Arnold Kling  

Colander on Complexity

John Green on Freedom... War: What Is It Here For?...

Since I brought up the topic, a commenter pointed me to the Wikipedia article on complexity in economics, and that in turn referred to David Colander.

[complexity theory] is highly mathematical, and, as I stated above, accepts the need for simplification. But it argues that the mathematics needed to simplify economics often involves non-linear dynamic models that have no deterministic solution.

He concludes,

The methodology of complexity economics is quite consistent with the methodology of standard economics and, as such, does not fit well with many heterodox traditions. Still, many heterodox economists have had a complexity vision, and thus complexity economics can be seen as a melding of heterodox vision with standard scientific methods, which is only now coming of age.

It strikes me that there is a deep question of what constitutes knowledge in economics. I do not know the answer, but my instinct is that the contribution of computer simulations of complex processes will not amount to much.

Comments and Sharing

CATEGORIES: Economic Methods

COMMENTS (14 to date)
Dave writes:

You can try reading The Origin of Wealth to see some examples of how complexity science has been applied to economics so far.

Jason Collins writes:

Your instincts may be right, but I would consider that there have already been some substantial contributions to economics using computer simulations of complex systems (noting that complex does not necessarily mean complicated). The works of Thomas Schelling and Robert Axelrod spring to mind.

I also expect that many other emergent properties of economies will come to be demonstrated in simulations of complex processes, sheding light on what factors may be driving those outcomes (in fact, some of this has been done, it just doesn't have traction in the journals). This probably won't require particularly complicated models, which in many ways is what complexity is about - complex outcomes from simple foundations.

Moving on from major contributions, however, computer simulations of complex processes could be very useful in economics education. As well as giving students nice, neat equations to solve for equilibrium, they could be given complex systems to simulate. It would give them a picture of a complex, emergent world rather than the mechanistic image they are often encouraged to develop.

Most of us who have done a fair amount of simulation modeling understand that nothing comes out of a model that isn't first put in. You have to put the rabbit into the hat somehow.

Computer simulations are often interesting and instructive, but the notion that the simulation is going to somehow reveal previously unknown, unconsidered truth is probably mistaken. At least, that's what my experience with simulation modeling shows.

The behavior of complex systems of matter and energy can be modeled often enough, because matter and energy appear to behave according to fixed relationships and complexes of variables and constants --- at least at the macro level of human experience. But humans don't. Humans can choose.

To my knowledge, none of the so-called laws of economics approach anything even remotely like the law of gravity. There appear to be no constants in economic science, notwithstanding the propositions of folks like Arthur Okun. Consequently, computer simulations of human behavior are not at all likely to be robust.

eccdogg writes:

Agree with David Kendall,

I do alot of finacial modeling and find that additional complexity is usually the absolutely wrong approach.

Usually you end up with too many parameters that you don't fully understand and too many possible outcomes given those parameters to make useful predictions or decisions.

The best approach is a very simple model that captures the core features of the problem. That model is certainly wrong in some respects some that you care about and some that you don't. However that leads to the second key to modeling complex processes, know your model's blind spots and don't put too much faith in it.

This has always been the approach I took to economics and what I believe what the strength of the field is. A few core ideas that are largely correct, but that are not a full explanation of the world that we live in.

Doommd writes:

Computer Simulation certainly is a value additive proposition within economics. It is not always (or often) true that you get out what you put in. Simulations often are not created as an economics problem [analytical eqns] but as stochastic decision trees or regression models, i.e. empirical methods. In other words, you can vary the input of a system knowing the empirical model within it is accurate to a certain degree and view the magnitude [or sign] differentials in the output. The human hand simply does not compete.
Moreover, to simulate economics is not conceptually very complex; defining a sufficiently sized representative market though and walking in with an all inclusive set of variables, thats a different story.

I am hoping you won't be so quick to turn away simulation as a tool, but rather place blame on the infant computer age for not learning how best to apply it yet.

Jeremy, Alabama writes:

I'm very surprised nobody has tried to apply fractal techniques to economic models.

@eccdogg. I agree with you, too. I have built several computer simulation models that end up with so many parameters that they are useless, really. No way to calibrate the model.

It's a bit like econometric techniques. If you have a scatter diagram, even in just two space, one fit a model that passes through each and every point. So what?

@Doommd, you wrote "It is not always (or often) true that you get out what you put in."

I disagree, almost by definition. No model behaves any differently than the constraints of its parametrization and structure. Using a stochastic structure doesn't change that; just makes the whole deal stochastic. But even then, the stochastic nature of the model is fully determined by the distributions the modeler allows.

