February 2021

Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
Blog powered by Typepad

« The Carl Menger Undergraduate Essay Contest for 2012 | Main | Living Living Economics »


Feed You can follow this conversation by subscribing to the comment feed for this post.

We usually also have expectations about inputs as well as outputs, which helps narrow down the field (unless you're a Friedmanesque positivist, of course). I agree with you that mainstream model building does this too.

Is there a reading list anywhere for someone curious about ABM?

Austrians have been deploying agent-based models going back to Menger -- by which I mean, of course, verbally-worked-out scenarios rooted in individual choice and reasonable assumptions about the goals, capabilities, knowledge, and constraints faced by acting individuals.

When it comes to MATH modeling, I think my concerns can be summarized under the headings of "DDT": Distraction, Diminishing returns, and Temptation. Distraction by the challenge of programming and tweaking parameters and statistically analyzing output, instead of thinking directly about purposeful human action. Diminishing returns, because you can't expect math to spit out results more accurate than the numbers you put in, and you can't really expect to characterize purposeful human beings in terms of a few accurate parameters -- quite the opposite of the situation in physics or engineering. Temptation to imagine that one can measure the relevant parameters, solve the equations, and engage in engineering; or, in other words, to imagine that one's clever facility with mathematically-characterized stick figures qualifies as wisdom for informing interventionist government policies.

I'm not saying that math models are completely useless, but I am highly skeptical that they can substantially augment, much less replace, verbally formulated reasoning about human action. They can serve as suggestive or illustrative Tinkertoys.

DDT, very interesting.

Puts me in mind of something Greg Lewis said when I was in his labor economics class at the U of Chicago: "Don't expect the math to do the thinking for you."

Allan -
But the way you describe Austrian ABM could be used to describe mainstream modeling just as well. Similarly, your DDT are each traps that "verbal" economists (and ABM economists for that matter) can and do easily fall into as well, all the time.

I think Peter hit on the right value-added of ABM already: avoiding the need for closure of models of emergent order.

My feeling on the value of mathematical models is that they keep track of everything that's going on that verbal renditions simply can't keep track of (at least I'm not smart enough to keep track of it as a reader or a writer of prose, and when people claim to be that smart I have my doubts).

The value added of verbal exposition is in dealing with more conceptual problems.

And definitely - if you try any of these without a strong intuition you're not going to get anywhere.

Richard Freeman once told me he often starts with a very complicated model, and as he refines it it gets simpler and simpler. He would develop an intuition of where the math was pointing so that he could strip the model down to its really essential elements. The more complicated the model, he said, the less sure a grasp he has on it.

Daniel, I don't see how DD&T as I presented them could possibly be traps for verbal reasoning. I referred very specifically to the distraction of math tweaking, the diminishing returns on math, the technocratic temptation posed specifically by the math, in comparison/contrast with verbal reasoning about scenarios. Someone employing verbal reasoning is not going to get distracted by the challenge of tweaking the math, is not going to run up against the diminishing returns inherent in the lack of precision with which one can determine the numerical parameters to put in equations or algorithms, and is not going to imagine that math prowess translates into a basis for effective technocratic control. This is not, of course, to say that verbal reasoning guarantees correctness or success, or that verbal reasoners can't be distracted, or find diminishing returns to their efforts, or be tempted by hubris.

My opinion on the math models is exactly the opposite of yours, that they invite us to LOSE track of what is going on with humans in the real world, that people (or at least entrepreneurs) would think outside the box of numerical parameters characterizing pathetic stick-figure "agents."

What AW said. As Daniel reminded us, the more you strip the model down the better, until you realise you didn't need the mathematical complications at all.

How is this going to get past old-fashioned data mining? Is it prompting new questions, or proposing new theories?

Maybe too early to say but it looks like the kind of fad, like POMO in a different field, that enables people to build careers doing something that looks different without actually delivering anything new.

Allan -
I know you talked about math specifically, but I see each of those all the time. The distraction of putting together a zinger polemic instead of getting to the actual economics of the question. Outputs of prose are definitely constrained by the quality of the inputs you are considering (verbal economists make assumptions and provide a structure to their arguments just like mathematical economists do). And the temptation to think you can actually grapple with all of the relevant parameters in prose seems to lead Pete Boettke, for example, into some pretty determinant views on what we ought to do in the realm of public finance.

You can't tell me these faults don't plague all versions of economics, Allan. I see it every single day.

re: "This is not, of course, to say that verbal reasoning guarantees correctness or success, or that verbal reasoners can't be distracted, or find diminishing returns to their efforts, or be tempted by hubris."

Hold on a second - so now you agree with me???

re: "As Daniel reminded us, the more you strip the model down the better, until you realise you didn't need the mathematical complications at all."

Well you added that last clause - I didn't say that.

One more thing -
Alan Isaac (my math econ professor, as it happens) is teaching an online ABM course this summer (website here: https://subversion.american.edu/aisaac/syllabi/496syl.xhtml). Anyone in the Washington Consortium of schools (such as George Mason) should be able to register. He's a very good professors, and ABM is his primary area of research.


Alan Isaac (my math econ professor, as it happens) is teaching an online ABM course this summer (website here: https://subversion.american.edu/aisaac/syllabi/496syl.xhtml). Anyone in the Washington Consortium of schools (such as George Mason) should be able to register. He's a very good professors, and ABM is his primary area of research.

Thank you Daniel. I'm seriously thinking about it.

Register soon - I know we had the opportunity to take it after comps (which is tomorrow) so I don't think it's started yet, but I'm not sure when the cutoff is.

You won't regret it - his class is the toughest I've ever taken, but he's the sort of prof that is up to the task of helping you master what at first glance seems impossible.

Daniel, no I'm not agreeing with you. My point (with which you are free to disagree of course) is that there are aspects of math modeling that are particularly likely to lead one astray. I've tried to explain the what and how.

I hate to disagree with Rafe Champion, but it looks like I'm pretty much on Daniel Kuehn's side on this one. All of our reasoning is subject to error. I don't see why "math" or "verbal reasoning" gets a pass on things like equivocation. (I have to wonder if that dichotomy even makes sense, but let that go . . . ) We say that math is less fuzzy, but I don't think that's remotely true. Diran Bodenhorn had a great article way back when exposing the ambiguities and equivocations in a piece of mathematical economics by none other than Harold Hotelling. Bodenhorn's point was that math does not magically save you from the sort of errors we all recognize as par for the course in verbal reasoning. OTOH, that does not leave verbal reasoning in any pristine state either. We're all pretty stupid and the world is big and complicated. There's no way we're gonna talk about the world without regularly saying stuff that's not even wrong, but just incoherent gobbledygook.

