Deep Learning Reveals Regulatory Impacts on European Banking

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jul 19 2024

In a paper published in the Journal of Risk and Financial Management, researchers examined the regulatory impact on the European banking sector using advanced deep learning (DL) techniques by analyzing the relationship between sustainable finance guidelines and the index from January 2012 to December 2023. They employed long short-term memory auto-encoder (LSTM-AE), variational autoencoder (VAE), and convolutional neural network (CNN) for anomaly detection.

The study found significant anomalies aligned with regulatory events, particularly highlighting the strong correlation between European Banking Authority (EBA) sustainable finance events and anomalies using VAE. This research advances the understanding of DL applications in financial markets and offers insights for policymakers and financial institutions.

Background

Past work has extensively examined the intersection of regulations, sustainability, and efficiency in various industries. Researchers have explored the impact of regulatory measures on banking efficiency and stability, highlighting the importance of sustainability in banking strategies and regulatory contexts. Additionally, the application of artificial intelligence (AI), machine learning (ML), and DL in finance has been thoroughly reviewed, demonstrating their effectiveness in predictive systems, anomaly detection, and risk assessment.

Anomaly Detection Methodology

Data was collected from Yahoo Finance, focusing on the STOXX Europe 600 Banks Index (SX7P). The period chosen to capture the dataset encompasses periods marked by impactful economic events, regulatory shifts, and technological advancements that influenced the banking sector, such as the recovery from the financial crisis, Brexit, and the pandemic of coronavirus disease 2019 (COVID-19). The SX7P index was chosen for its inclusive representation of major European banks, providing a strong foundation for comprehending the European banking sector's development, challenges, and possibilities.

Three DL models—LSTM-AE, CNN, and VAE—were used for anomaly detection. The LSTM-AE model excels at learning temporal relationships in sequential data. CNNs are adept feature extractors recognized for analyzing spatial patterns in time series data, and VAEs offer a probabilistic framework designed to learn latent representations and capture uncertainty. The training dataset comprised 1864 samples, while the test dataset included 1028.

The approaches were selected for their ability to capture complex patterns and structures in time-related data, making them highly suitable for anomaly detection tasks. Standardized parameters were utilized throughout the experimental process to ensure consistency and comparability across the methodologies.

The analysts used a time step of 20 for each input data point, applied a dropout rate 0.2 during training to prevent overfitting, and established a standardized threshold value of 2.30671 to distinguish between normal and abnormal instances. The training phase spanned 100 epochs with a batch size of 32.

This methodological consistency guarantees an equitable and rigorous comparison among the three DL approaches, facilitating a thorough assessment of their effectiveness in identifying anomalies within the time-series data under scrutiny. The objective is to determine the optimal algorithm for detecting anomalies in the European Banking Sector dataset, offering insights into the strengths and weaknesses of LSTM-AE, VAE, and CNN methodologies.

Anomaly Detection Analysis

The study employed three DL methodologies—LSTM-AE, CNN, and VAE—to detect anomalies in the SX7P dataset from September 2012 to December 2023. Each model exhibited unique strengths and patterns in anomaly detection. LSTM-AE identified 27 anomalies, primarily clustered around periods coinciding with significant global events like the COVID-19 pandemic, showing minimal correlation with EBA sustainability guidelines. CNN detected 23 anomalies, notably concentrated during the pandemic's onset, with varying degrees of correlation to economic indicators such as unemployment rates.

VAE detected 63 anomalies and demonstrated sensitivity to economic output fluctuations, showing alignments with specific EBA sustainability publications. Analysis of macroeconomic variables revealed insights into each model's sensitivity. VAE displayed a moderate positive correlation of 0.36, suggesting economic conditions influenced its anomalies. Conversely, it exhibited weak negative correlations with inflation and interest rates, indicating potential inverse relationships during stable economic periods.

CNN demonstrated strong sensitivity to unemployment rates (0.4), while LSTM-AE's anomaly detection relied less on macroeconomic factors, suggesting robustness to latent variables and temporal patterns in capturing banking sector anomalies. These insights enhance anomaly detection strategies amid economic and regulatory changes in financial markets.

Conclusion

To sum up, this study uniquely integrated DL methods with analyzing EBA regulatory impacts in the transition to a green economy. It emphasized banks and regulators adopting ML tools to enhance anomaly detection and support sustainable financial practices.

Future research was suggested to address methodological limitations and explore additional economic variables to improve accuracy. Overall, the study underscored the efficacy of LSTM-AE, CNN, and VAE in detecting regulatory-driven anomalies and offered insights into the nuanced impacts of EBA regulations on the European banking sector.