A point I've made in old blog posts:
Deviations of target variable from target are the controller's forecast errors, which should be unforecastable, unless the controller make *systematic* mistakes.
Deviations of inflation from 2% target should be uncorrelated with the policy instrument and with all indicators (with a lag equal to targeting horizon).
And the Phillips Curve should be flat (+/- random noise) at 2%.
Etc.
The idea predates him, but Friedman came up with the great "Thermostat" metaphor.
worthwhile.typepad.com/worthwhile_can…