Bibliometric Analysis of Bankruptcy Prediction in Financial Institutions: Themes, Evidence, and A Future Research Agenda

Research Question: What are the dominant themes and methodological gaps revealed through co-word and bibliographic coupling analyses of Bankruptcy Prediction in Financial Institutions? Motivation: Starting from the need for an accurate and auditable early warning system in financial institutions, this study maps the bankruptcy prediction research landscape through a Web of Science-based bibliometric analysis (1970-2025) and VOSviewer visualization. Idea: This article combines performance analysis (publications, citations, and h-index) with science mapping (bibliographic coupling and co-word analysis) to explore intellectual foundations, research frontiers, and methodological gaps. Data: Publication record from 1970 to 2025 in Web of Science with a total of 2,410 articles. Method/Tools: The methods start by defining the scope of the topic (Bankruptcy Prediction in Financial Institutions) and the time horizon (1970-2025). The list of keywords is compiled iteratively by combining key terms and their synonyms. Boolean operators and wildcards are used to expand/detail the outcome. This study implemented the co-word and bibliographic coupling analyses using VOSviewer. Findings: The results show 2,410 publications, 42,247 citations, and an h-index of 89; output and impact have increased sharply since 2015, indicating an acceleration of interest in this topic. Four main clusters are identified: (1) Machine Learning (ML) ensembles for credit risk & bankruptcy, (2) sovereign-bank nexus, capital, and credit risk systems, (3) capital risk, governance & bank failure prediction, and (4) network-based systemic risk & tail-connectedness. The analysis also reveals divergent findings between conventional logit models and ML algorithms, the importance of addressing class imbalance, the need for temporal (out-of-time) validation, and feature selection contextualized to regulatory and systemic dynamics in the financial sector. Contributions: This study presents a concise taxonomy of methods and proposes a research agenda to address identified gaps and promote more accurate, transparent, and testable models across market regimes. This study provides a compass map for designing research that is both relevant and replicable.