Economics with Fewer Equations

The theme that economics is less like physics and more like biology is not new. However it is more a fringe view than a mainstream one. Even now, I would guess that the general expectation of an economic journal in publishing a paper is what mathematical basis it has and how robust the Econometrics is. Several recent things have pointed to a better direction.

• The economist article on agent based models
• The book – growing artificial societies
• The book – origin of wealth, especially the early sections
• One recent article in Bloomberg by a theoretical physicist 

The essential argument is same. We can’t draw up pretty equations to predict people’s behavior. Generally it is argued that these equations are trying to predict the behavior of only the “model” or average individual or the behavior of the collective. The implicit assumption in the usefulness of this approach is that this “average” study is genuinely the median behavior and the deviation around this median is noisy but controlled. Also it is assumed that the noise is averaging to zero and has no impact on the overall behavior of the economy. 

It is not very different from saying that a ball of steel has electrons inside it moving in all directions. However, at no point is the ball moving anywhere on account of this movement. The diverse directions of movement of the electrons are random and thus cancel each other out at the aggregated level. Similarly it is argued implicitly that in equationalizing economics, individual differences of behavior around the “mean” are random and they cancel each other out. As it turns out, more often than not, they do not.

Some systems in economics may indeed be amenable to a linear model of this sort. However most useful phenomena are non-linear and dynamic and thus do not readily lend themselves to this approach. Imposing this approach on those systems then is bound to yield erroneous forecasts. I have elsewhere argued with the example of the billiards table where the physicist refuses to predict where each ball would be after the first strike. Physics concerns itself with questions like conservation of momentum in each interaction and the inertia and friction and so on. No physicist would try to build a model of the average ball and then hope that some contained linear variation around it would be a good way to explain how the strike leads to the evolution of the table.

Conventional economics is routinely trying to do this. I don’t know how it came to be here. The book 'Origin of Wealth' offers some explanations. Historically Walras and his contemporaries were quite enamoured by the success of physics with equations and tried to use the same in their work. Since then, almost as a historical accident, economics has continued to progress in that direction. The author of origin of wealth even goes ahead and calls this a century long detour. Audacious maybe? But most likely quite accurate description of what has gone on since.

The beer game example in origin of wealth is quite illuminating. (http://en.wikipedia.org/wiki/Beer_distribution_game)

A simple trigger at the customer demand end causes all sorts of fluctuations in the supply chain although after the initial reaction, the variation in customer demand is taken out. It goes to show that in absence of perfect information and strategic gameplay between transacting parties, the supply chain can exhibit very dynamic patterns – which are far from equilibrium. The Growing artificial societies authors call it far from equilibrium economics.

The opposition to the equationalizing of economics earlier was countered with TINA. What do you propose, the proponents would ask. Since the opponents never really had much of a proposal, the conventional equilibrium economics continued. Now in computational economics, complex adaptive systems and agent based modeling, there might be a genuine alternative.

This approach can open up new areas of dynamic modeling which were intractable for analytical solutions. This can also help learn emergent phenomena which are otherwise blackbox to top down modelers.

What are the limitations?

One needs to start somewhere. Where one starts can significantly impact the answers one gets. Hence the approach is somewhat prone to curve fitting. Some intellectual discipline and robustness inducing techniques are required in this case.

The micro leading to macro is an interesting theme and a long cherished dream of economists. However, conventionally the two have stayed separate. The agent based modeling with inclusion of complexity approach can start to make this reality.