Since the consequences have to be predicted, there is a close relation between decision and forecast. When it is assumed that there are no cognitive limitations, agents behave in conditions of full or substantive rationality. The best action, which maximizes expected utility, is conditioned by the complete knowledge of all its possible consequences, of the degrees of value ascribed to that action and also to all its consequences; in addition, all the possible alternative actions, their consequences and their values have to be known. It would seem that in order to know the best action at a given moment, an agent should know the entire future of the world. In other words, decision makers are assumed to know the utility function, to have complete knowledge of the available means, to be able to foresee the relevant future with accuracy and to have no computational limitations. In these conditions their problem is purely logical. They will choose the action that maximizes expected utility, and they behave as if the expected value is the certain value of utility. The problem of choice has been transformed into an equivalent certainty problem, because the calculus of probability is 'supposed to be capable of reducing uncertainty to the same calculable status as that of certainty itself' (Keynes, 1973c , pp. 112-13).
Individual and strategic rationality are distinct. The rational expectations hypothesis (REH) well describes individual substantive rationality. It assumes that (a) agents maximize their utility functions subject to some constraints, and (b) the constraints are perceived by all the agents in a mutually consistent way (the mutual consistency of beliefs). In the sense of Muth (1961), this means that decision makers are able to represent their beliefs as probability distributions, and that their subjective probability distribution coincides with the objective probability (also called frequency probability) distribution. This definition of rationality is stricter than that admitted by Bayesian rationality, which does not require subjective probabilities to be equal to objective probabilities. In a static situation, the REH equilibrium is that of a perfect competition model where all the agents have the same information set because they know the structure of the model and the true value of parameters. In a dynamic situation, in contrast, it is assumed that agents behave in recurring situations, which they have experienced before. Therefore, they know the laws that govern the economic system, and their forecasts are correct estimates of its future trends.
When we move from individual rationality to strategic rationality, agents' interactions are modelled through game theory. Agents have to consider that the economic environment is composed of other agents, that they are part of other agents' environment, that other agents are aware of this, and so on. This means that they must also be able to foresee all the possible actions undertaken by others according to a probability distribution. Von Neumann and Morgenstern (1947, p. 19) assume that, in a condition of risk, each agent knows the objective probability distribution that governs the random process. They justify this choice by maintaining that
probability has often been visualized as a subjective concept more or less in the nature of an estimation. Since we propose to use it in constructing an individual numerical estimation of utility, the above view of probability would not serve our purpose. The simplest procedure is, therefore, to insist upon the alternative perfectly well founded interpretation of probability as frequency in long runs. This gives directly the necessary numerical foothold.