Abstract
Purpose This paper aims to investigate whether the amount of local governments' debt can be predicted by the level of political competition. Design/methodology/approach The study uses the artificial neural network (ANN) to test whether ANN can "learn" from the observed data and make reliable out-of-sample predictions of the target variable value (i.e. a local government's debt level) for given values of the predictor variables. An ANN is a non-parametric prediction tool, that is, not susceptible to the common limitations of regression-based parametric forecasting models, e.g. multi-collinearity and latent non-linear relations. Findings The study finds that "political competition" is a useful predictor of a local government's debt level. Moreover, a positive relationship between political competition and debt level is indicated, i.e. increases in political competition typically leads to increases in a local government's level of debt. Originality/value The study contributes to public sector reporting literature by investigating whether public debt levels can be predicted on the basis of political competition while discounting factors such as "political ideology" and "fragmentation". The findings of the study are consistent with the expectations posited by public choice theory and have implications for public sector auditing, policy and reporting standards, particularly in terms of minimising potential political opportunism.