It is true that the human mind often doesn't understand what the consequences of a particular structure and parametrization will be, but that's a different story, no?

I haven't given up on computer simulation at all. I think it's fascinating, really. But I've built too many models to think the modeling of any kind is magical.

fundamentalist writes:

I agree with Kendall. I have done some econometrics in business and I have found that I have two extremes to deal with: one group thinks modeling can solve any problem and forecast anything; the other group thinks it's all a waste of time.

I like what I learned about structural equation modeling. Most software used for SEM includes statistical measures to guide the modeler in adjusting his model, but an author of a text on SEM wrote that those measures are rarely useful. Theory is a much better guide to model adjustments.

As I have written before, if computers could sift through the data and give us good theory then neural networks programs would have solved all debate in economics decades ago. But that hasn't happened. Why? Because NN programs produce dozens of different models all equally good but which produce very different forecasts.

Watch the weather channel and notice how their models forecast the movements of a hurricane. They have dozens of models predicting very different paths and the use them all to create a forecast of possible paths.

The weather is much simpler than the economy.

Jason Collins writes:

Reading the comments above (@David L Kendall, @eccdogg), most appear to be addressing computer simulations in general rather than in the area of complexity. The focus of complexity research is generally on simple agents interacting in simple ways from which the complex behaviour emerges.

Take Schelling's model of segregation. Simple preferences as to neighbours resulted in segregated neighbourhoods. Axelrod's research into the evolution of cooperation also involved agents with simple decision rules. Complexity research is not about increasing the number of equations or parameters.

On the point that there are no economics laws that are like the law of gravity, that is one of the features of economics that complexity research seeks to address. Instead of looking at perfect calculating machines, complexity research usually involves simple decision rules that could be argued to be closer to actual decision making processes.

@Jason Collins, Yes, that's what complexity modeling is supposed to be about. Problem is, humans don't make choices according to a simple set of rules. Propensities, yes, but rules, no.

Jason Collins writes:

@David L Kendall, I think I am on your page now. Would you apply that critique across all economic modelling?

eccdogg writes:

"[complexity theory] is highly mathematical, and, as I stated above, accepts the need for simplification. But it argues that the mathematics needed to simplify economics often involves non-linear dynamic models that have no deterministic solution."

This is the text I was working off. I don't believe there is much to be accomplished in economics by using "highly mathematical" modeling that involves "non-linear dynamic models that have no determinisic solution".

If that is what complexity ecomonomics is about I don't hold out alot of hope.

Most of the truly great economic theories can be explained using almost no math.

Also I do not view simulation modeling as "Scientific". Science is empiricism, making predictions about the real world and then testing them against the data not exploring the consequences of your computer created world.

And yes much of "standard" economics does not fair well against this critique either.

Troy Camplin writes:

Understanding what complexity science says, whether you use the math or not, is greatly beneficial, I think. From the models one can develop general principles, and then one can compare them to what one sees in the world. The complexity models are exactly that -- models -- and that is their strength. They don't claim to be accurate representations of reality, but rather models. One can set up parameters and see if they match something about reality. One can add and subtract and, thus, see the outcomes. Such modeling acts as something in between pure theory and hypothesis creation with empirical investigation. I see that as nothing but beneficial, especially for a complex process like an economy.

I disagree with the author about heterodox economics. Complexity science disproves neoclassicism and lends great support for Austrian economics. There is no place for equilibrium in complex systems, which are in far-from-equilibrium states, which is the realm of greatest creativity. If complexity science draws our attention to nothing else than far-from-equilibrium states, it will have greatly benefited economics as a science.

I wish people would read Stuart Kauffman and Hector Sabelli and really understand what they are saying and use their ideas in economics (both authors try, but really real economists need to do it).

eccdogg writes:

Let me preface this with saying that I do not buy into everything in neoclassical economics.

But, how can complexity economics disprove it?

I do not believe neoclassical economist ever believed their assumption were true, just that for a limited set of problems a theory derived by assuming they were true yielded predictions that matched reality.

The way to disprove this claim is not by saying, your assumptions are false, by using a different (also simiplified and unrealistic) set of assumption in a simulation I get a different answer.

The way to disprove neoclassical economics is to instead say your theory predicts Y but instead we see X.

Now the second step may be my new theory developed from insights of complexity economics makes a prediction that better fits reality.

Comments for this entry have been closed
Return to top