And we all do "mathematical economics" all the time. A lot of our marginalist arguments were first hammered out with the calculus and other bits of classical analysis, including Mises's ERE. Which reminds me of a good example of why Austrians should just totally avoid sweeping anti-math comments. We really like Mises's argument about how the exchange value of gold (or whatever) grows with its liquidity, so that it acquires this specific utility as a medium of exchange which is reflected in its augmented purchasing power. Yeah, Lu! Well, guess who came out with that one first? It was some math-head name Walras, and it's right there in the Elements. That's just one example of the indefinite host of propositions we use all the time that first came out of mathematical economics. We went through this sort of thing when Samuelson died and we considered his legacy.

So if math is not unclean does that mean ABM is sanctified? Well, that way of putting it demands a negative answer. BUT, as we discussed in the comments to an earlier post by Peter L., I certainly think there have been some successes in this area, with Gode and Sunder's famous paper on zero-intelligence traders being the poster child. I might also have mentioned the evolutionary models of Boyd and Richerson and related work of Bowles and Gintis. This stuff has helped to show that it makes sense to say that pro-social instincts could have evolved in the Pleistocene. Fehr and others have experiments with oxytocin tending to support the pro-social view. So here we are on about what Allan has called "thinking directly about purposeful human action," and we are doing through the mathematics of ABMs.

And as I also said in that earlier thread, the boundaries of ABM are really unclear, so that there is no bright line separating, like, Rob Axtell on one side and Carl Menger and Adam Smith on the other. So sweeping statements about ABM have ambiguous referent and should therefore be avoided. Let's talk about specific models such as Gode and Sunder or El Farol or Clower and Howitt.

For me, ABM is interesting and promising, even though, of course, the label ABM is no guard against error. I also stand by the theme of my pro-complexity "Austrian Economics at the Cutting Edge," where I emphasized the importance of what Caldwell calls "basic economic reasoning." Let a hundred flowers bloom, but remember that "basic economic reasoning" is the holy grail.

Forgive for the double post, but I would just like to record my opinion that we should not judge a method by the conclusions it yields unless those conclusions are strict logical contradictions. You have not criticized my argument by pointing out that it seems to have a policy implication that you don't like. If careful study showed that socialism was more productive than capitalism while posing no great risk of tyranny, then we'd better all become socialists. Only this view is consistent with liberalism, btw. As Hayek and others have pointed out, liberalism is *first* a theory of society and only *second* a political program. It's science first, politics second.

I've also been reading books on chaos theory and complexity recently, trying to understand the topics in a context that can apply to economics. I think Roger makes a great argument about the promising aspects of ABM.

The second note "that we should not judge a method by the conclusions" is an interesting contrast to Friedman's methodology that I blogged on this past Friday (http://bubblesandbusts.blogspot.com/2012/06/studying-reality-new-path-for-economics.html).

The post was inspired partially by a recent EconTalk podcast with Ronald Coase, in which he says "I think the time has come when we should study what actually happens." IMO, ABM is a move in that direction and well worth exploring.

Allan, you link to the Alan Isaac website doesn't work.

ABM is an unaesthetic way of understanding human action. I realize this is a subjective-sounding statement. But the goal of understanding is to understand in certain particular ways. That is why we have different disciplines. So, for me, I will let this method pass.

Woj's link also does not work.

Is the difference between DD&T and a certain pesticide just the "&"? :-)

Mostly agree with Roger (as is often the case). Won't bother pointing out minor disagreements.


Whatever ABM is, it is not "data mining."

And Peter's final question about the burden of proof is serious. It is not always easy to prove or disprove a given ABM against data. Nevertheless, I think that they are useful and maybe even the way to go to really get a handle particularly on the microfoundations of macro.

"If careful study showed that socialism was more productive than capitalism while posing no great risk of tyranny, then we'd better all become socialists"

Roger, I think your statement turns out to demonstrate the paradox of "value free" economics that I tried to get at in some earlier comments. A "value free" analysis can never actually be held in a vaccum, but always morphs into an ethical system, unless of course the user of the "tool of economics" is a machine and not a human being. Now public policy designed by a sovereign king who turned out to be a "values free" machine would be a really scary proposition for mankind. (Kind of like an airplane flown only by a machine with no crew in the cockpit-while possible, not many paying passengers want to climb aboard such an airplane.)

At the end of the day, "value free economics" while serving as a tool of analysis in the hands of a values laden human being, can never by itself determine which system is most ethical. That is a deep philosophical position based on our personal views of individual liberty vs the greatest societal good, and which system most enhances the position we deem most moral. Such a values system is one of the things that makes us human.

I strongly suspect Mises already carried a deeply held ethical position on socialism, but to be pursuasive to peers he tried to argue from the perpective of positive economics. Hayek did the same, but with more technical rigour IMHO.

This in no way should be taken as a lessening of the usefullness of the economic way of thinking in understanding how social order forms, or if a spontaneous economic order resulting from prices generated in a relatively unfettered market brings less poverty than an order designed by a board of planning engineers armed with numerical code run by a powerful machine.

Yet Mises and Hayek already knew that the Walrasian model as employed by Lange could indeed show the superiority of the calculus of central planning, yet they were not willing to conceded. IMHO, a utilitarian case for Mises's prior committments (which rubbed off on Hayek in his early career) turned out to be bolstered by the problems in Lange's positive arguments, but
I seriously doubt Mises would have ever conceded the socialist planning argument even if Lange's theory was without holes, since Mises likely strongly believed it meant death for classical liberalism.

This is why I am not willing to arbitrate an ethical position *soley* on value-free grounds, despite the theoretical potential for cold mathematical rigour. To agree to such a postulate would indeed make me a type of radical utilitarian, and I reject this ethical position on the grounds that I do not like it, and I am free to choose.

Here is an updated link: http://bubblesandbusts.blogspot.com/2012/06/studying-reality-new-path-for-economics.html

Thx for the pointer Barkley.

K Sralla:

Well, yeah, you gotta have at least one value judgment in there somewhere for sure. And with some issues, I suppose, you can't avoid diving into "philosophical" issues about, like, what the good life is and stuff. Totally. In the main however:

value-free social science + good will to men = liberal political philosophy

Of course that equation works only if value-free social science comes out more or less the way Adam Smith, thought, which was pretty much my point. You don't care for that line. You have (strongly felt?) "personal views of individual liberty vs the greatest societal good." Those views are moral views you think to be independent of science and they are more specific than just good will to men, right?

Well, please let me just go on record as a "radical utilitarian" in the sense that my ethical baseline doesn't go much deeper or more specific than good will to men.

