What is The Rational Expectations Assumption For

No it is not a guide to proper grammar, and I know perfectly well that a preposition isn’t the sort of thing I’m supposed to end a sentence with.

The case for assuming rational expectations rests on two valid arguments.

1) Expectations matter a lot.*
2) If you allow people to assume whatever they want about expectations, then you allow them to explain anything and nothing and predict anything they want (see the Confidence Fairy).

This leads to the conclusion
3) If we want to use models to prediction what will probably happen, we should assume rational expectations.

The logic is not that actual agents actually have rational expectations, but rather that some discipline is needed to keep people from fiddling away until they have fit the past impressing the gullible but obtaining no ability to predict the future. The rational expectations assumption just has to be the least bad well defined rigid assumption.

My problem with this argument is the hypothesis “If we want to use models to prediction what will probably happen”. Back in the bad old days, economists would use one set of (primitive) models for short term forecasting, but not at all assume that the factors they left out (left as disturbance terms) weren’t important. This is still true of many economists.

Some economists, however, clearly think that the models can be used to determine what might happen. Indeed that if something couldn’t happen if the model were reality, then it couldn’t have happened even though it sure seems to have happened.

My view is that, if we have to make simple assumptions about expectations, then the rational expectations assumption is the least bad assumption. But if we don’t have to make such assumptions, as when we are asked if something might happen, we shouldn’t.

One fan of the rational expectations hypothesis is named O.J. Blanchard. He is a leading new Keynesian. He has recently proposed a target inflation rate of 4% not 2%. The logic is that, if we find ourselves in another deep crisis, it is better for central banks to be able to set the safe short term real interest rate to -4% rather than -2%. He doesn’t have a model which predicts that there will be another deep crisis in say 2018. He doesn’t need to. All the argument requires is that we can’t rule out another deep crisis (or put a low upper bound on the probability) and that the cost of getting caught in a liquidity trap dwarfs the added cost of 4% inflation over 2% inflation.

Over at the investment banks (including those who failed) they claim to practice “stress testing” where, even if a model suggests that a trading strategy will give high risk adjusted returns, they try to think about the lower tail — what would happen in the case of extreme events which haven’t occured in their data set but which might occur in the future.

There doesn’t seem to me to be any stress testing of policy rules based on the idea that things might happen which can’t happen in the model and haven’t happened in the data set. I think this is partly a consequence of ignoring the conditions for usefulness of the rational expectations assumption and that it has caused vast suffering.

* On 1) surprisingly many new Keynesians seem prepared to argue that we have advanced beyond Keynes, because we have figured this out. By Keynes they mean the old hydrolic Keynesian models which are very different from anything Keynes wrote. I think I understand what is going on. Krugman has been praising the IS-LM model. They want to debate Krugman. He is busy. So they sometimes treat me as a junior varsity Krugman. This is flattering, but it is not responding to my challenge to find something introduced to macro theory since Keynes which is demonstrably empirically useful. The key point is that Keynes hedged his bets, so it is hard to take him to the cleaners.