Two small points following up on my previous post on microfoundations.
1) Adopting rational expectations as the default expectations model has never meant (for me at least) ignoring the possibility of non-random expectations errors. As Lars Syll points out, the informational demands of rational expectations are very strong. However, we need to model expectations by some means. What rational expectations allows you to do is think about expectations errors in a structural way. We can think about deviations from rational expectations, just as we can think about shocks to behavioural relationships. The problem with what went before rational expectations (e.g. adaptive expectations) is that expectations errors were built in, and in most situations these built in errors were not terribly plausible.
Hopefully models of learning will eventually allow expectations errors to be analysed in a more plausible, systematic and routine way. I was interested to see that Michael Woodford, in his defence of microfoundations methodology here, nevertheless pinpoints rational expectations as a key weakness, and learning models as the way forward. As my next door neighbour in the department at Oxford works in this area (see for example Ellison, Martin & Pearlman, Joseph, 2011. "Saddlepath learning," Journal of Economic Theory, Elsevier, vol. 146(4), pages 1500-1519), I couldn’t possibly disagree. However I think it is highly unlikely that learning will negate the advances in understanding monetary policy that I referenced in my previous post.
For some variables in some situations, a baseline where expectations were formed in a naive way might be more appropriate. Some aspects of risk maybe? However inflation in a business cycle with an independent central bank is not one of these. David Glasner talks here about the “the tyrannical methodology of rational expectations”. I just do not see it that way. Rational expectations do not prevent us understanding sustained periods of deficient demand when an inflation targeting central bank hits a lower bound. Indeed they help, because with rational expectations inflation targeting prevents inflation expectations delivering the real interest rate we need, as I have argued here.
2) I talked about both rational expectations and the New Keynesian Phillips curve (NKPC) in providing the theoretical impetus to inflation targeting by independent central banks. A comment asked why I put the two together. The latter goes with the former, because rational expectations with the more traditional Phillips curve imply deviations from the natural rate are random, which is totally destructive of Keynesian theory. (If inflation at time t depends on the output gap and expected inflation at time t - rather than t+1 as in the NKPC - and the difference between actual and expected inflation is a random error because expectations are rational, then the output gap is also a random error.)
The traditional Phillips curve has always seemed to me to be an advertisement for the dangers of not doing microfoundations. It seems plausible enough, which is why it was used routinely before the rational expectations revolution. But it contains the serious flaw noted above, which almost destroyed Keynesian economics. I know this is not realistic, but imagine that Calvo (1983) ‘Staggered prices in a utility maximising framework’ Journal of Monetary Economics Vol 12 pp 383-398 had been published a decade or more earlier, as a direct response to Friedman’s 1968 presidential address. Who knows what would have happened next, but it is difficult to imagine the history of macroeconomic thought being worse as a result.