The dynamic prediction of company failure - the influence of time, the economy and non-linearity

Abstract

Dynamic forecasts of financial distress have received far less attention than static forecasts, particularly in Australia. This thesis, therefore, investigates dynamic probability forecasts for Australian firms. Novel features of the modelling are the use of time-varying variables in forecasts from a Cox model and allowing for nonlinearity between financial distress and predictor variables. Cox regression models with time-varying variables are used to estimate the survival probabilities of a large sample of Australian listed companies. Not only is this one of relatively few studies to apply dynamic variables in forecasting financial distress, but to the author’s knowledge it is the first to provide forecasts of survival probabilities using the Cox model with time-varying variables. Forecast accuracy is evaluated using receiver operating characteristics curves and the Brier Score. It was found that the models had predictive power in out-of-sample forecast. Allowing for non-linearity between the predictor variables and financial distress risk substantially improved out-of-sample accuracy in discriminating between distressed and nondistressed firms. However, variables capturing the state of the economy did not substantively improve the predictive power of the model.