Answers are grounded in this paper and cite their sources.
This study examines whether employee perceptions of business outlook provide incremental predictive information about corporate bankruptcy beyond traditional financial, market-based, and audit variables. Using employee-generated data from Glassdoor, we analyze 20,757 U.S. firm-year observations from 3,830 firms during 2012-2021, including 96 bankrupt firm-year observations. Because bankruptcy is a rare event, we evaluate predictive performance primarily using the area under the precision-recall curve (PR-AUC). We compare baseline models with and without one-year-lagged employee outlook using Random Forest, XGBoost, and a leakage-resistant Stacking model under firm-grouped cross-validation and an out-of-time evaluation. In the out-of-time test, adding employee outlook increases PR-AUC from 0.279 to 0.335 for Random Forest and from 0.281 to 0.338 for the Stacking model, with both improvements statistically distinguishable from zero. The gains are smaller under firm-grouped cross-validation and are not consistently observed across alternative robustness tests. Overall, the findings demonstrate that employee outlook can enhance near-term bankruptcy prediction under certain modeling approaches, particularly in out-of-time forecasting. However, the magnitude and consistency of these improvements vary across classifiers and validation settings. These results underscore the importance of rigorous validation procedures and appropriate evaluation metrics when assessing predictive performance in rare-event classification.
No takes yet. Share an insight, caveat, or question.
Cheng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: