Predicting financial distress as repeated events

Purpose The purpose of this paper is to model financial distress as repeated events using discrete-time hazard analysis and evaluate which method that accounts for dependence on prior event history provides superior predictive performance. Design/methodology/approach Using quarterly data from China’s listed firms (1995–2021), the authors estimate financial distress prediction models in discrete-time hazard models using, in turn, cluster-robust standard errors only, generalized estimating equations, random effects, fixed effects and a hybrid approach. Models are validated using both cross-sectional and longitudinal test samples and assessed via areas under the curve and recall. Findings Random effects and the hybrid method consistently outperform other methods in predictive accuracy and better capture within-firm correlation and event history. Research limitations/implications The methods to control for dependence upon event history are generalizable to other recurrent events such as credit rating changes or mergers, supporting broader adoption of repeated-event modelling. Originality/value To the best of the authors’ knowledge, this is the first study to model financial distress as repeated events using time-dependent methods with time-varying predictors and prior event history.