The Federal Reserve Board published a research paper that builds a comprehensive dataset on US colleges and universities from 2002 to 2023 and uses it to develop predictive models for institutional financial distress and closure. The paper finds that missing data is pervasive among institutions that ultimately close and is a major obstacle to identifying at-risk colleges, while modern machine learning approaches using richer inputs outperform both simple statistical models and existing federal screening mechanisms. The analysis combines operational, staffing and financial characteristics including revenue and expense patterns, sources of revenue, liquidity and leverage metrics, enrollment and staff trends, and prior indicators of significant strain. Model performance is benchmarked against federal tools such as financial responsibility scores and heightened cash monitoring, with the preferred model pairing an off-the-shelf machine learning algorithm with the richest set of explanatory variables and delivering the largest gains, particularly where data are incomplete. Simulations based on the estimates indicate that enrollment pressures linked to an approaching demographic cliff could materially increase annual college closures under reasonable scenarios.
Federal Reserve Board 2025-01-01
Federal Reserve Board research shows richer data and machine learning better predict college financial distress and closures
The Federal Reserve Board released a research paper developing predictive models for financial distress and closure of US colleges from 2002 to 2023. The study highlights pervasive missing data among institutions that close and demonstrates that machine learning models using comprehensive data outperform existing federal screening tools. Simulations suggest that demographic shifts could significantly increase college closures.