You say: "I seriously doubt Mises would have ever conceded the socialist planning argument even if Lange's theory was without holes, since Mises likely strongly believed it meant death for classical liberalism." If you are right, then so much the worse for Mises.

Great comments! Allow me to probe a bit further, since it appears to me that
the "liberal equation" might be a bit on the simplistic side.

"Good will to men" really does not have a clear meaning unless we more closely define it with some specific ethical grounding. It might mean anything between live and let live to love your neighbor as yourself. I suppose most common Joes (if they are not Hitleresk) possess "good will toward men" in fair measure.

But stick a more specific meaning (of good will to men) into the equation given a similar value-free economics, and you might come up with quite an interesting variety of "liberals".

For the sake of argument, suppose a machine could genuinely measure everyone's happiness utility function, and though not being able to maximize any particular person's happiness for the sake of maximum aggregate happiness of society as a whole, it indeed had the power to numerically calculate an economic central plan that minimizes poverty and produces the maximum level of aggregate happiness by planning hour by hour the daily tasks of everyone in society, and the means to punish those who willingly get out of line.

According to the loose formula of liberalism above, if I had 1) good will to men (whatever that means) and 2) the correct value-free economic understanding (centralized numerical computer planning minimizes poverty and maximizes aggregate happiness), I might indeed be scientifically compelled to support a centralized authority composed of a computer to rule our lives for the maximum aggregate happiness and equality of society.

So in light of such an example, it seems clear to me that "good will to men" must contain some specific content that at the end of the day can give firm ethical application to value-free economics. These go beyond economics into areas where economics tell us very little. These are big narratives and themes about the very essense of what it means to be human, and may involve the very motivation of humans to perpetuate themselves and survive as a species. These surely transcend formal calculations on a page.

Finally, I just do not believe for a minute that either Mises or Hayek would have ever supported such a ghastly notion of a brave new world run by computers, regardless of any proof (mathematical, apriori, empirical or otherwise) that such a deliberate tyranny of an individual's liberty achieved a kind of aggregate greater societal justice.

Such of model of humanity is dehumanizing, and reduces every human soul to that of a slave.

If it turns out I am wrong about Mises, and he indeed would have supported such centralized economic planning based on value-free economics + good will to men, then shame on him, and I reject his brand of liberalism. So maybe by your definition, it turns out that I am not a classical liberal.

P.S. Hope this does not come across personal or too intense, since as usual your comments force me to think about this more.

All the best.


You did not say so, but in case you are under the impression that ABM implies "a brave new world run by computers," this is not the case.

Is agent-based modelling similar to cellular automata?

I used to work with cellular automata a bit - using that to simulate biological processes. Cellular automata has been around for a long time - von Neumann supposedly did a lot of work on it.

In many ways, biological modelling is similar to economic modelling - it is about understanding how micro-phenomena can generate macro-phenomena when aggregated. And this macro-phenomena can exhibit behaviour which is significantly different to the underlying micro-phenomena - and this is the crux of the difficulty. Such problems are thus hard.

Applying this 'new' simulation technique to economics does not guarantee improved modelling outcomes.

From my experience with using cellular automata to model biological processes, all such models would suffer from the usual difficulties - including simplifying assumptions, introduction of various 'non-existing' parameters, calibration, etc.

My point is that modelling outcomes would still be dependent on the mathematical construction of the model, regardless of the mathematical techniques used - be it differential equations or agent-based modelling.

There is nothing which guarantees agent-based models would lead to superior, more realistic economic models than the conventional general equilibrium models in use today. In fact, it would present other difficulties - and personally I am dubious of agent-based modelling being a promising pathway for either economic or biological modelling.

Much of the enthusiasm over agent-based modelling is that conventional macroeconomic models - be it econometrics based models, or the dynamic stochastic general equilibrium variety - have failed. So someone suggested employing agent-based modelling as a new 'tool' - a new 'hammer', so to speak. The problem is that the real economy is not a 'nail', which calls into question the suitability of using a 'hammer' at all.

Having said that - unlike many Austrians, I don't have anything against mathematical modelling per se. I believe there is merit to such modelling activity. The problem is when viewing mathematical modelling as the primary way, or the exclusive means, to understand economics - and this is where I believe the mainstream economics profession has gone wrong.

I have posted (under various pseudonyms) on this blog that proper pedagogy is to teach economics both from a (i) philosophical perspective; and (ii) mathematical perspective. To link the two would require a course in the philosophy being mathematical modelling/philosophy of statistics (statistics is hardcore epistemology after all - much more interesting and profound than what Kant or Popper has written).

I think this idea is worth exploring, especially as the economics profession is wrestling with these issues from a pedogogical perspective. MIT's Callabero got close in a rant against mainstream economic modelling circa 2010.. but i.m.o. he has yet to fully grasp the philosophy and proper usage behind mathematical models in economics.

Also, I want to say that I agree with Mario Rizzo on agent-based modelling being 'unaesthetic' in understanding human action. As I had alluded to earlier, due to its lack of mathematical tractability I am personally dubious that it can lead to improved understanding of economics.

However, they can generate very entertaining computer simulations - c.f. http://en.wikipedia.org/wiki/File:Gospers_glider_gun.gif

@Frederik Marain: It was Daniel Kuehn who gave a link to Alan Isaac.

Somebody in an earlier post asked for a reading list on ABM. I could oblige by providing a rough head start.

Don Lavoie's chapter on "Austrian models" in The Elgar Companion to Austrian Economics (edited by Peter Boettke) is a good piece to commence with.

I'm also aware of a couple of papers using ABM methodology in some recent edition of The Review of Austrian Economics.

@Roger Koppl: "OTOH, that does not leave verbal reasoning in any pristine state either."
>>Pure straw man, Roger. Nobody said that. Who are you arguing with?

"A lot of our marginalist arguments were first hammered out with the calculus and other bits of classical analysis, including Mises's ERE."
>>I don't recall Menger developing marginalism via calculus or math at all. Without going back through Human Action, I don't recall Mises laying out ERE with math. Math in Walras? Yes, and the differences are striking. With Menger we get a process stemming from human choices. With Walras we get a dreary fixation on simultaneous equations and whether they have solutions.

"We really like Mises's argument about how the exchange value of gold (or whatever) grows with its liquidity, so that it acquires this specific utility as a medium of exchange which is reflected in its augmented purchasing power."
>> Yeah I found this in Human Action. But Mises is referring back to Menger. The argument you are talking about is found in Menger, and I doubt Menger got it from Walras. Did Walras discover monetary value in marketability via math? Or was it an ordinary reasoning argument, as with Menger?

