The authors previously developed a dynamic Cox hazards model with time-varying covariates to estimate and forecast the events of financial distress for Australian firms (Kim and Partington, 2014). Yet, in common with most prior studies on financial distress, they neglect the potential non-linear relation between individual predictor variables and the event of interest (the risk of financial distress). This paper extends Kim and Partington (2014) by addressing both the effect of non-linearity and the impact of the state of the economy on the financial distress risk. Specifically, each predictor variable is transformed non-parametrically based on the univariate empirical mapping with the observed failure rates. With the use of data on publicly listed companies on the Australian Securities Exchange (ASX) from 1995 to 2006, the models are calibrated using an estimation sample ((1995-2002) and the out-of-sample discriminatory power (based on the ROC curve) and accuracy (based on the Brier Score) are assessed on the holdout sample (2003-2006). It is found that the discriminatory power (but not the accuracy of probabilistic predictions) of the models is substantively improved by catering for the non-linear relations between the risk of financial distress and predictor variables. However, it is shown that variables capturing the state of the economy do not add to the predictive power when timevarying firm-specific variables are already included in the model.