The first of these is computational irreducibility. You may be able to reduce the behaviour of a simple system to a mathematical description that provides a shortcut to predicting its future behaviour, the way a map shows that following a road gets you to a town without having to physically travel the road first. Unfortunately, for many systems, as Stephen Wolfram argues, you only know what is going to happen by faithfully reproducing the path the system takes to its end point, through simulation and observation, with no chance of getting to the final state before the system itself. It’s a bit like the map Borges describes in On Rigor in Science, where “the Map of the Empire had the size of the Empire itself and coincided with it point by point”. Not being able to reduce the economy to a computation means you can’t predict it using analytical methods, but economics requires that you can.
The second characteristic property is emergence. Emergent phenomena occur when the overall effect of individuals’ actions is qualitatively different from what each of the individuals are doing. You cannot anticipate the outcome for the whole system on the basis of the actions of its individual members because the large system will show properties its individual members do not have. For example, some people pushing others in a crowd may lead to nothing or it may lead to a stampede with people getting crushed, despite nobody wanting this or acting intentionally to produce it. Likewise no one decides to precipitate a financial crisis, and indeed at the level of the individual firms, decisions generally are made to take prudent action to avoid the costly effects of a crisis. But what is locally stable can become globally unstable.
The name for the third characteristic, non-ergodicity, comes from the German physicist Ludwig Boltzmann who defined as “ergodic” a concept in statistical mechanics whereby a single trajectory, continued long enough at constant energy, would be representative of an isolated system as a whole, from the Greek ergon energy, and odos path. The mechanical processes that drive of our physical world are ergodic, as are many biological processes. We can predict how a ball will move when struck without knowing how it got into its present position – past doesn’t matter. But the past matters in social processes and you cannot simply extrapolate it to know the future. The dynamics of a financial crisis are not reflected in the pre-crisis period for instance because financial markets are constantly innovating, so the future may look nothing like the past.
Radical uncertainty completes our quartet. It describes surprises—outcomes or events that are unanticipated, that cannot be put into a probability distribution because they are outside our list of things that might occur. Electric power, the atomic bomb, or the internet are examples from the past, and of course by definition we don’t know what the future will be. As Keynes put it, “There is no scientific basis to form any calculable probability whatever. We simply do not know.” Economists also talk about “Knightian uncertainty”, after Frank Knight, who distinguished between risk, for example gambling in a casino where we don’t know the outcome but can calculate the odds; and what he called “true uncertainty” where we can’t know everything that would be needed to calculate the odds. This in fact is the human condition. We don’t know where we are going, and we don’t know who we will be when we get there. The reality of humanity means that a mechanistic approach to economics will fail.
So is there any hope of understanding what’s happening in our irreducible, emergent, non-ergodic, radically uncertain economy? Yes, if we use methods that are more robust, that are not embedded in the standard rational expectations, optimisation mode of economics. To deal with crises, we need methods that deal with computational irreducibility; recognise emergence; allow for the fact that not even the present is reflected in the past, never mind the future; and that can deal with radical uncertainty. Agent-based modelling could be a step in the right direction.
