Accurate modelling of earnings dynamics is critical for projections of poverty dynamics and inequality, taxation revenue and savings, and public policy costings. Despite extensive analysis of earnings dynamics using the PSID, there has been little research using Australian longitudinal sources. This paper seeks to partially address this gap by using the first seven waves of HILDA to estimate and compare dynamic wage models. Residuals from a mean fit to hourly wage are decomposed into permanent and transitory components, and it is found that the variance of the components differ by employment state. A random walk for the permanent component can be justified on economic and empirical grounds, and the models are extended to allow for non-Gaussian errors and serial correlation in the transitory component. Short-term predictive ability of the models is mixed, and possible improvements are discussed. The implications of the models developed are illustrated through projections of outstanding debt and repayments through the Higher Education Contribution Scheme for a hypothetical population.