“Microfoundations” ain’t so microfounded

What’s really wrong with modern macroeconomics (or at least, one of the things wrong with it)

One of the things that I started doing when I was writing my book about economics with @TessRead was to start making a list of the things that seemed really obvious to me, but which didn’t seem to feature in the economics literature at all. One of these things, which ended up being too techie and specialist to put in the book but which still strikes me as very important, is about the whole basis of the debate over microfoundations in macroeconomics.

For those few people who claim to have better things to do with their time than keep up with such things, the “microfoundations debate” relates to a particular way of doing economics. Specifically, the idea is that if you are making a macroeconomic model, you need to be able to disaggregate it — to show, in detail, how the macro-level relationships that you’re assuming can be reconciled and made consistent with the optimising plans of a presumed rational individual agent. So it’s not on to have concepts like “wage rigidity” or “investor confidence” in your model unless they can be explained in terms of individual optimisation.

There are all sorts of arguments one way and the other on this subject, but the one that has always seemed to me to be fatal to the microfoundations project is — a lot of the time, the proposed “microfoundations” also look like they need microfoundations. For example, plenty of allegedly microfounded models will assume that corporate investment is carried out by firms which aim to maximise profits.

On the face of it, not an unreasonable starting point for modelling, but it’s hardly “microfounded”, is it? The actual investment decision is carried out by a corporate manager. He wants to maximise his own income, so he is going to do what he thinks will generate a bonus for him. This depends on his boss, who reports to the corporate board, and who therefore wants to do what he thinks will get a bonus from them. The board are responsible to the shareholders, and it might be thought that the shareholders, at least, want to maximise profits. But actually, the shareholders are several levels of abstraction away from the actual investment decisions of the company. Most shares are held by fund managers, who want to beat the market, and who therefore want companies which do things which the rest of the market will value higher at a later date.

So, the investment decision, rather than being simply based on “what will maximise profits?”, is actually based on the manager’s perception of the CEO’s perception of the board’s perception of one group of fund managers’ perception of a larger group of fund managers’ perception of what the underlying owners of the shares are willing to pay for. There are six or seven levels of principal / agent problems between an already quite intractable profit-maximisation problem, and the individual decision-making agents who are meant to carry it out.

The amount of credit in the economy is determined in roughly the same way, as loans have to be made by loan officers, who have bosses too. Government spending would also need to be microfounded in many cases, possibly as the maximising decision of a contractor’s view of a civil servant’s view of a ministers’ view of the party leader’s view of what some part of the electorate wants. In other words, hardly any part of a “microfounded” model is actually microfounded. You can just about defend the assumption of a “representative individual” to model household consumption and savings decision, but all sectors other than the household sector aren’t made up of individuals — they’re made up of complicated organisations which have all sorts of interactions, and where the one thing we know about the congeries of game-theory problems that are going on, is that they don’t end generally end up with outcomes that look like the solution of simple maximisation problems.

So, there’s no such thing as microfoundations. Any attempt to be genuinely rigorous about macroeconomics (and a lot of microeconomics too) is going to lead straight into a combinatorially insoluble knot of interactions. Everyone who actually wants to model the economy has to accept that their modelling of investment spending, at the very least, is going to be based on some kind of simplifying assumption to boil all these tiny and human interactions into a mathematically tractable composite.

And that tractable composite of all the principal and agent problems in the whole economy — what’s it going to look like? Well, it’s going to look like an “animal spirits”, from a classic Keynesian model, or one of Roger Farmer’s. We know that all the different levels of game theory interactions do sort of work out in the real economy, because investment behaviour does take place. But we also know that they’re subject to fairly wide swings and overreactions.

Another thing we can probably say is that if we think of the animal spirits factor in this way — as a proxy for the true but incomputable microfoundations — then we can guess quite a few things about its statistical properties. Which is to say — whatever the process which generates the “pure” expectations of the individuals, the animal spirits factor is the outcome of a tightly coupled and largely self-organising network. And one thing we know from the work of people like Duncan Watts is that such systems tend to have distributions of outcomes which look much more fat-tailed and prone to extreme over-corrections than the normal. Which makes sense — in investment bubbles, every level of decision-making wants to impress the level above with their optimism, while in slumps everyone wants to be ostentatious with their caution. If you think that your boss basically doesn’t want to invest, then you’ll shade downwards the estimates that you give to him. Then he shades them down a bit more when he talks to his boss, and so on and so forth, until by the time it reaches the final level of approval, even the best investment project in the world has been made to look like a three-legged dog.

I think this is why people like Roger Farmer’s and Nassim Taleb’s economics have immediately appealed to me. It’s not that they’re ignoring microfoundations, or trying to get away from the difficult business of making a model of the world that makes economics sense. It’s precisely because they’re actually providing a vision of what macroeconomics might look like if it really did start thinking about the actual system it’s trying to model, rather than proving valid theorems about imaginary people and hoping for the best.