Contradictions within economic theory. All well known but still important and, I think, not taken as seriously as they should be.

Posted on by

Asher Meir writes:

Economists (like me) love models with “rational expectations”. In these models, all agents have mutually consistent expectations of some future economic equilibrium. If everyone expects that the outcome (which can be stochastic) is X, then it will be optimal for everyone to engage in choices that will result in X.

There can be no doubt that there is a certain logic in an equilibrium like this. So we spend a lot of effort and computer time to solve them. Because we don’t know what these equilibria look like.

Uh, wait.

If we don’t know what the equilibrium looks like, how is it possible that every single agent in the economy does know what it looks like?

I was the programmer for a model like this in the 1980s. This paradox bothered me a lot. Perhaps there is some satisfactory resolution to the paradox, but what drove me crazy even more was that no one was bothered by this obvious paradox. Alan Auerbach and Larry Kotlikoff were very happy to have me as a programmer because I was good at making it converge. Often you can’t prove that these models converge, but usually they are pretty well behaved and a decent programmer can nudge them towards an equilibrium. Then lo and behold, Larry and Alan and I would discover the equilibrium growth path is that had already previously been known to every single agent in the US economy and which underlied the US economy. . .

In the 1980s the models were pretty simple but today they are amazingly complicated. The stochastic ones are gridded and get you into the curse of dimensionality. Researchers expend a lot of effort trying to show that the model has a stable equilibrium, hopefully only one. But nobody seems to notice that while the researcher can impose a transversality condition on the model, s/he can’t impose one on the world.

As far as I can recall the only economist who was interested in this gaping black hole in our profession was Herb Simon. And nobody talked to Herb Simon.

My reply: Yes, I’ve heard this general argument made before. On one hand, individuals are modeled as Bayesian optimizers. On the other hand, actual Bayesian optimization is really difficult and requires Ph.D. statisticians etc. I had the impression that the usual resolution of this problem is to say that economic incentives push people toward optimality. Even though people do not individually behave optimally, any instances of non-optimality will be caught out by the market. This is the “as if” argument, no?

Meir:

No, that is not the problem I am pointing out.

The “as if” argument works not badly if our only problem is to optimize given known constraints. Human beings can’t maximize anything, all they have is crude climbing algorithms (as Herb Simon kept on trying to remind us), but they have a bunch of pretty good climbing algorithms and these could give a good approximation to maximization if that was what these algorithms were meant to emulate. (Fortunately for the survival of the human race, the algorithms are not actually meant to maximize our utility but rather our progeny.)

I am not at all denying that the problem you mention is grave and that practitioners extend the methodology beyond the places where it works “not badly” to places where it works badly. But the problem I am discussing is much worse.

In most economic models, the constraints depend on future outcomes, and those future outcomes depend among other things on the current expectations of millions of people. Once upon a time economists liked to actually ask people what those expectations were. That approach worked surprisingly well actually.

But then we had a rational expectations revolution and we adopted a new approach which has iterative models. These algorithms ask the following question: what current expectations are consistent with the current+future behavior which is optimal given the common knowledge of those expectations? This implies also that the expectations are shared. My expectations include my evaluation of what other people’s expectations are, and all these have to be self-consistent.

It is generally an incredible numerical headache to find those. When we do find a “rational expectations equilibrium”, we proceed to assume that those actually have been people’s expectations all along. We basically assume that RIGHT NOW each of millions of people know exactly what each of millions of people expects the future to be like. But we economists don’t know. But if we are extremely clever and solve an extremely intractable numeric iteration, we will know what people think RIGHT NOW.

So there are quite a few issues:

1. Economists like to assume that people are endeavoring to maximize something, but as far as I know there is little evidence that they do or should act that way. People have a variety of sophisticated algorithms which do lots of cool things but they don’t approximate maximizing any one-dimensional function. This assumption is not terrible if we are limiting ourselves to one narrow area of human endeavor. People do choose jobs to maximize their economic well-being and choose houses to maximize some intuitive basket of housing characteristics.

2. Then we assume that people actually find the maximum. Again, this is not a terrrible assumption as long as we don’t assume some kind of ridiculous precision. Though often we do, as your email points out.

3. The assumption that people have “rational expectations” in the sense of expectations that are common knowledge and self-consistent and are close to what some small-dimensional iterative numerical model with highly artificial assumptions will solve for does not seem to me to be remotely defensible. Maybe I am wrong.

I don’t want to spoil your fun, but I am guessing that lots of what you want to say was already said by Herb Simon and they even gave him a Nobel Prize for it but nobody paid attention and anyone who did I guess forgot. Do read some of his stuff.

While I’m in the business of selling gaping holes, here is another one, To the best of my recollection this doesn’t date to Herb Simon but rather back to the early 19th century and the early criticisms of utilitarianism. The following are the two most important insights Jeremy Bentham was selling:

1. A person’s well-being can be reduced to a one-dimensional number which is a function of his/her experiences. “Pushpin is as good as poetry”. Human psychology is basically the understanding that each individual is endeavoring to maximize his/her sensory pleasure.

2. Ethical conduct, to which all human beings should strive, is to maximize the sum total of humanity’s utility.

Wait, Jeremy, do people get utility from their utility or from everyone else’s utility? If the first, then how exactly do you expect them to strive to maximize the second? If the second, then perhaps people are going to spend little time playing pushpin and lots of time trying to bring about prison reform (a favorite of Bentham’s).

Once other people’s utility becomes part of my utility the whole thing blows up. Hopeless to expect you can converge. Consider the following sentiment which is common and perhaps universal: A person sees that his/her significant other is not really investing effort in making him/her self attractive: Feeling inside: “I want you to want me to want you”. This is an everyday sentiment which we all acknowledge and probably almost all of us experience. Try to put it into an equation to maximize! Our consumption is an argument to our utility function, OK. Other people’s consumption is an argument, nu. Now make other people’s UTILITY FUNCTIONS an argument. Sound easy? Now make the dependency of B’s utility function on A’s utility functions an argument in A’s utility function.

