Forecasting Risk Elevated industrial in Sri Lanka using Machine Learning Techniques

Abstract

The stability of a country's financial system is pivotal for economic growth, with banks serving as its backbone.However, recent challenges, including a rise in Non-Performing Loans (NPLs), have necessitated proactive strategies to maintain stability. This study addresses the need for systematic risk evaluation in identifying Risk Elevated Industries (REIs) to mitigate NPL inflow. Through a comprehensive analysis of NPL trends, credit growth, and impairment charges, the study highlights the urgency for accurate industry risk assessment. It aims to forecast REIs using Machine Learning techniques, identifying key factors influencing industry risk and enhancing credit risk forecasting. The research's significance lies in reducing NPL portfolios, increasing lending opportunities in low-risk industries, and improving impairment provisioning accuracy. Additionally, it offers insights into industry risk dynamics, aiding lenders in decision-making and regulatory compliance. Despite data quality and generalization limitations, the study's deliverables provide actionable recommendations for industry stakeholders, fostering a more resilient banking sector in Sri Lanka. The methodology employed in this research encompasses a comprehensive approach aimed at identifying the factors determining Industry Credit Risk. The conceptual framework, developed through a synthesis of literature and expert opinions, categorizes these factors into three main groups: Financial Risk Parameters, Quality of Lending Parameters, and Macro-Economic Parameters. The framework employs the Average Non-Performing Loan (NPL) ratio as a proxy for Industry Credit Risk. A total of eleven independent variables are identified, and model building involves the utilization of multivariate models such as Vector Auto Regression (VAR) and Long Short-Term Memory (LSTM) networks. The performance of these models is evaluated using metrics like Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The study compares the performance of various models in analyzing industry risk. It finds that when applied to non-normalized data, the VAR model outperforms the model with normalized data. Additionally, after removing outliers, the LSTM model demonstrates high accuracy, indicating its efficacy in risk prediction. Furthermore, the LSTM model surpasses the VAR model in performance. Applying these findings to industry analysis, the LSTM model forecast the Construction- Infrastructure industry as Medium risk (NPA 8.40%) , while Tourism (Hotels) (NPA – 5.48%) and Export – Team Industry (NPA – 4.48%) are deemed Low risk, necessitating continuous monitoring by lending officer for future adverse situation.