Comment by washedup | Hacker News Reader

Comment by washedup | Hacker News Reader

Parent Full thread

We really need a way to simulate how the economy will react under certain policies. The current model is based on one hypothesis about how the economy works, when such policies may have been only tested once or twice before, AND when the test conditions were different in a lot of ways. Because of the complex nature of economic systems, prediction is extremely difficult.

We have yet to see if the monetary policies being used today will be a success or not. Many people point to signs of weakness and cracks in the economic foundation. Until interest rates are brought back to 2007 levels, we cannot say that the economy is stable. What happens to high levels of debt when interest rates begin to go up? People begin to default, credit is restricted, and spending drops. It will certainly bring higher volatility across the entire global economy, and the FED is nervous about how much.

While MIT might be today's most influential powerhouse of economic thinkers, we still don't know how successful they were at applying theory to practice. The debate about Keynes is still up in the air.

I don't see any fundamental difference between what your call simulation and what economists call models. The only difference is that your believe in the possibility of models so accurate that one could easily verify that they give correct predictions, as opposed to existing models which are subject to continual debate.

The reason it is so hard to stimulate the economy is that it represents the sum total of a large part of all human activity. So for example to test the impact of a particular kind of fiscal stimulus, you would have to accurately model the response of individuals to their new circumstances, as well as the response of markets (which involve individuals with conscious knowledge of the new policy, and thus their own beliefs about its effects). It is this last factor that makes the economy so hard to stimulate. You are not only simulating actions, but beliefs, and beliefs about beliefs, etc. It's ironic that economics is criticised for following physics to closely, when it is precisely this last factor that makes economics fundamentally unlike physics. The closest thing to modern macro if mean field theory.

I mean computer simulation. Something that you can test many, many times that accurately reflects (or at least within reason) the affects of tweaking the parameters. I understand it's a difficult problem, but there has been some success in this field, and with cheap computer power it becomes easier.

Check out this paper: http://arxiv.org/ftp/arxiv/papers/1412/1412.6924.pdf

Economic models are computer simulations!

The difference between conventional models and the kind of simulation you would like to see lies in the kind of assumptions that are made about agents and the environment they operate in. What they have in common is that both try to simulate a world in which individuals act in response to their environment.

The simulation in your paper might be more complex, but that in no way guarantees not faithfulness to reality.

I would suggest your check out mainstream micro and macro, eg a first year graduate textbook. Then you will see that the "representative agent" models of economics are really just simulations based on a specific set of assumptions.

All computer simulations are models, but not all models are computer simulations. What's taught in graduate school are text-book models which are descriptive at best. I am arguing that more thought should be given to trying to develop computer simulated economies now that computing power is much cheaper. The paper I linked to you shows initial success is at least being able to describe how an economy behaves. It's worth exploring, especially when we have huge test data sets in the form of online gaming economies.

In graduate-level economics (got my masters) there is very little connection between the micro and macro world. Macro economic behaviors are emergent properties of the underlying agents who are guided by a set of incentives, and graduate economic programs generally teach a fractured system. You either apply steady-state models to the macro world or you run behavioral experiments. There is (or there was in 2010, anyway) very little emphasis on modeling both systems as one.

I'm a theoretic physicist who does computer simulations of laser plasma interactions. I'm just curious, how much of a scale difference (orders of magnitude) exist between your micro and macro worlds? We deal with femtosecond (1e-15 sec) and nanosecond (1e-9 sec) dynamics by using different simulations and feeding the short time sim results into the nanosecond ones. Is such a thing possible? I'd think doing something like that would be obvious, no?

> ...you would have to accurately model the response of individuals...

but macro and micro are two separate disciplines within economics because of the belief that large scale economic activity can be modeled without modeling the individual agent. it's more akin to newtonian physics where the model breaks down at small scales.

but i agree with your main point about why it's hard to model -- basically humans are wiley creatures who defy predictability, and observation itself creates a feedback loop that changes the model =)

Mathematical descriptions of large-scale static properties like equilibria are patently not the same as, say, dynamic discrete-time simulations with evolving state.

I don't think we have anything satisfyingly like the latter for economics, but to claim that there is no fundamental difference is just inaccurate.

DSGE models are precisely "dynamic discrete-time simulations with evolving state" and they are the workhorse of modern macro.

You should take a look at the Open Source Policy Center: http://www.ospc.org/ We just released an application for predicting the effects of changes to the tax code.

https://github.com/OpenSourcePolicyCenter/webapp-public

http://www.debtdeflation.com/blogs/minsky/

May be of interest.

We also could use a way to simulate how the various stock and commodity markets will perform in the future.

This is just the same problem, looked at from a different angle.

Reply on news.ycombinator.com