This paper shows that regime shifts in Full-Information Rational Expectations (FIRE) models generate predictable regime-dependent forecast errors in macro aggregates. Hence, forecast error predictability alone is neither sufficient to reject FIRE nor informative about alternative expectations theories. We instead propose a regime-robust test of FIRE and apply it to a medium-scale New Keynesian model with monetary policy regime shifts that is estimated on US data. While the test fails to decisively reject FIRE, the model conditional on macro data implies expectations that are generally different from observed survey forecasts, thus providing a new empirical motivation for alternative expectations theories.
