The only thing we can be sure of about the future is that it will be absolutely fantastic. So if what I say now seems to you to be very reasonable, then I have failed completely. Only if what I tell you appears absolutely unbelievable, have we any chance of visualizing the future as it really will happen.

- Arthur C. Clarke (h/t Brin)

Wednesday, January 7, 2009

Can Computation Help Solve the Economic Crisis?


Thanks to "Tidal" for pointing out the fascinating essay by a similar title, CAN SCIENCE HELP SOLVE THE ECONOMIC CRISIS? by Brown, Kauffman, Palmrose and Smolin. I mostly agree with their perspective (I have a few quibbles about what "equilibrium" means) enthusiastically, and I appreciate the kick it gave me to express what I have come up with so far myself.

As I admitted in the discussion on a recent thread here, I have come to the conclusion that a workable, resource-aware economic theory is absolutely indispensable in the current circumstances. Unfortunately, we don't really have one, so in our new circumstances, coming up with one is about the most important thing to do.

One thing about complex systems is that you don't ever completely understand them; you just get to know them. The way you get to know things about the economy is exactly the way you go about getting to know things about oceanography, glaciology, geophysics/planetary physics and astrophysics, and to some extent meteorology, among many other disciplines I don't have much of a feel for.

That is, you build a complex artifact that behaves similarly to the complex system in nature that is of interest, and you study it as a proxy for the unique real system, constantly seeking to improve its resemblance to the real system.

In other words, you build a science around computer modeling.

Actually, the best thing to do is to build a whole lot of computer models, and have them compete to give you the best explanation.

Climatological aside: Note that nobody has created an otherwise useful dynamic model of the atmosphere that has the small sensitivity that many people are urging you to bank on. I don't know if this is for want of trying. It seems, actually, like something very much worth trying, so it surprises me that nobody wants to take it on seriously.

So how do you build a useful computer model of economics?

Well, let's consider how you build a useful model of climate.

Is there some code that says "here is the El Nino kicking in every few years?" "here is the PDO" "time to add a cold front"? No, you let the properties emerge from the low level description: here is what a parcel of air will do, in this environment, heated in this way, pushed that. Then you put all the parcels together and watch them behave. Once they settle down and act reasonably, the process couples back, and you start learning things from the modeling process. The 3 degrees C global warming stuff is a byproduct. It is not the main output of the system; the system is the partnership of minds and complex simulations; the output is understanding about the real system.

There are models of the atmosphere with much smaller numbers of parameters, and they can be didactically useful and useful in coming up with hypotheses. The general equilibrium models of conventional economics are similar to those. They are like the undergrad matlab exercises in meteorology classes. Except that they seem carefully designed to be anti-scalable; if I recall correctly, computation cost for economic equilibrium models actually scales exponentially with the number of free variables!

What we want is a huge system where the computational cost of each individual event is very small and the scaling is nearly linear. And we want huge amounts of observational data to calibrate against them. And we want to build lots of models. Lots and lots.

The analogy to a continuum model for economics is an agent model. The simulation thus needs make no assumptions of perfect information, price equilibrium, discount rates, anything. No approximation for the whole system is necessary and it seems likely that none is even possible. What we can model is how individuals make decisions in certain circumstances and see the large scale economic activities that emerge.

I've met a few people who both understand this much and indeed are pursuing it. Yet they are still hypnotized by the idiot passivity of the "neoclassicists" (H/T Erik Conway) and thus, think of economics as a branch of pure science,.

It is anything but. Economics is the most applied of sciences.

Of course, for almost thirty years we have been assured that everything would go just marvelously if we refrained from designing the economic marketplace at all. The result was, well, we'll see. Everyone agrees that the times for hands off is over.

Nowadays, we don't need to just understand economics as a matter of some urgency, we need to apply the understanding that we acquire. Computational science applied diligently and competently is the only tool that is up to the job.

I say this even though, of course, computation was at least an accelerating and exacerbating factor in the current mess.

The purpose of the computations is to represent how humans, acting individually and collectively as economic actors, make decisions, and to design the system to reward benign behaviors and increase the cost of behaviors that might be harmful, even sociopathic. Such experiments performed, appropriate incentives and disincentives can be written into law after they have been written into a particularly appealing model.