"I would just like to record my opinion that we should not judge a method by the conclusions it yields unless those conclusions are strict logical contradictions."
>> If I have what I take to be good reasons for doubting the efficacy of the technocratic state, and if a certain economic approach seems to lend credence to the technocratic state, and if I am commenting in a blog where I take it most others share my lack of confidence in the technocratic state, then it is a perfectly valid warning on my part to point out that the approach in question tends to lend false support to the technocratic state. By rough analogy, if I have good reason to trust the law of conservation of energy, and if a particular approach to analyzing natural phenomena seems to lend credence to non-conservation of energy, then it is a perfectly valid warning on my part to point that out. (And please, no shallow barbs about how I'm equating free markets with the laws of physics.)

We observe a particular reality. So we wonder what kind of world might be producing this outcome and we try and put the essentials into a computer model. How do we know we are right? We see what comes out. If it is close, we are happy. If it is not, we tweak it until it is close, then we are happy. Where does this really get us? There are probably a very large number of agent-institution-profiles that will produce outcomes close enough to make us happy.

Yes, but the ABM method is not about finding the right causes ("right" as in real-world existing), but about analyzing the sufficiency of already known or hypothesized micro foundations or causes of macro phenomena. Epstein & Axtell discuss this in Growing Artificial Societies (1996):

"The surprise consists precisely in the emergence of familiar macrostructures from the bottom up--from simple local rules that outwardly appear quite remote from the social or collective phenomena they generate. In short, it is not the emergent macroscopic object per se that is surprising, but the generative sufficiency of the simple local rules." (pp. 51-52; emphasis in original)

Per Bylund: “…about analyzing the sufficiency of already known or hypothesized micro foundations or causes of macro phenomena.”

I’m still trying to figure out what the point of ABM is and I’m not sure what you mean by “sufficiency.” Do you mean that ABM will be able to separate good from bad macro models?

The main problem I see with ABM is the same problem with all models – getting the right data. As Hayek wrote, sometimes the needed data just doesn’t exist and may never exist.

Econometrics was supposed to end arguments about models, but I have never seen an econometric model resolve any arguments. The competing modelers just gripe about the flaws in the other guy’s model or selection of data, etc.

As Mises wrote, models are just prose translated into math symbols; they can’t tell us anything more than the prose did. And it may tell us less if we don’t have the data we need.

I can see the value of ABM in forecasting if modelers can get the data they need.

I guess I need help understanding what the purpose of ABM is other than translating prose into math.

K Sralla:
If a computer could make the calculations you describe, then people and society would both be radically different than they in fact are. So, yes, why mightn’t it be best for a radically different sort of creature to have its every move controled? But such creatures would not be human beings.

Given your forceful rejection of my consequentialist perspective, I think it’s fair to ask: Is there is any scientific evidence about anything that would bump you away from liberalism?

Can you please provide a link to the Callabero “rant”?

Nice point, nice quote.

Questions for those familiar with ABM:

1. Is it true that these models presuppose agents whose knowledge is known to the modeler?

2. Given that these models produce unpredictable outcomes, does anyone out there have opinions about whether, a fortiori, models in which agents' knowledge is *not* known to the modeler would also produce unpredictable outcomes?

3. Is there any application of complex-systems theory to social interaction apart from ABM?

I see a lot of promise in ABM. For example, these two pieces by Doug Kenrick, et al. show how one can get a variety of cultural expressions from human psychological universals:



It is a very powerful form of modeling. Part of its power is that, like in the real world, complex outcomes result from simple rules. Also, one can tinker with the inputs until one finds a model that most accurately matches what we see in the real world, meaning it can help us discover the rules, which we can then add in to other models to further test.

Eric Beinhocker also sees connections between ABM and Austrian economics:


@McKinney: ABM is not formal modeling in the same sense as econometrics--it is a means to dynamically "grow" societies and thereby investigate the effect by single or combinations of characteristics and/or behavioral rules. For instance, I have modeled a simple market in order to test the transaction cost explanation for firms. I do not program firms; I program agents and give them "personalities" through assigning random values to a set of characteristics and specify their behavioral rules (what they do when they meet somebody else, how they trade, etc.). Then I let these actors interact to see the effect of certain sets of "personalities" and I can identify if e.g. a certain interpretation of transaction costs are sufficient to create firms (properly defined). I can tinker with the values of agents' characteristics as well as their behavioral rules to figure out if the TC need to be high or low, what other conditions are necessary, etc.

@Jeffrey Friedman: I don't think one needs to know the knowledge of modeled agents. In fact, I think it is common practice is to randomly assign values to agents (their characteristics, such as risk aversion or "vision" etc.) as well as their behavior. Also, one can for instance program agents who "learn" over time through saving and interpreting (gaining) their experience of interactions and choices made. This experiential knowledge can also be programmed to have an effect on an actor's behavior, which makes the actors' behavior and "knowledge" unknown (in the sense, not predictable) to the modeler. Of course, one can always observe and keep statistics of changes to agents, so their "knowledge" is not unknown (unknowable) to the modeler--but it may be neither predictable nor predetermined; actors' knowledge may be a result of a very complex causal chain of interactions and reactions.

Jeffrey Friedman,

I do not think your distinction between whether the "knowledge" of the agents is "known" to the modeler or not matters. These are agents in a model, and it is meaningless to speak of them having "knowledge." Actual human beings have knowledge.

What the agents have is certain rules of behavior and one can endow them with the ability to remember past events that occur during a simulation and "learn," which is indeed one of the more interesting things in many ABMs. But, this is not a matter of agents having knowledge that the modeler does or does not "know" of.

Regarding other complex systems approaches, there are others, although this is a matter of how one defines complexity, an ongoing conundrum and controversy. I would egotistically suggest looking at my 2009 edited volume (available in paperback) from Edward Elgar, _Handbook on Complexity Research_ to see discussions of both what complexity is and is not, as well as examples of a variety of applicatioins and approaches to doing it. I note that one Roger Koppl has the next to last chapter in the book, titled, "Complexity and Austrian Economics."

The other authors (in some cases I shall list just the prominent lead) are me (three chapters), Brian Arthur, K. Vela Velupillai, Cars Hommes, Michael Kopel, Alan Kirman, Richard H. Day, Thomas Lux, Joseph McCauley, Frank Westerhoff, Peter Allen, Herbert Gintis, and David Colander.


Here's the Callabero 'rant':



Yes, cellular automata are a part of ABM and thus also a part of the broader complexity scene, but not the whole story by a long shot on either, although some of the questions you raise are serious for any kind of simulation modeling, quite beyond CA.

Ah! Caballero's 2010 JEP paper. Thanks, John. I'm not sure he goes wrong on the use of math models. What do you have in mind?