Meir continues:

A lot of these critiques have been around for a long time. Some for hundreds of years.

A lot of them turn on the normative/positive tension. Going back and forth between what we claim people do and what we think people should do and what we think is good for them.

If human conduct is not maximizing, then people aren’t making mistakes in the maximization and so utility maximization can’t really be an adequate normative framework. If as Herb Simon points out it is literally impossible that people should be “maximizers” since we are just computers with algorithms, then it is impossible that utility maximization can be an adequate normative paradigm.

Another problem is identifying utilitarianism with libertarianism. There is no particular reason to assume that freedom maximizes people’s utility. If North Koreans weren’t exposed to broadcasts from South Korea they would probably be a lot happier. Their belief that they are living in a heroic paradise led by a demi-god would be unchallenged. But libertarians are the first to favor giving people full information.

Lots to chew on here. I have more to say but I’m not quite sure what, so here’s everyone’s chance to discuss this in comments.

Sometimes it’s hard to talk about economists about this sort of thing because they get all defensive about it.

87 thoughts on “Contradictions within economic theory. All well known but still important and, I think, not taken as seriously as they should be.”

  2. This is hard for economists because they apparently don’t understand dynamics in the slightest. If they stopped thinking in time-free models and started thinking about “given where everyone perceives they are at the moment which direction are they walking in?” A lot of economists confusion would disappear.

    Of course, then economics becomes something a little like turbulent fluid flow and a lot less like engineering statics. Economists don’t like the answer “everything is sensitively dependent on everything else” even though it’s true.

    Reply ↓

      • Maybe. Here’s the course catalog for UC Davis https://local-resources.ucdavis.edu/local_resources/docs/catalog/GenCat20222023.pdf

        Page 185 suggests you can get an undergraduate econ major with a “General” specialty at UC Davis with two semesters of calculus, no course in ODEs, no course in computational dynamic timeseries simulation modeling, actually no course in a general computing language of any sort (ie. R, python, Julia, C++, Java) (I’m assuming the data analysis course would be SPSS based), no course in SQL, nothing in partial differential equations, nothing in feedback control, nothing in stability of dynamical systems, not even a linear algebra course. The one quarter required course in statistics teaches more or less binomial and normal distributions, and 4 or 5 types of confidence interval/tests. Nothing about stuff like simulation of time to failure or even poisson processes or such. In other words basically no course on probability modeling. I’d bet the data analysis course is almost entirely linear regression with OLS in SPSS so it’s not things like diffusion or time to failure or general “probabilistic stuff” especially “processes”.

        Of course, the Econ student is going to get a bunch of labor economics and theory of banking and whatever. lots of “substantive” work in economic ideas, but without even an ODE course, if they discuss dynamics in those courses it’s likely to be at what I would consider a very “basic” level. Very possibly PhD students get into the weeds and start learning a bunch about dynamics, but I’d bet mainly if they’re emphasizing macro-econ.

        Compare that to say a bachelors in Mechanical Engineering

        A graduating undergrad mechanical engineering student interested in mechanical design emphasis (pg 303 of same catalog) would have taken 3 sessions of calculus, a session of vector calculus, a session of linear algebra, a session of differential equations, three sessions of classical physics, a course in electrical circuits a course in computer programming for applications involving simulation, a course in engineering dynamics, a course in fluid mechanics, a course in thermodynamics, a course in thermo-fluid interaction dynamics, a course in heat transfer, a course in automated control systems, a design course where they would build some engineered controlled system, a course in one of (PDEs, advanced scientific computation, numerical analysis of ODEs, probabilistic methods …) and then for their specialized design emphasis they’d take one of {mechanical vibrations, engineering applications of dynamics, stability of flexible dynamical systems, mechanical design, mechatronics, and analysis and simulation of mechatronics}, they’d also have to choose some courses in something like propulsion, orbital mechanics, vehicle stability, HVAC, statistical methods in manufacturing, … etc

        So basically they’re doing dynamics in something like what 10-15 classes? They start with dynamics in their very first quarter, and continue to take one or more course involving dynamics every quarter for 4-5 years. At least 2 or 3 of those involving programming computers to set up and solve dynamical systems. At least several of those courses involve feedback control. Likely 1 or 2 courses involve probabilistic methods or statistical methods. And several of these involve **designing** systems to meet dynamic goals.

        Along with these textbook courses, they actually do laboratory courses where they measure real world vibrations, forces, changes in temperature, control systems for temperature or vibration of wings or whatever.

        I’d guess that the average mechanical engineering undergrad from Davis has had 10x the exposure to dynamics than the average PhD in Econ.

        That’s a gulf that seems HUGE to me.

        I don’t think this is because Econ students **can’t** do dynamics, I think it’s because Economics apparently doesn’t put a lot of importance on topics in dynamics. It seems to be a niche topic for those interested in macroeconomic modeling for the Federal Reserve or something similar. This is admittedly an outsider perspective, but I’m an insider when it comes to Civil Engineering at UCD, so I can say confidently that a graduate in the top half of engineering at UCD understands a considerable amount of dynamics.

        If you wanted someone to build a computer model describing how things change in time including some probabilistic variation through time, which would you hire, an MS in economics from UCD or an undergrad in Mechanical Engineering?

        I don’t think it’s even close.

        Reply ↓

        • “If you wanted someone to build a computer model describing how things change in time including some probabilistic variation through time, which would you hire, an MS in economics from UCD or an undergrad in Mechanical Engineering?”

          Obviously it depends on the topic. If those “things” are related to economics then I’d hire the economist every time. You’d be a fool not to.

