In the following pages we examine the generality of the proposition (Buiter (I98I)) that innovation-contingent (or feedback or closed-loop) policies are always superior to non-innovation-contingent (or fixed or open-loop) policies in the design of stabilisation programmes unless (i) they arbitrarily increase the randomness of current and future policy instruments or (ii) the authorities either pursue the wrong objectives or else pursue the right objectives in an inept manner. What is at stake here is the general presumption that acting on more information is necessarily beneficial to the economic agent. In order to tackle the question, we bring together two strands of the literature that have so far remained largely separate. First, the optimum-control strand, which discusses policy within a framework where the authorities are assumed to be Stackelberg leaders and the public Stackelberg followers (e.g. Buiter (I98I), Miller (i984)) and in which the proposition under scrutiny has been proved. Second, the recent suggestion by Barro and Gordon (I983 a, b) that the policy game in the absence of pre-commitments may be better approximated by a Nash non-cooperative equilibrium between the authorities and private agents on the grounds that it is the only incentive-compatible notion. In this case, the solution to the game is constructed in such a way that, in equilibrium, each agent is maximising his utility given the strategy chosen by the opponent, i.e. each player has chosen his best reply to the other players' moves. It will be shown here that once a Nash equilibrium is assumed, contingent policies do not dominate fixed policies. In the process, it will also become apparent that once innovation-contingent responses are allowed for, Barro and Gordon's argument against discretionary policy is strengthened: even if the deterministic component of the Nash policy is non-inflationary, its innovationcontingent part is counterproductive. In other words, even if the authorities aim at stabilising output at, and not above, the natural rate (or, equivalently, do not derive utility from positive deviations from the natural rate), the absence of precommitment entails a welfare loss. In section I we briefly discuss the relationship between information and precommitment and rules versus discretion to avoid misunderstanding of the thrust of our analysis. In sections II-IV we demonstrate the central proposition in two different classes of models. First, in models, as in Buiter (i98I), wlhere the decision-takers are restricted to react to exogenous shocks with a one-peried lag. Second, in models (e.g. B. Friedman (I977) and Courakis (I984)) where, while still unable to observe directly the behaviour of the ultimate goal variables,