The idea of achieving such an option by the sort of deal-cutting and difference-splitting that normally is how laws are put into effect will fail. The entire package has to be designed coherently and implemented both urgently and fastidiously. The response of the real system has to be monitored carefully, and feed back to tune the system.

People need, essentially, the maximum freedom that is consistent with a sustainable totality. That is a hard problem.

It also has to get past the politicians intact. Also hard.

A lot of people want very much for it to be an easy problem, but it ain't.

Economics is about to begin to undergo a transition very much like the one climatology is still struggling with. Suddenly, it is consequential, and as a consequence of its sudden consequence, it is not up to the task. It's a matter that should engage whatever creativity and intelligence that can be mustered.

Computer generated glowing fossil Created by R Neil Marshman in 2003 - using a fractal generating program and then enhancing it with Paintshop to create a glowing fossil image. Released under GNU GFDL 1.2.
See Wikipedia for details.

Update: My enthusiasm for the Edge article has waned upon reading the commentary to the article, though I'm still glad of the incentive to state my beliefs about this.

Of course, different people have different ideas of what the "crisis" actually amounts to. They seem to be talking not about what I'd call the economic crisis but merely about the financial crisis that hastened the inevitable economic crisis.

I am looking at a longer term than these people are. In my view the economic crisis is about reconciling the creative power of the marketplace and the desire for prosperity with the constraints of sustaining the planet as a suitable habitat for life indefinitely.

Regarding the short run, I don't know whether a theory of economics can properly regulate the financial sector without substantially diminishing it (my own preference). Anyway I very much doubt that such a theory could emerge in time to help with the present tangle. And in any case, finance is just the froth; I am interested in the cappuccino. The fact that some people like too much froth is neither here nor there. As far as I am concerned, the crisis is reconciling economic policy with sustainability with political viability.

Second, they refer to Eric Weinstein who in his comments and on his website refers to some advance in economic theory involving differential geometry. I don't doubt that a better general model could emerge from some mathematical legerdemeain, but better is not necessarily good. Climatology teaches us the limits of general theories of messy systems. What I'm advocating here is a pragmatic approach that lets go of the idea that economics is a discipline that is subject to a eureka-style generalization at all.

Regarding the ambitions of this posting, I don't really know on whose behalf I'm being ambitious. It's hard for me to see a scenario where I actually end up working on such a project. I'd be willing to participate if I could see how, but I have shown no signs of developing the political skill or political capital to start such a huge ball rolling.

I am hoping someone else reads this who can pull it off. (Youth would be a tremendous advantage here.) This is also a long-term project, and its fruits probably won't emerge for decades.

I have ideas sometimes. Some of them are good sometimes. I am not too fussy about taking credit for them either, so just go for it if you can. A hat tip in my direction along the way would please me but is in no way required.

What I am saying is that, at least in part, this is what a realistic science of economics would look like.


20 comments:

gravityloss said...

There's the issue that the investors are aware of such research and try to use it to their advantage, hence changing the equation.

Ie the model results affect the behaviour.

That's one additional problem.

In theory you could even account for this feedback loop (from control theory). Ie would this economic theory be valid in case it would not only make measurements and predictions but also be used for attempted profit making, changing the system.

But then, people would even try to take that system into account and build something beyond that.

It's a never-ending problem.

As long as people want to make more profit by having different expectations of the future than the rest...

In a sense, betting with non-fixed coefficients is a good model for this. How could you make a scientific model for betting wins?
And if the people betting would use this model (that would show possible places to make profit by investing differently), would the model be valid anymore?

tidal said...

Well, that is quite an ambitious post!

I'll try to return with something more substantive later, but I thought I would add the following now, which I think is apropos.

The Jeffrey Sachs quote that you pulled from the edge article appears to come from a longer quote (which in turn comes from a conference discussion at The Earth Insitute at Columbia University, amongst Sachs, George Soros and Nouriel Roubini. I have not listened to this.):
"Now what I know about our training since the early 1980s [...] the way we train people to think [...] in mainstream economics and in mainstream politics, has left them almost unable anymore to distinguish the surface from the underlying reality. Not only was it the age of Reagan and the beginning of market fundamentalism that came in the early 80s, and the rational expectations revolution and all the rest, but a fundamental break in how we actually train our students to think [...] the way we see it in our universities.