          Apparently engineers’ obsession with dynamics has warped their perspective so much that they think “everything [in the economy] is sensitively dependent on everything else.” (Do you have any evidence whatsoever for that claim? Do engineers not learn how to use data? Do engineers not learn how to do a literature review?) And that confusion is going to cripple them in the face of easy problems that an economist can solve in a few minutes by drawing a shift of a demand curve.

          And yes, if you hire a PhD economist to do macroeconomic research, they will do dynamics.

          (Just to avoid confusion, we are two separate Wills.)

        • Robby, sorry I meant to say Stata. Thanks

          Will, the price of eggs to take a current example, is dependent on the choices made by Russian oligarchs, the virus load of migrating birds, the legal policies on enforcing antitrust laws, which are dependent on donations made by special interests to particular political campaigns, the location where you want to purchase the eggs, the local availability of fuel for refrigerated transport, whether you grow the chickens in your backyard or not, how long til the expiration date of that batch, etc etc.

        • “everything [in the economy] is sensitively dependent on everything else.” (Do you have any evidence whatsoever for that claim? Do engineers not learn how to use data? Do engineers not learn how to do a literature review?)

          I dunno about “sensitively dependent” but “everything is correlated with everything else” should definitely be the default.

          To continue the egg price example though, evidence should be provided that something *does not* influence the price of eggs. The standard burden of proof is the reverse of what it should be.

          But really whether there is a correlation/causation or not is the wrong type of question altogether. We want to know the magnitude of the various relationships under a variety of circumstances.

        • A couple misunderstandings

          There’s a peculiarity to economics in that economics undergraduates are generally terribly prepared to become economics PhDs. More than half of an incoming class of PhDs at a top institution have a bachelors in math, physics, or statistics; undergrads who only have an econ degree are at a severe disadvantage. To make up this gap, the summer before classes begin the incoming class of PhDs will attend a crash course called “math camp” to catch them up. Anyways, the undergrad curriculum in economics tells you much less about the capabilities of its researchers than it would for other fields. It’s definitely not ideal, but it’s an artifact of Samuelson’s mathematicization of economics paired with the role of economics as a “default major” for aimless drifters becoming bankers at their father’s institutions.

          There may also be a confusion about the phrase “time free” and “dynamic”. An intertemporal choice model has time in the model, but the way physicists use it, it is still a static model. Any model in which you solve for an equilibrium like “demand = supply”, or “utility is maximized” or “excess demand function goes to zero” and your prediction is the consequent price or consumption bundle or price vector would be a static model. In a DSGE model, the fluctuations in macroeconomic variables are driven only by stochastic shocks. So yeah, macroeconomic variables will evolve in time, but I wouldn’t consider it “dynamic” in the sense of a math class in “nonlinear dynamics”. A model in which markets do not clear, but firms increase prices when there’s excess demand and decrease supply/increase prices when there’s spoilage and agents hill climb towards their optimal consumption bundle would be a “dynamic model” in the sense that Daniel means it, and it is a bit heterodox for economics.

          I wouldn’t agree that it’s because the economists lack the understanding, though I do think economists’ have been slow to train their students in flexible computational methods. As I understand it, there were plenty of early investigations into “how do we get to equilibrium”, both with simulation and mathematics. But my understanding is that these were pretty discouraging. Simple exchange economies with tantonment would never reach equilibrium. Sonnenstein Mantel Debreau showed the excess demand function could be arbitrarily poorly behaved. In my (very weakly informed) opinion, the subsequent abandonment of dynamics was less due to an inability to adapt and more due to a desire for models to be beautiful, at least quasistable, and pareto optimal. Contrast this with weather forecasting, where it’s acknowledged that the problem is just ugly, unstable, and basically impossible to forecast more than 10 days out.

        • somebody.

          Thanks for all those clarifying remarks. It’s interesting to me that Econ undergrads are not generally qualified to be on track for Econ PhDs… perhaps that’s a lot of what concerns me! Undergrads should come out of an Econ series prepared to at least do some kind of Economics data analysis and dynamic modeling. Like they should be able to work for a toy company and at least help more experienced economists decide how many action figures they should contract to produce based on the upcoming release of the latest Toy Story movie or something.

          Also all the commentary about static vs dynamic is spot on. Thanks for clarifying that. In reality, the spot price of corn looks like: https://markets.businessinsider.com/commodities/corn-price

          There is no “clearing price” it just moves all over the place constantly, between 5.75 and 8.50 over the last year. If you’re “solving for the price” as opposed to “solving for the rate of change of the price” then you’re doing a static “time free” model as I and “somebody” understand that term.

        • Daniel: 1) Econ Undergraduates are indeed not well prepared to be econ PhDs, and haven’t been since at least 1970. Of course, I would also argue that *most* undergraduate majors are not at all prepared to be PhDs in their subjects, either. Since only a tiny fraction of them actually try to go grad school, that’s just supplying courses to meet the demand. If udergraduate econ actually required their graduates to be good enough when they graduated to go to grad school, economics programs would be about 3% of their current size. That would put a lot of econ PhDs out of work!
          That doesn’t mean that you can’t graduate from a good institution prepared for econ grad school… you can. But you’ve got to take a lot more rigorous courses than the typical major does.
          2) Like I said on this thread, dynamic models of the sort you’re talking about are rare. What you get are regression models and static predictions of those regression models. So when you say that the corn market is never is in equilibrium, what you will be told is that it’s in equilibrium every damn day, or indeed, if you believe the Efficient Markets Hypothesis, every damn minute, at least within a margin of error caused by underlying transactions costs. That point of view is, somewhat famously, completely unfalsifiable (check out the Grossman-Stigler Paradox) but that doesn’t mean it’s wrong. It just means that it’s a way of looking at the world and it’s a way economists are entirely comfortable with.
          3) So there are indeed MA theses on electricity price formation…. Lots of them. I’ve hired around ten people over the last 30 years who wrote them, and failed to hire a few dozen more! None of them are dynamic in the Lakeland Sense.

