Whether the equilibria towards which economic variables gravitate are stable or exhibit historic path dependence (i.e. hysteresis) is of central importance for the efficacy of stabilization policy. Given that such hysteresis is a natural consequence of nonlinearity in the conditional mean of the data generating process, we lest for the presence of hysteretic path dependence by examining the time series properties of monthly U.S. aggregate and disaggregate data over the post Bretton Woods period 1971-1993. Our results suggest that such nonlinearity is indeed present and capable of representation using bilinear and threshold autoregressive models, estimates implying strong asymmetry in the behaviour of our series with recessionary periods being inherently more‘'noisy’. Finally, we consider generalized threshold vector autoregressive models where threshold regimes are conditioned on the sign of the change in the growth rate of industrial production (i.e. the ‘business cycle’). These models reveal that cyclical behaviour of unemployment and the real wage is complex, being procyclical during business cycle expansions but countercyclical and independent, respectively, during contractions.