Because the new kind of economic modeling for 25 years, the one that won all the Nobel Prizes said: you don't have to understand the deep picture, you have to look at the prices you see on the surface and infer the deep picture from that. In other words, the surface tells you the depths in fact because the surface and the depths couldn't be two different things. You don't really have to know underlying mechanisms in the economy because the prices reflect the underlying mechanism. In fact you fit the deep model, if there is one, by looking at the surface and then inferring what the deep model must be, that's literally how students are trained. They're not trained to be skeptical, they're trained to fill in parameters of a system which assumes from the start that what you see on the surface is what you really have. Whereas what fundamental proper skepticism of the Popperian sort [...] and also what the logic of economic bubbles shows, and what the logic of biophysical bubbles show, is what you see on the surface can completely hide and obscure what's happening below.

And it's our job as scientists, and I would say as responsible citizens, to understand the deep points and therefore help to rectify the discrepancy and keep ourselves away from trouble."

Michael Tobis said...

Ah, so neoclassical economics is a branch of seismology...

It's sort of amusing. I read your submission immediately after a half hour conversation with an exploration seismologist. They try to decide where to drill for oil based on evidence at the surface only. But to do that, they construct a model of what is under the surface.

I'm not sure I agree with Sachs' diagnosis, or perhaps I odn't understand it.

I think what the economists do wrong is better captured by the Brown et al essay. As explained there, economists develop a detailed theory that depends on assumptions that are always not exactly true and in some important cases not even approximately true.

Crucially, of course, as ecological economists are always complaining, sources and sinks of actual physical mass, and the viability of biological resources are casually dismissed. This approximation is increasingly invalid. The problem is starting to bite, and this will only get worse.

bi -- International Journal of Inactivism said...

gravityloss makes an interesting point. Can we say that such 'recursive' behaviour is partly what led to this current huge hiccup in the first place? Lehman Brothers invested in insecure "securities" backed by subprime mortgages, due no doubt to borked predictions of the mortgages' values.

Alas, it's probably not very feasible to ban all financial products with unknown future worth (though it does sound attractive -- I mean, why can't people simply buy and sell real stuff, for goodness sake...).

Interesting times are ahead, definitely.

Michael Tobis said...

I definitely don't think the financial sector accounting for 15% of GDP makes any damned sense.

Would it help to institute 100% capital gains tax for trades made in a week, scaling back smoothly to say 50% in two years, 0% in five years?

I don't think this is my idea but I don't remember where I heard it.

gravityloss said...

Just a note that the idea of the problem of the system being aware of the modeling is not my idea, but one of a friend of mine (or I heard it from him) who went (or was he in already back then, can't remember) to study economics (kinda, at a technical university, where it's probably very very different from pure economics).

Another friend on the other hand is a physicist, but has studied some economics basics, and had to say how, like Michael has said, a number of the assumptions don't hold (ie a great number of independent rational actors on the market)...

I don't really know if the guys with economics degrees are really that intelligent in a logical rational sceptical sort of way, since I don't know that many. Their lifestyle, taste and objectives in life seem so different from mine that we don't really meet anywhere. (And the one I do know is very intelligent but he is sort of an exception anyway.)

American Psycho is a nice parody of the subject.

tidal said...

The "actors changing their responses in response to the research" is really just a manifestation of what I meant earlier when I referenced the realclimate comment that stated "there are no unchanging underlying relationships between economic quantities".

Ok, so there might not be any gold standard, inviolate f=m*a or PV=nRT analogues within the economic models. Maybe.

But I don't think that undermines what Michael is suggesting. In fact, trying to "force" similar behavioural "laws" into the fundamental assumptions of neoclassical economics could well be a large part of the problem. Computational/empirical modelling might be far more instructive about how the behaviours actually manifest themselves, and might not be ideologically biased.

Certainly there ARE analogues in physics, etc., where there are non-linearities or changing behaviours at extreme conditions. So I don't see how that could not be modelled if empirical economic observations suggested that's what is occurring.

One thing I am curious on from mt are his thoughts on what Paul Romer said in response to "the edge" article - referring, I think, to the work on complex systems (applied to economics) by W. Brian Arthur, et al, at the Santa Fe Institute.