        • Of course, I would also argue that *most* undergraduate majors are not at all prepared to be PhDs in their subjects, either.

          I think it’s a lot worse in economics. An economics department *might* suggest taking one proof based class. Then half of the first year of PhD economics is nothing but proofs, and assumes familiarity with real analysis! It’s a pedagogical embarrassment.

        • I’d have no qualms about hiring a good mechanical engineering undergrad to work on building a control system for drone cameras for nature photography. With appropriate mentoring they could pretty rapidly be writing code to reduce vibration of the camera image through extreme telephoto lenses or such. Industrial research and development basically. Sure, you might get them up to speed faster if they had a master’s but the undergrad would have the foundation.

          Now it sounds to me like you wouldn’t hire a typical Econ undergrad to do any applied econ at all, you’d wait for them to go get a PhD, and 50% of those PhDs would have an undergrad degree in Math or Physics or such. That’s just weird to me. There’s a two tier system in Econ I guess, you teach different stuff to the two tiers, and that seems like a major problem for those undergrads who think they’re learning economics, and the people who hire them thinking theyre hiring well educated “baby” economists. Sure all fields have had credential inflation, but I do think you can hire a good undergrad engineer, physicist, chemist, mathematician, computer scientist, biologist, and expect them to know enough to do industrial research supervised by a PhD perhaps but with the fundamental skills in place. If you want them to design the research program, sure you’d look for PhD level.

          Johnathan: When it comes to 40% of an islands power being provided by solar, and clouds coming and going throughout the day, the quantity of power being available changes on a minute by minute basis. Models of such things would have to be able to account for large and rapid fluctuations, and/or to quantify the value of mechanisms such as batteries which might smooth out variation, which still requires to be able to compute the value with and without those expensive devices. Something similar would need to be done on those “save power days” in SoCal where they’ll pay you to cut your power consumption and avoid rolling blackouts. How much incentive do you have to offer? And what timescale do you have to offer it ahead of the crisis level?

          When it comes to saying that the market is by definition in equilibrium at all times… now it’s just redefining words. Equilibrium has a well defined meaning… That all factors which might cause something to change cancel out and the state of the system remains constant through time until one of those factors changes. As you say, it’s unfalsifiable and hence nonscientific to make that claim. You can just say “look the price changed so demand or supply curves must have changed!” Since you can’t observe demand and supply curves noone can prove you wrong!

          In any case HFT traders trade commodities on millisecond timescales and they spend enormous amounts of money to make that happen, so it must be worth it to them. I can’t imagine they don’t have models for the direction and magnitude of predicted price changes over the next few epsilons of time… Dynamic models in other words.

          Economics and Finance are not the same thing obviously, but they’re related, and it seems some cross pollination might help things.

          You might ask why bother with dynamics in economics? But there are lots of interesting dynamic questions. A few years ago Phil asked about how the distribution of rents in SF apartments would change if various quantities of new housing were made available. Unless we look at agent level models and cascading avalanches of people moving and prices of vacated apartments resetting to market rates in rent controlled locations, and then other people moving into those apartments and such there’s not going to be a good model of that distribution change.

          There are tons and tons of interesting questions about what would happen if things changed … Policy changes, natural disasters, Ukraine invasions, pandemics, etc. If an event causes a cascade of things to happen you can say at every moment it was in equilibrium but it’s not going to answer the interesting questions like “how long will it take until the supply of eggs reaches its previous level again?” or “what will be the trajectory of the SF rent distribution probability density function over the next 12 months after we change x policy?”

        • 1. My former firm hired lots of BA economics majors as research assistants. We never expected any of them to be baby economists, except in the broadest possible sense, e.g. familiarity with concepts.
          2. Supply and demand for electricity change every to minute, but prices don’t. The effects of electricity storage, and transmission line losses, and the full AC power solutions of electric systems are, to my knowledge, not solved even today in the models that actually generate prices. Those prices come from static approximations to full AC models.

        • In places like the US the maximum supply and demand change every minute, but it’s a smooth function where the derivative magnitude is never something like 3% of maximum capacity per minute, probably more like .04% per minute based on quick calculations from https://www.caiso.com/TodaysOutlook/Pages/default.aspx

          Nevertheless, spot prices for the western united States range from -$5 to $125 (per megawatt hour I think) the negative prices are mostly in low density regions like southern oregon where maybe solar is producing excess, or reservoirs need to dump water, not sure? And the high prices are near Lakeport, or Ukiah (cannabis growers?).

          https://www.caiso.com/todaysoutlook/Pages/prices.html

          However the generating capacity mix is not concentrated into an island where 100% of the solar capacity can be covered in clouds in minutes. Instead the generating capacity is diversified across not only all of CA but actually all of the western united States since CA can import and export power to other states. It’s super convenient to have diversification, and it sucks to be Texas that has isolated it’s power grid from the rest of the US. If it sucks to be Texas, it really could suck power wise to be Rarotonga, in the cook islands, subject to the whims of cloud cover. That’s what I was going for.

          A Civil engineer out of an undergrad program is not expected to sign off on beams and columns and footings, but they are expected to follow design procedures and do the design and a more senior person would check the design calcs. In fact sometimes the licensed engineers who sign off are often considered to be people you wouldn’t want anywhere near a calculator.

