The verbiage-to-insight ratio is very high everywhere, so short, smart statements have to be noticed whenever they pop up. I found one from the great tribe of Anonymous, who posted a comment over at Noahpinion:
In macroeconomics, we don't have microfoundations to the point of the individual consumer (i.e. we don't literally model the decisions of every agent in the economy, or every particle in the system). I don't think macroeconomists want to go in that direction at all (hence the resistance to agent based modelling). What we do have is a model for the aggregate behavioural response to a policy as a function of fundamentals. An important difference between physics models and economic ones is that expectations about the future affect decisions today, and so it is important to capture this channel. It is not necessary that microfoundations be completely analogous to the micro level decision however, because we are modelling the aggregate behavioural response, not the individual level one. See for example much of Prescott's writing on interpreting the Frisch elasticity in macroeconomic models. I think a lot of the discussion here implicitly assumes that microfoundations refer to particle level interactions, they do not. Whether they are structurally invariant under the policy considered, well, that depends on the model and the policy.My main purpose is to get you to read and think about this comment, which is more subversive than perhaps its author realizes. As for my reactions, here are a few:
1. The term “microfoundations” is fundamentally misleading. What we really have are aggregate behavioral functions. Economists feel more comfortable if the functions that predict collective behavior mirror those that they are familiar with at the individual level, but since collective behavior is not derived from individual behavior, this preference has no theoretical basis.
2. It is possible that optimization assumptions that are so flawed at the individual level may work better at the collective level, in the sense of better explaining the data. I don’t think that’s the case on my planet, but I grant that the failings of this behavioral model at the individual level are not in themselves dispositive.
3. Since we are not deriving aggregate behavior from individual behavior, we are free to play with models that might not be applicable to individuals. Thus we can consider models of the formation, competition and dissolution of norms and conventions, or herd behavior at the level of the herd.
4. While a full specification of the state of the world at time 0 would enable us to predict, perhaps with error, behavior in time 1, no one is trying to do this. Instead, we have radically incomplete specifications with models that are essentially heuristic, somewhat better or worse at explanation and prediction under particular circumstances. Thus the goal of forecasting has to be scaled back. What we can do if we are really at the top of our game is generate forecasts that are conditional on a possibly large number of future circumstances which themselves cannot be forecasted. We don’t know whether a major war or natural disaster will disrupt the economy over the coming months, or whether the “mood of the market” will shift substantially, or even the full extent of exposure of the financial system to the systemic risks implicit in their various derivative instruments. We can’t put percentages on them either. All we can do, at best, is arbitrarily identify a large set of assumptions and forecast conditionally on them. (And one catch-all assumption is that none of the consequential unknown unknowns will materialize.) For this reason, forecasting may be a false goal. A more serviceable one would be to identify processes with known dynamics in as close to real time as possible. Example: I can’t forecast the effect of “fiscal consolidation” (austerity) on European growth rates, but I can possibly track the process by which falling credit and output demand on the part of the state is generating reduced income and monetary growth in the present. A lot else is going on that will affect how Europe progresses, but do I need to construct forecasts that are sensitive to it?
5 comments:
I agree. I've said before that what's generally called "microfoundations" is actually "optimizing foundations", because that's what macro got a large dose of with the so-called "microfoundations" revolution.
It's understandable why optimization would be labeled "micro", of course, but it is a little misleading.
"For this reason, forecasting may be a false goal. A more serviceable one would be to identify processes with known dynamics in as close to real time as possible."
At the INET conference Gerg Gigerenzer spoke about "What Can Economists Know?" In his speech, he discusses why optimization does not work in an uncertain world. Instead individuals uses heuristics to make decisions. In that sense, I think heuristics is a better description of the topic discussed and more reasonable goal for understanding.
I don't really understand this post.
When you say "What we have here are aggregate behavioral functions," what do you mean "what we have here"? Why have you just tossed out the idea of deriving aggregate behavior from individual behavior?
Noah, I mean this quite literally, in the sense of Anonymous. A representative agent is not an agent; it is a collectivity, as in "200 million households". If you were really a methodological individualist (strict microfoundations), you would derive that collective behavior from the behaviors of constituent individuals. We know this can't be done; it's much, much more than an adding up exercise. But the fact that it isn't being done means that there is no theoretical reason the model of collective behavior has to have any particular relationship to a preferred (or in my case, doubted) model of individual choice.
This seems so simple and straightforward to me that I fear that something important is being missed, either by you or me or both of us.
Peter, you’ll soon be getting comments pointing out that some DSGE models aren’t inhabited by representative agents, but by heterogeneous agents with different preferences, endowments, etc. So, I’d like to add another consideration to your complaint.
Agents aren’t presented with a complete list of possible futures to which they attach probabilities. As G.L.S. Shackle emphasized, people must imagine alternative scenarios, and as Frank Hahn stressed, from a slightly different vantage point, people interpret data with a theory. Now, I’m wondering how Roger Farmer’s “belief function” (let alone rational expectations) is able to capture the diversity of views about the future generated by diverse theories and the many imaginative propensities at work in the economy.
Post a Comment