I remember being vaguely intrigued by how this all sounded, way back when I read Wired last century, and it sounds like Romer is underwhelmed with what they have eventually delivered. (Although, frankly, I'm underwhelmed by what Romer delivered - which was largely to extend Solow's economic growth theory, so that technology and innovation were given even more pre-eminence, and any material or energy inputs/outputs/constraints were further marginalized/dismissed.)

Michael Tobis said...

"actors changing their responses in response to the research" does seem like an emergent bug in mathematically driven trading schemes. If enough people trade that way, the software will mostly be about predicting the outcome of other software.

That isn't what I'm talking about here.

I admit that I have never properly understood "finance" although the way my brain is wired I feel I ought to be able to get it. That in turn just makes me suspicious of the whole thing.

And perhaps modeling economics without modeling finance is a fool's errand. You'll note though that conventional macroeconomics doesn't seem very interested in it.

What I'm after, though, is figuring out how to keep people happy while constraining their aggregate impact. That seems to be the right definition of our core problem, and I don't see a lot of people thinking clearly about that.

Present company excepted, of course.

Eric Hacker said...

The problem with complex models is, well, uh...., the complexity. In "figuring out how to keep people happy while constraining their aggregate impact." the first step is to try to get them to simplify their lives. Relocalization, Think Globally - Act Locally, Transition Towns, etc.

Sure, work on better ecological economical models too. However, complexity is the beast that has caused all this risk and chaos. I really don't think modeling it better will help humans make much better decisions about how to live. There is already plenty of science to back up what needs to be done and much historical religious thought as well.

Simplify and reunite our spirit with nature.

Michael Tobis said...

The only way to simplify ourselves out of the present mess is by cutting our population 80%, unfortunately.

Individual actions are well and good, but as Gore said in his Noble lecture, and as Obama sed about his lightbulbs, they aren't enough. Not even close.

John Mashey said...

Economics is an important part of all this, but economists vary widely, and shouldn't all be tarred with the same brush.

At the risk of sounding like a broken record, I strongly recommend:

Robert Ayres @ ASPO 2005.

Ayres and his colleague Benjamin Warr have done interesting work modeling the effect of energy (or more precisely, work = energy * efficiency) on economic growth, i.e., they think that energy and resources actually matter to an economy. They think the thing called "Total Factor Productivity" (TFP) is actually ~work, with a little help in last few decades from computing. The last page of Ayres' paper models US GDP as a function of efficiency.

Ayres and Warr have a substantive book coming out in April - get your library to order one.

I'd also recommend reading Vaclav Smil.

Finally, I'd recommend Charlie Hall, see especially ERROI Balloon Chart at TheOilDrum.

It seems a fundamental issue that the estimated growth rates for the next 100 years are based on past experience, and past experience is based on $20-30/bbl oil and other cheap fossil energy (see Hall's EROI chart).

If the "Solow residual" or "technology progress" or TFP is *really* work (in Ayres/Warr sense), there is no earthly way there is a nice, several %/year GDP growth indefinitely... but that assumption seems built into most of the climate/economics studies.

I'm no economist (although I've studied Stern, Nordhaus' DICE, the MIT models,and some others), and I've asked around on economics blogs, but have yet to get a great answer explaining why Ayres/Warr would be wrong.

This makes me nervous...

Michael Tobis said...

John, I am sure there are smart people who think about economics for a living. I apologize for conveying otherwise. Many of the most quotes about economists that I feature are made by other economists.

I am confident that the things economists tend to think of as macroeconomic whole systems computational models are useless artifacts, not even good epicycle gizmos.

I suspect that this is surmountable to the point where such models might be useful. The idea is that they are built up from a large of collaborating simple and mostly similar models. There are people around Chicago pursuing this.

Given a sufficiently useful modeling strategy it may be possible to develop incentive systems such that what people are rewarded for is realigned with the collective interest.

This is one of the main challenges for modern civilization, and it seems many people don't quite get it. (I know, quelle surprise!)

I second your recommendation of Ayres and of Hall and I will look into Smil. Thanks!

ac said...

We can do millions of particles in MD, there's no reason we can't do millions of agents in an economy.

The hard part is writing coming up with the equivalent of Newton's equation for your individual actors: as Michael puts it: "how individuals make decisions in certain circumstances". I think this might be what some folks call a 'utility function'.

Maybe that's not as hard as then coming up with a set of constitutive relations (how does the current system state feed into my utility fuction) that result in something resembling an economy.

