Draft paper: A Systematic Literature Review on Graph-Based Models in Credit Risk Assessment

This is the draft paper: A Systematic Literature Review on Graph-Based Models in Credit Risk Assessment The paper has niot been published yet. Thus, it is at the moment restricted. The abstract of this paper: This review provides a comprehensive analysis of graph-based models in credit risk assessment within the financial industry. It systematically categorizes and evaluates various models, including factorial network models, Graphical Gaussian Models (GGMs), Graph Neural Networks (GNNs), network centrality measures, community detection methods, dynamic multi-layer networks, and advanced techniques like graph attention networks and hypergraphs.The analysis highlights the comparative advantages of graph-based models over traditional approaches in capturing complex relationships and contagion within financial networks. Factorial network models and GGMs excel in understanding latent factors and systemic risks, while GNNs and network centrality measures enhance predictive accuracy and explainability. Community detection and dynamic multi-layer networks offer insights into risk transmission and systemic risk. Advanced techniques such as graph attention networks, hypergraphs, and knowledge graph models integrate diverse data sources for holistic credit risk assessment. Additionally, the review underscores the potential of graph-based models in handling imbalanced data, improving credit scoring for thin-file borrowers, and mitigating financial contagion. The findings emphasize the need for future research to integrate early warning systems into customer segmentation frameworks and extend the utility of graph-based models to identify positive financial behaviors and lending opportunities.