We construct a dynamic error correction model of the Australian labor market using a macroeconomic panel across seven states from 1972:3 to 1999:1. Medium-run equilibrium estimates support a real wage-productivity gap and an unemployment gap. The dynamic short-run estimates support expectations-augmented Phillips curves for wages and prices, and demand-led employment growth. We compare three procedures – pooled, aggregate and mean group estimates. Considerable heterogeneity existed across states in the pooled procedure, and state-level variables had a significant impact in the aggregate procedure. Out-of-sample aggregate forecasting for the pooled, aggregate and mean group procedures suggests that the pooled one performs best.