Some economists are trying to get macroeconomics off the hook by arguing that by their very nature crises are unpredictable. Thus David Levine aggressively argues that "our models don't just fail to predict the timing of financial crises – they say that we cannot."
There are three problems with this argument. First, it assumes what is it at question – namely whether what Levine calls "our models" are good models. Perhaps behavioral models could better predict the timing of financial crises. I will not push this argument but I do believe that current events call for a greater than normal willingness to think beyond the confines of the models that one defends.
Second, it's not true that "our models" tell us that we can never predict a financial crisis. In some cases, our models predict the exact moment that a crisis will occur and these models are perfectly consistent with, indeed require, rational expectations. It is perhaps no accident that Paul Krugman has specialized in these types of models.
Third, the word timing is misleading. Let's accept that a crisis cannot be predicted to the day or even to the year. Nevertheless, it is perfectly reasonably and fully consistent with rational expectations to predict an increased probability of a crisis.
If you play Russian Roulette with 1 bullet and 100 chambers in your pistol, I can't predict when the crisis will occur. If you play with 10 bullets, I still can't predict when the crisis will occur but I can say with certainty that the risk has increased by a factor of ten. Analogously, nothing in modern economics makes it theoretically impossible to forecast that greater leverage and higher than normal price to rental rates, to name just two possibilities, increase the probability of crisis. Nor does modern theory make it theoretically impossible to forecast that conditions are such that if a crisis does occur it will be a big one.
All of this is true even in the context of stock markets. Efficient markets theory implies that any two stocks will have similar risk-adjusted returns it does not imply that the risk of bankruptcy is the same for any two firms. It is perfectly reasonable to say that Google revenues are going to have to increase at a historically unprecedented rate or the stock will plummet. It is even consistent with efficient markets theory to predict that the probability of Google stock falling is much greater than the probability of it rising (but if it rises it will rise very far, very fast).
Thus the "we could not have predicted the crisis even in theory" argument is a weak defense–even with rational-actor, rational-expectations models there are plenty of senses in which economists could have better predicted the crisis and, although this is yet to be seen, perhaps they could and will do even better with other sorts of models.