Caballero has very interesting twist on Hayek's "Pretense of Knowledge syndrome" (certainly different from what Hayek meant).

It's been a while since I last read his paper, but he seems to suggest that the way to go about things is to use new mathematical modelling techniques (that would include agent-based modelling), and use more realistic modelling assumptions.

The problem is that incorporating more realistic assumptions can lead to worse outcomes (like my agent-based modelling project). The reason is that each model has many simplified ('false') assumptions, and these assumptions interact with one another in generating the model's outcomes. If you improve some assumptions while retaining the rest, the resulting model could lead to a worse outcome.

This is one of the reasons why I am dubious about the prospects of agent-based modelling - researchers would still be tweaking their agent-based models to generate the results they want.

The other problem is that agent-based modelling have simulation outcomes which are hard to grasp - you only get simulation results, not nice snappy equations (as per generic general equilibrium models). It would thus be difficult to interpret and predict the model's outcomes - i.e. these models might not be very illuminating.

And even for successful agent-based models, perhaps the same outcomes can be generated using conventional techniques such as dynamical systems, etc.

Back to Caballero... I think at the core he still agrees with the mainstream approach to economics, i.e. primary by way of mathematical models. I disagree with him on this count, since models can only capture so much of reality... and most people seem to construct their models so as to generate the results they want.

My view... (parts of which are cut-and-paste from an old post):

I think there is scope to consider teaching economics in two components - both which are integral, and connected:

1. The first would be having economics taught as philosophy - Smith, Hayek, Ropke ... even Keynes. Economics is inherently intertwined with politics, history, philosophy... which is probably why economics used to be known as 'political economy'.

2. The second would be on mathematical modelling - the building of models of the economy. This would be contingent on your economic beliefs, so the first component would guide you in your modelling endeavours.

A course on the philosophy of statistics would be relevant too - after all, statistics is but epistemology couched in mathematical language. And to interpret any statistical data, you need to come up with a mathematical model of some sort. You can reduce the number of assumptions grafted into your statistical model if you have tremendous amount of data - but that's often a luxury in the social sciences.

Models should be recognized for what they are - because they can only be approximations of the real thing, their applicability is limited. To what extent are the models useful - over what time frames, and how precise are the forecasts - these are usually very difficult questions.

In much of statistics, the honest answer in many situations is that we just cannot make any definite conclusions from the data. However, such conclusions do not allow you to publish papers...

Modelling can be a useful exercise - and to interpret empirical data, you cannot refrain from modelling activity. But it is an 'art' - which is not well communicated in economic pedagogy. The situation is better in Maths & Stats departments - but these usually come from hands-on experience... at my old university, there was only one course dedicated to such aspects of applied statistics.

What I think is dangerous with modern economists is that they have been taught to do #2 (mathematical modelling), without #1 (economics as philosophy). There is a need to recognize that models are not truth - as a famous statistician once said, "All models are false, but some are useful". And in many ways the modelling outcomes have been 'assumed' from the outset when the model was constructed.

Try asking a economics student, especially from reputable institutions - "Should you cut taxes during a recession?", "Why are countries poor?" And he could respond, "Well, according to Krugman's JPE paper... Now, the Fafchamps model assumes you have n regions, so you have... But the Matsuyama model which assumeasserts that..." Might be difficult to elicit a 'yes' or 'no' answer.

ABM is used for sports simulation and prediction on such sites as Fox Sports' Whatifsport.com.

It seems that the technique is good for simulation, which might make ABM a good teaching tool, and possibly for forecasting.

So, John, do you have the answer? Should one cut taxes or not during a recession?

Guess I am getting a bit frustrated here. You have done some work you claim with CA in biology unspecified, which led you to identify some well known problems with simulation modeling in general. You think that GE might be better in econ, even though you admit that its main app in macro in the form of DSGE models is a "failure." You sort of think math modeling might be OK, maybe, if people learn their philosophy. Wow. Is that it?

So, if you are able to give a definite "yes" or "no" answer to the question you sneer that "a economics student" would refer to various models in attempting to answer, what is the basis of your answer, given that you sneer at ABM and DSGE (and did not obviously support either some form of ABCT or some obvious Keynesian alternative or any other approach for that matter)? Do you just "know" the answer after perusing the works of dead economists of various schools, or was it reading Hume and Wittgenstein (or maybe Plato and Sartre) that led you to it? Or do you think there is no general answer, but find it amusing to sneer at students who cite various models when considering the question?

You totally missed my point, Barkley.

The starting point is to understand what your goals are in constructing mathematical models, and be transparent and honest about the assumptions that are grafted into it. And also be humble about the limits and applicability of it.

It is not that models by itself are superfluous, as I said earlier they do have their place. The problem is when people are not honest about the assumptions which they make, they don't fully grasp the simplifications embedded in the model, and they overestimate the applicability scope of the model.

This applies to analysis of empirical data too, there is thus no such thing as fully 'objective' interpretation of data. Assumptions have to be made when interpreting data... usually the disputes over data interpretation can fundamentally be traced to disputes over the assumptions adopted.

But the truth - economic truth, in this case - is out there, we're all grasping for it. It may or may not be beyond our reach, but let's be honest in our pursuit of it.

Per and Barkley:

What I mean by "knowledge" is the means by which a human agent would interpret new information, including experiential information.

Among different agents, these means are not random. They are determined by past information, experiences, and interpretations. We cannot know what the past information, experiences, and interpretations of a given real agent are. But there is no reason to think that they are randomly distributed. They are unpredictably but not randomly distributed.

Therefore, my question is: Given that ABM confers on agents either random behavioral rules for reacting to experience, or rules known to the modeler (right?), yet produces unpredictable results, then if we try to think about the implications for real-world agents, a fortiori, do we get equally or even more unpredictable results?

Jeff: I don't think I'm getting the picture. Maybe it would help if we moved to a specific example. In your way of thinking, are the agents in Arthur's El Farol model governed by "random behavioral rules," or by "rules known to the modeler"? And in what sense are the results "unpredictable"?

"Is there is any scientific evidence about anything that would bump you away from liberalism?"

As far as I am able to tell, the science of economics indeed shows that relatively unfettered markets guided by a good set of rules and institutions operate more effectively than centrally planned economies.

But.... even if a system of computer model-based central planning were shown to give more economic efficiency, I would still be in favor of a market economy, since it maximizes individual freedom and allows life to be worth living in my view.

One of the reasons for my viewpoint is that I actually think that numerical models, given the correct set of rules, can in fact appear to replicate many of the patterns we see in complex social orders.

