Simon Wren-Lewis asks heterodox economists a question: how do you answer the question "what do consumers do if they are told that taxes are rising temporarily?" without some appeal to representative agents?
My answer is: I would start from a representative agent model (which predicts consumption smoothing) but I wouldn't stop there. On this issue, as on most other macro ones, I'd ask two further questions.
One is: do we have any reason to suspect that the representative agent perspective might be wrong? For example, some households might not have savings to run down in response to a tax rise, and might be unable to borrow; this is true for a minority (pdf).These people might be forced to cut spending.In this sense, heterogeneity matters, because models in which everyone is credit-constrained or nobody is are both wrong.
Heterogeneity also matters for firms. Geroski and Gregg's study of the 1990s recession found that most of the fall in output and employment was concentrated in a minority of firms, and Xavier Gabaix has shown how a few large firms can generate productivity fluctuations.Real business cycle theorists should make more of this.
The second question is: are there any cognitive biases which bear upon this issue?
For example, if consumption is determined by habits then agents' desire to smooth spending might (or not!) be even stronger than the conventional representative agent model predicts. But if agents have limited attention and so don't listen to the message that the tax rise is temporary (or don't believe it), then spending might fall by more.
Of course, models with both heterogenous agents and cognitive biases very quickly become fiendishly complex*. But then the economy is complex and as Wittgenstein said, a clear picture of a fuzzy thing is itself fuzzy.
But in the realm of policy advice, do we need to build such models? Isn't it better to be roughly right than precisely and elegantly wrong? As Stephane has complained, perhaps there's a cost to formal modelling - it can distract us from other ways of thinking about the economy.Worse still, in the wrong hands (which are not Simon's), it can distract us from the facts.
* I suspect one way forward is the use of agent-based computational models; this sort of thing might be the future of economics.