Manipulation Detection of the Bank Sector

This paper presents a novel and explainable approach to detecting manipulation in the banking sector, par-ticularly spoofing in financial markets. Using the LOBSTER dataset [1], we apply a Hidden Markov Model [2] (HMM) to identify rapidly placed and cancelled orders indicative of market manipulation. Our method emphasizes interpretability, making it suitable for compliance and regulatory settings. Compared to tra-ditional machine learning baselines, the HMM balances accuracy with transparency. We provide a detailed methodology, including data preparation, model training (via Baum- WelchlEM), feature extraction (order size anomalies, placement- to-cancellation latency, and price impacts), and baseline comparisons against ML models such as Decision Trees and XGBoost. The results demonstrate the effectiveness of the HMM in capturing hidden states of market behaviour, providing transparency and reliability in spoofing detection.