We can produce numerical models of schools of fish and flocks of birds that behave on a macro-scale in very life-like manners, despite very little understanding of the micro-scale. Prediction of future behavior of the macro-economy will always be a problem, but patterns in a numerical simulation can appear very life-like, and as a process study, indeed give us excellent insight into the nature of complex order. Mises and Hayek in the 1930's had no conception of the type of computational power that would now be at our disposal.

However, I worry that these very scientifically valuable models may one day be used by despots who may impress an ignorant public to cede their liberty, thereby consodidating power and restricting individual liberty based on "so-called" science. The public will not understand that to use these models to predict the required imputs to achieve a desired future economic state will always be beyond what science can deliver. Yet, nevertheless, the models will appear very compelling.

For this reason, I am not sure that a liberalism based mainly on which economic model looks best in the peer-reviewed literature is a vibrant libralism. At the end of the day, we must believe in the values of individual liberty, and support the economic systems that bolsters this the most. This goes beyond the economics journals.


Actually, I would argue that because one can manipulate ABM agents to get the answers that you want is one of its main benefits. After all, if you create agents that create/work well in a socialist economy, one can then look at the rules/traits involved and ask oneself whether or not real human beings are that way. If not, we now have some good, explicit reasons for why socialism doesn't and cannot work.

On the other end, one can also manipulate the agents' rules until one gets an outcome that resembles what we see in a real economy. And isn't that one of the goals of scientific models, to most accurately model reality?

And then we should expect to do a bit of both. That is the very strength and power of ABM.


Let us be careful to distinguish "real human agents" from those in models. The latter will never fully mimic the former, who really are too complicated. However, I repeat, a major thrust of many ABMs is to incorporate learning by agents. This is at the core of the much-studied and imitated Santa Fe stock market model. It goes on in a model I published on with Gallegati and Palestrini in Macroeconomic Dyhamics last year on the period of financial distress, something never previously shown to be possible in any model, although it is a widespread empirical phenomenon.


Sorry, but you are verging on being insulting without perhaps meaning to be. Obviously I have no idea what you are talking about, given that I apparently "completely missed the point." Right, all of us who do ABMs are foolishly ignorant of the goals we are pursuing; I am personally opaque as all get out and lie through my teeth about the nature of assumptions being made in models I study, and of course I am just as arrogant and pompous as all get out about everything I do and its applicability to whatever (which actually is pretty true, as many here can attest, :-)).

Whether or not an "objective" analysis of data is possible or not is indeed a philosophical question, but I do not think you have the answer to it, even if you think you do. I shall not point out the applicability of your own arguments to your own arguments any further.

As I've pointed out on another thread, the El Farol Bar model and its variants serve as good examples of the problems with ABMs. The "agents" are constrained to think within a very narrow box, namely, trying to use the previous pattern of attendances to outguess the others as to whether the bar will be crowded tonight. And the purveyors of these models generate all sorts of clever variants on thinking within that box. But humans think outside the box, and economists who employ ordinary verbal reasoning are in my opinion less likely to go on and on with this silly zero-sum game without stopping to think about how people might start calling ahead, or how the owner might set up a reservation system or expand the bar or how other entrepreneurs might establish similar bars nearby. And of course, doesn't it seem so obvious that a bnevolent dictator could smooth things out by telling which people where to go when?

Thanks Barkley and Allan, I think you've answered my question.

Well, Jeff and Allan, a bit of a note on the El Farol bar model. It attempted to model a real phenomenon that Brian Arthur and some of his friends had observed, that the El Farol bar tended not to get too crowded on any particular nights, with an occasional exception. They did not report that this was being achieved by lots of people clling or all these other "out of the box" methods that Allan suggested, although he is correct that people will resort to all kinds of things that do not show up in any models, no matter how open to learning or other variations on their narrowish rules of behavior.

Barkley is in a snarky mood today.

I used to be a mathematics and statistics major, and I have written my thoughts down on the philosophy of statistics. So I'm not saying things off the cuff.


Point taken. Though on the flipside, it also means that ABM has the potential to validate crackpot ideas... but I guess that would be true of whatever variety of mathematical modelling. Here's looking forward to what ABM churns out.

I was going to ignore this thread, but then I had a couple of glasses of wine.

Many of the claims about models, math and economics in this thread are simply absurd. Math does not have a monopoly on the ability to obfuscate or provide rationales for harmful policies. Neither Taylor nor Lange had any trouble making their case for socialism without the aid of mathematics. On the other hand, Gordon Tullock was quite capapble of analyzing repression with the aid of equations and graphs. If Tullock believed that it enabled him to more clearly think about or explain political behavior, I am not going to argue.

One thing math does not enable you to do is hide your assumptions. The assumptions have be to be stated, and a mathematical approach enables you to do determine the degree to which the conclusions are sensitive to changes in the assumptions.

Models are tools. A good model is one that does what you ask it to do. Does Ricardo's model help you understand the benefits of trade? If so its a good model. If not, it is not.

Models are not about math; they are simplified versions of reality. Coases description of what would happen in a zero transaction cast world is a model. It is a simnplified version of reality that enables us to understand what happens in reality.

If something helps you to think aobut an interesting problem I don't care whether its words, graphs, or equations.

As for the particular issue of ABM, I believe that many people regard Schelling's model of segregation as one of the early examples of the appproach. If so, it seems to me that it fits the criteria of helping us to think about an interesting problem.

@Per: (further to McKinney).
Yes, but the ABM method is not about finding the right causes ("right" as in real-world existing), but about analyzing the sufficiency of already known or hypothesized micro foundations or causes of macro phenomena. Epstein & Axtell discuss this in Growing Artificial Societies (1996)

"The surprise consists precisely in the emergence of familiar macrostructuresfrom the bottom up--from simple local rules that outwardly appear quite remote from the social or collective phenomena they generate. In short, it is not the emergent macroscopic object per se that is surprising, but the generative sufficiency of the simple local rules." (pp. 51-52; emphasis in original)

Interesting. I will look at this article – I saw it referenced in the RAE articles on ABM. My comments were not necessarily to criticize ABM, as to raise a concern. Your comments don’t really assuage that concern. Why the criterion of sufficiency? Why not plausibility? Is this some sort of Occam’s Razor idea. If we can “explain” (generate) the outcome we seek with minimal agent-behavior assumptions, is the is a good thing? Why? Naively, don’t we want to know what is really going on?

I can tinker with the values of agents' characteristics as well as their behavioral rules to figure out if the TC need to be high or low, what other conditions are necessary, etc.

