents, particularly in scenarios characterized by high message volatility.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Copyright (c) 2025 Zhiqiang Zheng (Author)