          It sounds like economists consider PhDs in a completely different class than the masters and undergrads. That’s not the case in math, biology, engineering, chemistry, physics, etc. If you tell an undergrad mathematics major “here’s an ODE that a chemical engineer came up with for our reactor, can you choose some ODE solving software program up this system of equations and run it for a variety of conditions and give us a summary of it’s behavior, by the way it’s bad if the temperature gets above 400K or reaction product C goes above 1 micromolar can you give us an idea of what conditions could cause that?” I would be extremely disappointed if that wasn’t an easy task, they’d probably choose Matlab, Julia, python, R or Mathematica. It might take them a couple days to produce a nice report, with graphs.

          If you asked an undergrad economics major straight out of UCD to grab the ACS microdata for the last 5 year period (multi gigabytes) and determine which zip codes of the United States have the most couples with two masters degrees or more, and model the AGI in those regions as a function of age, income, number of children in the household, and educational attainment, and answer some questions about whether households with high educational attainment who have children are under employed compared to those without, it might be not something they actually could do? Especially if they didn’t have Stata available? If you asked them to estimate how much a childcare allowance of $1000/mo from the govt would affect tax revenue by increasing the employment of highly educated stay at home parents… Would they have a chance of doing a good job?

        • We’re too far down in the comments to have a real discussion of price determination in the CAISO, or PJM, or ERCOT. (By the way, the diversity advantages of ERCOT expanding their interconnections is surprisingly small.) But the huge difference between Rarotonga and California is in the reserve margin required to keep the lights on. Small islands routinely have more than twice the capacity they need at peak, while reserve margins in large systems are optimally in the range of about 10 percent. (To find out what optimal means in this context, look up Telson, Rand Journal of Economics, 1975)
          But the point I was making was that the prices in these systems are actually derived by running static approximations of the constrained power flows and then calculating marginal costs at each node of the network. There are very few even pseudodynamic constraints — things like the speed by which individual generating units can ramp up or down to match load are actually compensated outside of the prices you show.
          And to your final question, the answer is no friggin’ way.

        • Daniel
          From my experience, the tasks you outline could only be done by a small fraction of econ undergrad degrees. Most Masters level students could probably do these tasks, with varying quality. Virtually all PhDs could, but again with varying quality. If you would advocate that all people with econ degrees should be able to do these things, I wouldn’t disagree. You should realize, however, that much of the undergrad economics coursework is of a different nature – focused on topics such as environmental econ, labor econ, history of thought, economic development, money and banking, international econ, and a set of theory courses. Some of these require empirical work, but not all. Too many, in my opinion, apply a standard set of economic theoretical tools to a wide variety of issues with the basic message that markets are good, and government interventions should only be when there are large market failures and even then, often are counterproductive. The limiting factor in the amount of empirical work in these courses is the ability of the students – and that varies tremendously across universities. The best will have undergraduate coursework that looks a lot like graduate work. The worst will look nothing like that.

          I’d speculate that engineering degrees vary a lot less in quality than econ degrees. Similarly, accounting degrees show less variation. My belief is that having standardized professional testing forces a lot of standardization. Economics has no such external licensing. Nor does statistics (although actuarial science certainly does). Would such licensing improve things? I believe it would, despite my dislike for much of the objective testing that is done.

        • It just occurred to me that Supply Function Equilibria creates a great example of what you’re talikng about. Klemperer and Meyer (1989) invented the ODE explaining the equilibrium of firms bidding supply curves into a market characterized by a known demand which varies with time (sounds like electricity markets!). Economists did virtually nothing with this model (beyond talking about it a lot) and the first glimmerings of a practical solution appeared in 2008 not in an economics journal, but in Operations Research (Anderson and Hu, 2008)

        • Dale, Jonathan,

          Thanks for the discussion. Overall, I think this is more or less tragic.

          Let’s go back to the UCD catalog. A Math major + CS minor (that’s what I did in my first time through undergraduate degree) would take several quarters of calculus, a linear algebra course, a course in differential equations, UCD has MAT 107 a course in probability and stochastic process with applications to biology, several PDE courses available, several numerical methods classes, an optimization course, a complex functions course, a course on database theory MAT170, maybe some sort of thesis project, and then in computer science likely a couple programming classes, data structures and algorithms, computing theory and formal languages, maybe a couple courses on scientific computation or bioinformatics, scientific visualization, maybe statistical data science 035A and 035B and 035C which would go well beyond what the Econ course requirement was and would include Bayesian methods, PCA methods, clustering, bootstraps, regression, etc.

          Ok, so by this point the person would have ZERO coursework in economic theory. But math+CS students need “breadth” requirements, so let’s assume the person has some interest in economic, and give that person let’s say ECN001A and ECN001B (principles of micro and macro) + ECN100A and ECN100B intermediate micro-theory

          If you hand them a Linux machine and tell them to install Julia, add CSV.jl, DataFrames.jl, Plots.jl, GLM.jl and a couple other packages, download the ACS, read it in, comprehensively munge that data into a dataset relevant to the question about dual advanced degree households in various zip codes and ask them to run OLS regressions on those variables and give naive estimates of the tax revenue change without concern for causal identification issues… I think that’d be pretty straightforward, and I think if you mention the causal identification issues and gave some suggestions as to what data to look for and how to do a basic job they’d take a stab at that without too much difficulty. If you asked them to use optimization techniques to try to find the optimal child-care subsidy to minimize the total cost to the govt they’d have a go at it and find it fun and probably would get some interesting results within a month or so.

          Would they understand all the nuances of various economic issues? No. But they’d be able to **do** economic data analysis and answer economic questions. They’d be much less able to **formulate** economic questions of course. Perhaps this is why most Econ PhDs wind up coming from Math or Stats or Physics.

        • The survey is almost 20 years old now, but Colander https://pubs.aeaweb.org/doi/pdfplus/10.1257/0895330053147976 reports that 81 percent of economics PhD students at the top 7 programs majored in economics. So it’s not like all economics students are math or physics majors. It’s just that those who go to grad school probably took a lot more math-oriented courses as undergraduates, whether in the economics deparment or elsewhere.

