Abstract Modern DSGE models are microfounded and have deep parameters that should be invariant to changes in economic policy, so in principle they are not subject to the Lucas critique. But the literature has already established that misspecification issues also cause parameter instability after policy changes in DSGE models. This paper will look at the implications of parameter shifts for econometric policy evaluation, to see whether policy advice derived from DSGE models would have differed fundamentally from that which the policymakers of the 1970s derived from their reduced-form Phillips curves. The results show drift in most parameters, including those that are supposedly structural (such as the share of capital in production, habits or the elasticity of labour supply to the real wage), and major shifts in the impulse response functions derived from the real-time estimation of the model. After the expansionary monetary shocks of the early 1970s, a standard DSGE model would have behaved very similarly to an old-style Phillips curve, with marked shifts in parameter values and impulse response functions.