Rational expectations as an optimal approximation

The rational expectations hypothesis has somehow fallen into disrepute because it is viewed as somehow failing to predict or account for the recent crisis. This is of course because of a fundamental misunderstanding of the hypothesis, as it does not imply that markets are efficient, or in equilibrium, or that the equilibrium is unique, or that bubbles cannot happen. But it is a hypothesis, and as any hypothesis it could be rejected by the evidence, for example by experiments showing people are afraid of Knightian uncertainty and yet are optimistic. But the fact that there was a crisis is not empirical evidence in this regard.

Kenneth Kasa actually shows that the rational expectations hypothesis is also a result, up to a close approximation. Indeed, if one is uncertain about the economic environment ("does not know the model"), one adheres to robust rules in the sense that among all possible models, one picks the one with the worst possible outcome. Call this the 'evil' agent. Add to this the assumption that one likes being optimistic. Call this the 'angelic' agent. Now assume that the evil and the angelic agents negotiate what to do in a Nash sense. They will then choose to behave in a way that is very close to rational expectations. This means that optimal expectations are rational, even though the model is not known and one has the documented psychological biases. In other words, rational expectations can be a useful approximation even outside the usual core of assumptions.