        • Daniel
          I find your description frightening. Anybody that takes all those math and computer courses will certainly be smart, at least in a narrow sense. The fact that they could calculate an “optimal” child care subsidy worries me greatly. All those “nuances” are actually important. Let me say it again: they are important, and not less important than the ability to code or optimize. Economics is a social science and must consider implications in a broad sense – as well as having a number of skills that you do mention. The hierarchical ordering of skills from math to whatever is not something I can agree with.

          Also, the list of courses you have for math major + CS minor is not realistic. These may all be choices available, but required courses for the major have limits – actual limits that institutions and accreditors impose (regardless of whether I agree with those). In a 120 or 128 credit degree, there is a limit to the number of courses a major can require, and while STEM subjects generally require more than humanities or social sciences, they still face limits. More importantly, some of the courses you list seem less important to me than some that are missing – philosophy, sociology, anthropology, and psychology for some examples [note here: I consider those subjects important, but not the way the introductory courses are taught at most places – so the current offering of these subjects might be something I’d be willing to forgo, but the lack of a serous exposure to these subjects is sorely missing among those optimizers trained as you describe).

        • We are deep in the weeds of the comments section, but I don’t disagree with you Dale. I am not trying to minimize the importance of the nuances. My point was more that if you had someone with a PhD mentoring and describing the nuances a math+cs person would be able to do the work. By admission of Jonathan the econ major would maybe have a better sense of the nuances but wouldn’t have any chance of being able to do the work… That seems problematic.

          Also there are plenty of examples of PhD econ analyses here at the blog where PhDs didn’t get the fundamentals right… For example the coal across the river analysis or the Mekong delta analysis… So… :-/

          Also I should say that the UCD system uses quarter system there are 3 quarters and a summer quarter. A Math+CS student would take 4 classes a quarter, so between 40-50 classes and probably 10-15 of them would be writing, literature, philosophy, social science, or a science series like biology or physics or something. The quarter system allows for somewhat more variability in interest but is rather fast paced.

          It wouldn’t be surprising at all for a math major to skip first and second semester calculus because of AP courses in high school. So

          Calc 3, odes, linear algebra, stats 1,2,3, probability and random processes, intro to PDE, numerical analysis, optimization, complex var calculus, Fourier analysis… That’s all very reasonable for math classes. Probably there would be 2 or 3 more.

          Programming 1,2, algorithms and data structures, theory of computing, database systems… Those are all very doable for a cs minor. So far we are talking 17 classes, add 15 more for general Ed requirements you still have 13 more. A person interested in econ could take the 4 courses I mentioned, some of which would count for general Ed, and as such probably still have like 10 classes free to choose.

        • Jonathan
          That old survey has a few notable things. First, 62% of the students in these elite PhD economics programs were foreign – the importance of that fact is that I think the economics major outside the US is quite a bit more rigorous than the domestic undergrad degree. Then there is this:

          “The large majority (81 percent) had majored in economics as undergraduates, while 21 percent had majored in mathematics and 22 percent had other majors. (A number of students had double majors.)

          I had a good laugh until I got to the parenthetical.

        • A couple of anecdotes.

          In my introduction to microeconomics class, I was taught that agents maximize their utility subject to a budget constraint. We had a cobb douglas utility function with elasticity of substitution alpha, we took a derivative, we set it equal to zero, it was great. I asked “so how does this happen? Like, people guess at a consumption bundle (x0, y0) centered on the optimum, and they get closer as they make more similar decisions?” The instructor stared blankly at me and replied “what they choose is the optimum.”

          I once asked Barry Eichengreen after a class on the history of the global economy whether one of a family of models would be helpful for understanding this class. He told me “I don’t think you’ll look at any point in this history and go ‘hey look! It’s general equilibrium'”

          I do think economists could stand to learn more history and philosophy. I don’t think they necessarily need to do more math — stuff like monotone comparative statics is impressive enough. I think they need to do different math, and stuff that’s often quite a bit easier than what they’re expected to do now.

        • somebody: thanks for those anecdotes. here’s one of my own.

          I took an intermediate microeconomics course in my 3rd year undergrad I think. I just found the textbook, Walter Nicholson: Microeconomic Theory (Dryden press). It was a 1990’s edition, I guess he first wrote it in the 70’s and updated it multiple times.

          The first … half of the course was teaching what was a function of 2 variables, how did a gradient work, how did lagrange multipliers work, how can you optimize a utility function of two quantities of economic goods subject to a linear budget constraint.

          then the second half of the course was looking at some examples… changes in prices, taxation and who bears the economic cost, monopoly pricing, oligopoly pricing… whatever I don’t remember exactly.

          My lasting impression was that the first 7 weeks or something of this course was a complete repeat of about 2 weeks worth of lectures I’d had in multivariate calculus three years earlier. The second half of the course was a set of examples that anyone who had taken multivariate calculus could have understood from a couple graphs and a nice 6 page handout one page per example. It was basically just moving the budget line around and looking at the shift of the maximization point.

          The ultimate outcome was that econ students got a poor and restricted understanding of multivariate calculus and a mathematical veneer over some intuitive concepts like “if the price goes up you buy less of that product” and “if you really want a good, and the government taxes it, then you’ll pay the tax and the supplier is unharmed, but if you hardly care about that good, you’ll just buy less and the supplier is harmed”.

          There was absolutely ZERO way you would use the content of that course to do something like figure out what would happen to the price of eggs if half the laying hens in the US were killed due to bird flu, except that you’d expect the price would go up and the quantity demanded would go down, the math added very little.

        • Daniel
          You nailed it. That is exactly what much of undergraduate economics is. The worst thing is that, armed with that veneer, students then are shown how you can take that apparatus and then find optimal policies for environmental protection, regulation of public utilities, income taxes, etc. But I agree with somebody here, more math is not what is needed. And I don’t think it is more dynamics either – except when the problem calls for it. Economics texts are the worst offenders.

