Since forecast failure is due to unanticipated deterministic shifts, "sensible" agents should adopt "robust forecasting rules". In such a non-stationary world, causal variables can dominate non-causal in forecasting, so "rational expectations" do not have a sound basis: agents cannot know how all relevant information enters the data density at every point in time. Although econometric models "break down" intermittently, that is not due to the Lucas critique and need not preclude policy analyses.