And then? How do you judge if the model is plausible – does it “explain” or simply show one kind of (many?) processes that could yield similar outcomes? Just asking, genuinely curious.
On the question of facts and values. I intend to address this when I discuss the burden of proof question. In the meantime see:


Actually, just looking at what I wrote, in response to what you wrote, I think I may have missed something. You are suggesting that the behavioral assumptions are already plausible or have some prior presumptive value - as, for example, being used in standard neoclassical models. And the ABM then checks the "sufficiency" of these assumptions. This makes me feel a little better, but ... .


Plausibility of assumptions is important. Some above have argued that mere plausibility is insufficient, that modelers can play tricks with them to get results they want that the unwashed masses trying to figure out what is going on will be unable to figure out (me being snarky again, :-)).

So, it is more than just that. For ABMs to be taken more seriously, one really has to have the whole banana: plausible assumptions with a plausible model where one can have at least some idea of what is going on, with some degree of robustness of results that also help to explain some real world empirical phenomenon, preferably one that other approaches have not done well at explaining or modeling. There have been some that have done that, with the famous example of the Schelling segregation model among the most famous and justly so.


Glad to see I am not the only one awake at this "strange" hour.

I think what you say makes perfect sense and will keep it in mind in my approach to ABMs. I have a paper that I am commenting on at a conference next week that I will subject to those criteria.

Coming back to what I said at the top, this is really no different from the way we evaluate traditional statistical models in practice - and ABM seems to have unexplored promising potential.

IMHO brad nailed it. I guess I would pick one nit: You can have hidden assumptions in a mathematical model too. But I'm sure brad's basic point even there was just that the math can sometimes help you to clarify your thinking. So right on, brad.

@Peter: Yes, I think that's right. To me, ABM is used primarily for testing the micro causes of macro phenomena in line with the Coleman bathtub view of inter-relatedness. It tests neither the micro level nor the macro phenomenon, but only the (possible) relationship(s) (/causes) between them.

The behavioral assumptions for the micro level are not tested but selected because they are plausible (or prescribed by theory). It is the interaction of agents according to simple behavioral rules that cause or do not cause the macro phenomena. In this sense, the sufficiency of certain rules or contextual effects (through variables) can be tested in terms of whether they cause the expected phenomena.

In the example I mentioned above, where I test the sufficiency of transaction costs on creating firm-like structures, I let TC vary in both magnitude and form (as a "tax" on search, on contracting, etc.--and other ways) and observe what effect this has on market structure. The question implied is "what kinds and magnitudes of TC cause firms?" Granted, the answer depends in part on the behavioral rules--which is why they need to be grounded in theory or generally accepted as plausible. As the macro phenomenon is complex and spontaneous, one cannot necessarily foresee the effect of varying inputs/parameters. In fact, I have been thoroughly surprised of the outcome several times.

Others may use ABM in different ways, but I am not sure to what extent I agree with broader applications than testing the micro->macro link.

I think the micro-macro link is precisely what ABM is all about. I would go so far as to say that ABM is the main pathway to a realistic macroeconomics and that it will overturn almost everything currently believed in macroeconomics. That will be its main benefit. Macropatterns after all are created by microinteractions, and it's about time economics reflected that fact.

@Barkley Rosser re El Farol: So what we have at first is a harmless little Tinkertoy, a diversion from which we can turn, after sufficient amusement, back to significant problems of econ theory in the real world. But no, El Farol keeps getting promoted as an example of "success" that should lead us to take ABMs seriously, and along with its variants keeps getting pursued in the "econophysics" literature with all sorts of choice algorithms being pusued by "agents" that are assuredly not being followed by many bar-hoppers in Santa Fe or anywhere else. And isn't it so obvious that a benevolent dictator could direct people in such a way as to avoid overcrowding. So here we have an example of what I summarized under "DDT:" Diversion from thinking about the choices of real people in the real world into tinkering with the parameters of stick-figure "agents" in a zero-sum game; Diminishing returns as one strays farther from the concerns and likely actions of human beings; and Temptation to draw collectivist lessons.

The thing with assumptions in models is that they are intrinsically 'false'. This is a necessary part of the simplification procedure - a simplified form of reality if you will, which is abstracted in such a way to tease out the elements you want to focus on.

As George Box (famous statistician) once said, "All models are false, but some are useful".

These assumptions, which are either "half-true" or "false", could interact in such a way as to give outcomes which are not sensible. As a friend (which had Machiavellian tendencies) once said to me - if you can get people to accept certain half-truths, they would be force to accept the logical consequences of these half-truths no matter how absurd the consequences.

Having said that, plausibility of assumptions are not always crucial. For instance, in some contexts the real probability distribution might not be a normal distribution - but assuming that it is will not affect the outcomes significantly. This concept is called "robustness" in statistics.

On falsifiability...

Most mathematical models are only approximate descriptions of reality - they are thus false, and would definitely be rejected given sufficient data. This raises the issue of how to judge the quality and applicability of each model... this debate goes back to Kydland Prescott.

This is unlike, say, for the laws of Physics, which are perfectly described by mathematical equations. The "unreasonable effectiveness of mathematics" in that discipline, as a Nobel laureate in Physics puts it. There is rarely such equivalents in most social sciences.

Also, the conclusions of an economic model is highly dependent on its assumptions, which in turn are usually 'false' (perhaps by 'necessity'). Further compounding the 'intrinsically false' nature of the economic model is that it is not clear how to statistically measure the parameters and variables in the model (e.g. inflation, natural rate of interest, etc.). The error bars on such things tend to be fairly large.

That does not mean that models cannot be illuminating. Many aspects of e.g. the Solow growth model doesn't fit with the empirical data, but it's still a decent construct and sheds light on a few key themes. My view is that we can aspire to a qualitatively meaningful model, but for a quantitatively accurate one would be difficult.

@Troy. I envy your optimism. But I don't see it at all. The thing is, it is possible to produce an ABM for whatever macro-result you want, and whatever policy you support. What independent method or criterion do we have for deciding between the Keynesian and Austrian ABMs and the policies they imply?


Ah ha! So, you are using this blog to help you make comments on a paper for a conference, eh? Fair enough, I guess, :-)

Regarding your pessimism, I may not be as optimistic as Troy, and I agree broadly with Per, but I stick with what I said before. One tests these models from all ends as it were, plausibility of assumptions, model, data, output, robustness, and so on, hopefully as honestly and objectively and fairly as possible, keeping in mind the genuinely profound difficulties involved in doing the latter.