          My favorite example is a Managerial Economics text (at one time the most popular) that had an example for estimating a demand curve – it had 10 made up data points with 3 variables that could affect demand (price being one, of course). The key was to recognize that one variable, advertising, had a regression coefficient that was not significant (p>.05), and the message was that it did not affect demand. If you looked that point estimate, however, it revealed a return on spending that was around $5 of revenue for every dollar spent. Since the coefficient was not significant, however, the confidence interval included some negative numbers. I used this example (not the text, however) to discuss with classes how crazy it would be not to increase advertising. And if your advertising was actually hurting sales, then surely you should be able to recognize that and change your advertising. After all, it was a textbook on Managerial Economics.

        • Dale, here’s my take, which isn’t super far from your take… In the case of that microecon course, it never should have existed. Instead, send econ students to a multivariate calculus class in their 3rd semester. With that under their belt, instead of a 15 week course on a subset of multivariate calculus + some examples. Start out with them understanding enough about calculus that you can set up that basic problem and solve it a few times under different conditions and then in week 4 you can say … All right that’s the idealized toy example… Now let’s look at some real world conditions… And pull data from commodities markets or real world example firms or changes in taxation or whatever, and start to LOOK AT WHAT HAPPENED you know, like a scientist. And compare the simplified models to what goes on in the world… And maybe talk a bunch about how you could estimate or predict things.

          As it is, economics at the undergrad level as I experienced it is more like a seminary school where you learn the religious doctrine. The holy ghost and the individual utility function have exactly the same relationship to the world… Unobservable.

          In physics class there are a ton of labs where people drop balls and roll things down ramps and heat water up or observe electron beams or waves in a string or whatever… Chemistry is the same. Biology etc. Of course you can’t do lab experiments so easily in econ but there is a TON of public econ data out there, so working with real world data and comparing idealized models to reality needs to be part of that I think.

        • Not to belabor the points, but you can do experiments in econ. In fact, there are a few online sites set up to run these experiments in courses very easily. One of the best was set up by Paul Romer (Nobel winner and somewhat well known for saying there was too much unnecessary math in econ) with slick experiments run on the web (Aplia – no longer exists). I used them when they first came out, but they never got updated nor were new experiments added. When I inquired about this, I was told that most economists that used that site only used the exercises (computer graded of course) and didn’t use the experiments. Apparently most (certainly not all) economists prefer their toy examples to running real experiments. After all, the latter might work out differently than the theory would predict (although most of the time they were consistent, but raised difficult issues, like what determines the time to find equilibrium).

        • To the extent you can do experiments even better… And if the experiments show that time-to-equilibrium is an issue, then… hey that’s evidence of interesting dynamics.

          Teaching undergrad econ majors to ignore the real world is bad, to say the least.

    • If only an entire field was astute enough to see things the way you do and avoid this incredibly simple flaw in their uniform reasoning. Alas, your incorrect assertions about the ubiquity of “time-free” models and comparisons to engineering statics will have to continue unabated.

      Reply ↓

      • Clearly some economists do dynamics. And clearly I’m being intentionally controversial. But I’m doing so not because I think economists are stupid or anything like that, it’s because I actually really care about economics and think they could be doing a lot better than they are.

        Most economists have a Bachelors or Master’s degree. They would go through a program similar to the one in the UCD catalog linked above. Apparently they would not have had a course in linear algebra, ODEs, PDEs, agent based models, computer programming other than Stata, numerical analysis, probability and random processes…

        If you are a company planning deployment of solar electric generation, and ask a recent UCD grad Master’s in Econ what is likely to happen to the spot price of electricity based on dynamic changes in cloud cover over a major solar farm about to go online providing 40% of the generating capacity of an island with population 2M and a number of fish canneries and related industries … How well would they handle that task? I don’t see evidence in course listings that the answer would be “sure that’s pretty typical of the kinds of projects we have 3rd year undergrads do as final projects in such and such a course.”

        That could be my ignorance, please show me I’m wrong. I would love to be wrong!

        Reply ↓

        • “Most economists have a Bachelors or Master’s degree”

          As it pertains to “what economists think”, I think it’s fair to focus only on PhD holding economists. A PhD in economics will require coursework in linear algebra, real analysis, probability and statistics, computer programming, some amount of measure theory and then additional math depending on what you specialize in. In fact, I find this particular criticism of economics kind of funny because another common criticism of economic pedagogy is that economics requires too much math! At this point it’s pretty common for PhDs to come into programs with at least a math major or sometimes a math masters degree.

        • I’ll throw my somewhat dated hat in here – I went to grad school a long time ago, so some of this may have changed. The top PhD programs in Econ are so mathematical that most econ undergrads are not adequately prepared – math or physics undergrads have an advantage there. Opinions will vary as to whether this is good or bad (I am more in the latter camp). As for dynamic modeling, there are many different degrees of dynamics. Fully dynamic models are rare – they are difficult and, in many cases, not necessary. I have criticized some models in an applied setting because they assumed away important dynamic considerations.

          Example: in telecom regulatory cases, the FCC has often applied a standard of what a network would cost if it were built instantaneously on a blank slate – but real networks are built over time, and this has implications for things like excess capacity that is inevitable. This is a dynamic consideration, but need not involve a complex dynamic model. For other cases, such as high volume stock market trading, a much more dynamic model is required. And, the people working on these models are usually not economists but have backgrounds in physics (the so-called “rocket scientists”). Finance PhDs are an interesting in-between case: they are rare (because so much money can be earned with a Master’s degree in finance) and most don’t have the math background to do that complex modeling. So, they are often working with a team that will have physicists (or comparably trained people).