I think forget about modern capitalist economies to begin with. Start with a simple idealised economy - the simplest possible system that could conceivable be called an economy. Move on to simple pre-captilalist economies. Easter Island comes to mind.

Yeah, I've thought about this before. It's a fascinating problem, and a thrilling challenge. To achieve a model in which economic behaviour emerges from simple laws would be a grand achievement. But... I don't see much practical use, or many applications. In particular I don't see how it would solve our current sustainability predicament.

The real system is too chaotic. Validation would be a nightmare. And capturing innovation and learning in the model seems like a lifes work in itself.

Eric Hacker said...

"The only way to simplify ourselves out of the present mess is by cutting our population 80%, unfortunately. "

I don't deny that at all. I'd like know how modeling anything any better would produce different results. I think the models we've gotten good at so far, say for global climate or fishery populations, pretty much indicate that a massive human population decline is necessary. If we can't learn from these models, how are better models going to teach us?

I'm playing a bit of a devil's advocate here. I did some modeling as an undergrad many years ago and even completed the annual collegiate competition that they had at the time. I beleive that modeling can illuminate useful information.

I don't understand what modeling can tell us now that we don't already know. The financial markets are, according to eco-eco, chained to nature. Energy makes the world go round, not money, and with a declining global EROEI, the financial bull market run hit the end of its chain and got yanked back.

How are better models going to either open the door to continued overconsumption or provide better information to encourage change?

Peter T said...

Hmm. Language - which is the medium through which we construct the many intersecting webs that make up an "economy" or a "society" is at least a 5 layer parallel network of networks. We cannot compute that yet. The climate, by comparison, is several orders of magnitude less complex, being constrained by relatively simple physical laws. This looks like a good approach, IF one accepted that it could only illuminate small parts of the totality.

Tim Worstall said...

This is fascinating. The assumption seems to be that no economist has ever thought about this problem before. That it needs a flash of insight from outside the field to put everyone straight after the last 3 centuries of thinking.

Problem is, all sorts of people run all sorts of models: every central bank does for example, every government.

We've even had highly respected economists (Hayek comes to mind) who state that the economy itself is so complex that the only model we can usefully use to model the economy is the economy itself.

"Just model it!" as a solution does seem to involve a certain amount of hubris.

Michael Tobis said...

Tim, do economists recognize the distinction between modeling and simulation?

I am suggesting simulation, not mere modeling.

I don't believe economic models will ever achieve predictive skill even comparable to the limited abilities of the best (simulating) climate models. But I do think we have reached the point where the general equilibrium models they keep proposing to use up our supercomputers have no utility at all.

I admit, there is hubris in criticizing a field in which one is inexpert. We in climate science are subject to it all the time.

Many believe that such criticism is useless as well as rude. I don't think that is ineviatbly the case.

Admittedly, it is difficult to be politic when one sees the landscape only from a great distance, and our critics take very little care, so I hope to take that as an object lesson. But the suspicions of economics coming from the physical sciences and especially the earth and biological sciences are real enough, and I think it's reasonable to voice them.

maninabarrel said...

Michael

One key difference is that every month, quarter, year, Governments release statistics on a number of economic aggregates (gdp, growth, inflation, unemployemnt, balance of trade, inventory levels etc etc etc). Most of nthese are qualkity-controlled. Any informed observer can make up her pown mind. Media pundits and professional economists can make a determination as to what is happening. On a coungtry-level basis.

Compare the world of climate science....4 unofficial metrics for a proxy called global temperature anomaly with unclarified quality-control procedures behind them; a few random metrics on Arctic ice coverage crop up on blogs but that is about it and no one says why this metric shoul;
d matter apart from making us grieve for polar bears; and every so often an observatory in Hawaii tells us about CO2 levels...and you claim that economics is modelled badly?

Economics has a lot of data which it tries to model. Climate science seems to lack data. Is that a fair assessment?

maninabarrel said...

would you like some tuition in economics?

Michael Tobis said...

"Climate science seems to lack data. Is that a fair assessment?"

No, not unless there are remote sensing devices in orbit that track financial transactions.

Lack of data is hardly chief among our problems, though there are indeed places where more data would be useful. Making use of the vast data we have is also a big issue.

Also, we have a strong basis in physics...