Leveraging Deep Learning for Anomaly Detection in the Interbank Bond Market

Abstract

 This paper explores the application of deep learning techniques to enhance anomaly detection in the interbank bond market, a critical component of the financial system prone to systemic risks. Traditional anomaly detection methods, such as manual checks and rule-based systems, are labor-intensive and often fail to capture complex abnormal behaviors. We propose a novel approach using temporal attribute network embedding and Long Short-Term Memory (LSTM) networks to analyze and detect irregular trading patterns among financial institutions. Our results show a significant improvement in detection accuracy, with an F1 score exceeding 0.7. This study suggests further enhancements through richer trading data integration and the implementation of attention mechanisms to refine detection precision, thereby contributing to the stability and health of financial markets.