Prediction of business bankruptcy has been a popular topic with both academics and practitioners for decades. One way to enhance predictive models is to introduce measures of corporate efficiency calculated by DEA (Data Envelopment Analysis). However despite of the attention DEA received in recent years, applications to bankruptcy/financial distress so far are only restricted to cross-sectional or static models. Dynamic credit risk models have several advantages over the static ones but obviously they require the covariates of efficiency to be measured in multiple periods. This paper bridges this gap by linking dynamic discrete hazard models and time-varying Malmquist DEA. It analyses financial distress in different industries separately and over different time periods with respect to assumptions of Variable Returns to Scale and the homogeneity of technology. Based on a sample of 742 Chinese listed companies over 10 years, dynamic DEA efficiency scores are successfully used as covariates to predict the probability of distress in discrete hazard models. Results suggest that Malmquist DEA can be used to make dynamic predictions over different time periods. The predictive accuracy varies depending on the efficiency reference sets: a global reference performs best in homogenous industry samples while a generic model is superior in heterogeneous samples.