No substantial disagreement. The Nobelist was Wigner. So, I agree with the remark by my old friend, George E.P. Box, but at the same time say he is oversimplifying a bit. Unlike many here I am not at all averse to pure theoretical modeling in which the assumptions may be drastic simplifications that are seriously false. But I also think that when one is really getting down to trying to explain what is going on, ceteris paribus, truer assumptions are to be preferred to falser ones in a model. As I said to Peter, models should be tested and judged and analyzed from all ends from the inputting assumptions to the outputting results and in-between.


Sorry, but I think you are really off the boat on the El Farol bar model. How do you see "collectivist conclusions" coming from it or any of its numerous descendants or relatives? I would contend that it is nearly an archetypal example of a model exhibiting Hayekian self-organization without any central planner or overseer or organizer. The people manage themselves in an orderly manner over time without anybody doing anything special, including all of that "out of the box" thinking that was being dragged in, such as people calling for reservations and so on.

You sneer at the econophysics descendants, although I see no particularly good reason to do so. What is your problem with these models? I am not aware of any collectivist conclusions obviously arising from those that descend from the El Farol bar model any more than I do from it itself. This class of models are now known at "minority games" models, and there is a large lit of them, all drawing on Arthur's original model. The basic intuition is that the winner of a minority game is the person who is not where the mob is, but in effect in front of it. That is like the person who succeeds in getting to the El Farol bar at a time when there are as few other people as possible. But, as in the El Farol bar model, people trying to win the minority game end up self-organizing the market to operate in an orderly manner. What is your grand beef about this, and what important alternative approach are these models "distracting" us from?

"All theory depends on assumptions which are not quite true. That is what makes it theory. The art of successful theorizing is to make the inevitable simplifying assumptions in such a way that the final results are not very sensitive. A "crucial" assumption is one on which the conclusions do depend senitively, and it is important that crucial assumptions be reasonably realistic. When the results of a theory seem to flow specifically from a special crucial assumption, then if the assumption is dubious, the results are suspect."

Robert Solow (1956)

Barkley, I stated explcitly where the collectivist lure comes from. Do you actually read things before commenting? Here you have all these purported agents employing various strategies to outguess the rest and show up on an uncrowded night, when obviously if they all submitted to direction by some planner the resulting fluctuations would be smoothed out and nobody would have to suffer arriving when the bar is crowded. What's my "grand beef?". I've laid out my concerns and reasons for skepticism at least twice on this thread alone. Furthermore, in my reading, I find hostility to markets endemic in the econophys movement -- in fact, the econophysics books sneer at the econ maistream's purported blind faith in markets, oblivious to the fact that the mainstream is always claiming fault wih markets and supporting a whole menu of gov't interventions.

Sorry, Allan, you are just misreading this literature. The bottom line on the El Farol bar is that it is amazing who relatively minor the fluctuations are. Do you not read papers carefully before you leap to unwarranted conclusions about them all the time?

Much of the econphys lit is critical of many of the assumptions made in much of the standard econ lit, particularly about ratex and EMH and so on. But except for some of it, most of it does not then pose that "Oh, big government can fix this," and this is certainly not coming out of the minority games lit particularly. You are seeing things that are not there. Sorry.

That quote by Robert Solow highlights the difference in modelling and data interpretation expectations in the physical sciences v.s. the social sciences. It also highlights that modelling is really an 'art', rather than science.

What is particularly lacking in economics is a coherent philosophical framework to modelling. Popperian falsifiability obviously cannot be applied in full here. One can agree with Solow as quoted above, but it barely deals with the difficulties encountered in modelling exercises.

"Except for some of it....". It's possible that what I've read is unrepresentative, in terms of the attitudes or interpretations of practitioners. When home and using more than an ipad I'll re-check my impressions on that. Nevertheless, I stand by what I take to be clear tendencies of math modeling to mislead, and this includes the econ mainstream's long fixation on general equilibrium and the recent ABM fashion that I associate with econophysics. Ironically, what the mainstream and the econophysicists have in common is in this sense more significant than their differences.

Ah ha! So, you are using this blog to help you make comments on a paper for a conference, eh? Fair enough, I guess, :-)

Not really. The paper is much more narrow than this discussion, and I had already written my comments. But what you said will help me add a few things. :-).

The El Farol Bar model actually demonstrates how equilibrium at the ideal number of patrons per night is never and could never be achieved, even as the number fluctuates in waves around the ideal number. Of course it is a simple model with a single bar and, thus, atypical (though I would bet it would be a good model of those few places that do in fact have a single bar). Add in another bar, and variations in the agents' tolerance of crowds, and you would end up with more complex patterns of behavior -- but even farther from equilibrium.

I am optimistic about ABM precisely because one can create Austrian or Keynesian outcomes. What one should do, however, is then compare the kinds of agents one would have to have to get those outcomes to how real human beings act. I would call ABM a first step in one's investigations. You can give your agents a variety of characteristics and see what happens rather than use, say, perfect rationality and perfect knowledge because that's the only thing that lets you do the calculations.

Not to mention that ABM allows you to do far more realistic economics of time and space.

Certainly my ABM optimism isn't without qualifications. Of course you can still come up with ridiculous conclusions and set about proving your pet theory because you're an ideologue rather than a scientist, but that's true of many other models and theories. One doesn't reject Darwinism because he was misused by the misnamed Social Darwinists.

I am optimistic about ABM precisely because one can create Austrian or Keynesian outcomes. What one should do, however, is then compare the kinds of agents one would have to have to get those outcomes to how real human beings act. I would call ABM a first step in one's investigations. You can give your agents a variety of characteristics and see what happens rather than use, say, perfect rationality and perfect knowledge because that's the only thing that lets you do the calculations..

Certainly reasonable, and I absolutely am not advocating a rejection of ABM on grounds of a multiplicity of possible outcomes. The proof of the pudding will be in the eating. But, here is my expectation. I expect that with not-so-different, plausible assumptions about agent-behavior and institutional-framework one will be able to get very different results - notably, Keynesian v. Austrian. I suspect it will depend much on the magnitude of the behavioral parameters and much less on the type. Even this would be quite informative, suggesting that there is something else going on, and would beg more, varied investigation. How far one can go with this remains to be seen. I counsel guarded optimism.

All scientific methods should be used with guarded optimism. Each aspect reduces bias.

In my last post I raise the question: Can we be certain, as a logical matter, that if certain conditions obtain, certain definite types of outcomes will result from the free-market process? As I understand it, much of effort in ABM-type investigations is into this type of issue. Not so much into the “logic” as we usually understand it, as into the necessary outcomes of certain kinds of processes, such processes being modeled by the behavior of the virtual agents of the model. By investigating these process-outcome variations one hopes to get traction onto the kind of economic-policy one feels appropriate – for example, supporting less intervention, support for property-rights, etc. – or the opposite.

The comments to this entry are closed.

Our Books