Reading Simon's defence of rational expectations made me wonder: what's the empirical evidence here?

One obvious data source is the Bank's survey of households attitudes to inflation. Every three months it asks people what they think the inflation rate has been in the last 12 months, and what they think it'll be in the next 12. My chart shows the median answers to these two questions.

It's clear that there's a close link between the two; the simple correlation coefficient since data began in 1999 is 0.88. Regressing inflation expectations upon current inflation gives us the equation: 0.87 + 0.6 x past inflation. Whilst this might be consistent with some rational expectations, it's also consistent with a simple rule of thumb: expect inflation to be what it is now, but adjust down if inflation is unusually high, and up if it's unusually low.

This in turn is consistent with Paul Ormerod's claim (pdf) that we should model behaviour as if agents used rules of thumb rather than full maximization.

So, is Simon wrong? Not necessarily. Look what happened in the recession. Inflation expectations then were lower than you'd expect from this rule of thumb. In exceptional circumstances, then - by which I mean the cases of most interest - agents form expectations in some other way.

How? Rational expectations are certainly one possibility. Another (non-exclusive) possibility is that, as Chris Carroll has described, expectations spread in much the same way diseases do: if enough people have (say) low expectations for inflation, they'll infect others.

This leaves me in an awkward position. On the one hand, agents sometimes (often) seem to have what Simon calls "naive" expectations. But on the other, they don't always have them.

Which brings me to something I've said before - that we should think not just about models, but mechanisms. We should ask: what is the mechanism by which inflation expectations are formed? It is perhaps reasonable to suspect that in ordinary times, the mechanism is a backward-looking rule of thumb. After all, in stable times, folk have better things to do than think about inflation. In abnormal times, however - such as recessions or high inflation - the costs of being wrong about inflation become higher, which leads people to abandon rules of thumb in favour of something else.

Now, this is much easier said than modeled; it leads us into Markov-switching models and other stuff I don't understand. And it might not be much use for forecasting purposes: can we tell in advance whether the expectations formation process will shift? Maybe the real world is more complicated - nay, complex - than any single model would suggest.