          The choice to approach markets as “never” in equilibrium or “always” in equilibrium is more than semantics. There is a range between these extremes, and I’d suggest that the appropriate modeling approach depends on the applied problem. I think there are some theorists (I’ve forgotten who, but probably had to study them in grad school but struggled due to my inadequate math background) who attempt to build fully dynamic models that would erase the distinction. To predict the impact of a tariff on the corn market, it might suffice to have a static model that assumes you move from one equilibrium to another, but ignores the transition path. To trade corn market futures, you would worry about that transition path and require a more explicit dynamic model.

          I don’t think it makes much sense to argue about whether economics has failed because the models are not fully dynamic or whether it is fine that economics focuses on equilibrium and ignores disequilibrium behavior. If there is a systemic failure in economics (and I think there is), it is that economists are inadequately trained to recognize the right degree of dynamism required for any particular problem they are trying to model. I think this results (in large part) to the focus on general solutions that can be solved mathematically rather than a focus on particular applications that rely on computation rather than closed form solutions (this last conjecture may have improved since I was in grad school).

        • Dale, again I agree with you. Choosing the right degree of dynamism is the important thing and you can’t do that if you don’t know much about dynamics.

    • I’m sympathetic to your discussion of dynamics, but it’s not so easy a problem.

      To your other points, pulling up courses and discussing what an economics undergrad should be able to do, I think it’s important to emphasize that only a small percentage of people who get undergraduate degrees in economics work as “Economists”. That specific line of work has relatively smaller demand (though well paid in industry).

      The undergraduate degree is more like an alternative to a business/management degree that is broader and less specialized. While there is less demand for “Economists”, there is a lot of demand for people with this background as people who get that degree tend to have similar average salary outcomes as those with accounting or finance agrees (which tends to come after engineering or computer science). The problem is that if you are actually interested in being an “Economist”, it doesn’t do the best job preparing you for a PhD level workload.

      Reply ↓

      • Yes, I’m learning that a lot of people apparently get economics degrees as kind of an alternative to business or operations research or something. Well, I suppose that suggests there should maybe be a BA vs a BS, or something with the BS having more emphasis on doing economics type stuff as opposed to just having some economics knowledge and then doing something else… Running a local restaurant chain or developing new clothing brands or handling contracts for construction companies or something… Who knows.

        I still don’t think courses like that microeconomic course I described above do anyone any favors. Divorced from reality they are a poor alternative to vector calculus, and they give the mistaken impression that there is a real mathematical basis for a bunch of policy ideas that isn’t actually supported by any of the course material. Perhaps supported by other info but its not presented in the course.

        Reply ↓

        • “they give the mistaken impression that there is a real mathematical basis for a bunch of policy ideas that isn’t actually supported by any of the course material”

          I’m sympathetic to this. If anything, Economists should be humble about their ability to predict the effects of changes in public policy, particularly those related to macroeconomics. However, there are some real insights of basic supply/demand curves that can provide a useful framework for thinking about the world. These more qualitative aspects of economics can sometimes get lost sometimes in a more quantitative course (at the same time it is also frustrating when you’re not quite sure what your professors model really is).

        • No doubt supply and demand curves and models of budget constraints in two good economies are worth seeing. Perhaps 3-4 weeks. But an entire semester course, half of which is a poor substitute for multivariate calculus and half of which is toy models analyzed to death with *literally zero* empirical connection… It feels like a physics class where you analyze the flight of a golf ball through the air after rolling down a ramp, after rolling down a ramp into a bucket of water, after rolling down a ramp with a ski jump at the end, after a ski jump into an oncoming stream of air, after ricocheting off an anvil…

          Oh and by the way, we never actually touch a golf ball, and in fact we never even watch a video of someone with the experimental apparatus, and also the models only work if the golf ball has a dense core and we ignore the moment of inertia, also the dimples aren’t there, also the apparatus is on the moon.

        • I largely agree with your complaints about the micro econ course – but I probably diverge from your opinion that replacing it with a good vector calculus course + some economics context is the solution. For managerial economics courses, I have eschewed the calculus approach entirely. It helps understand how the economic paradigm works (such as how a particular change might perturb the equilibrium – whether it be a comparative statics result or a more complex dynamic adjustment model), but not the more important question of whether that paradigm is appropriate or applies under the circumstances.

          Example: a trade war is threatened and soybean farmers want to understand the potential impact on soybean prices (this is an example from my text). Real price data can be used to construct a time series model for the behavior of soybean prices. Then a variety of sources can be used to estimate how much soybean exports might be impacted and what the possible elasticities of demand might be. A simulation model can be run to show probabilistic forecasts of soybean prices after a potential trade war (that limits exports to particular markets). All of this can be done in a spreadsheet environment. Of course it can be done via R or Python or Matlab, and there is nothing wrong with those modeling approaches. But I think calculus adds virtually nothing to exploring such a problem. It could be useful if an approach is adopted that requires the use of calculus – but there is nothing in such a problem that requires such an approach, and I personally don’t think it would be a more valuable way to approach such a question. An important part of such an analysis is to use realistic data and assumptions about the relevant variables – and these often conflict with the assumptions required to use calculus in the model.

          I maintain that many, if not most, microeconomic questions are like this. And, in 40+ years of consulting work, I have never used calculus in any of my models, I’ve only used it in publications.

        • Dale.

          I largely agree with you. My point was more **if** you want your students to understand multivariable calculus based models, **then** you’re better off just having them take multivariable calculus, and then the content of the later courses can use that background and you don’t spend a bunch of time just teaching a poor version of multivariable calculus and you provide a lot more real context to the economic components.

          On the other hand, I think you’re right you could argue that discrete time processes are likely fine for almost all economics modeling. The vast majority of what you’d be teaching undergrad econ students would be fine if modeled on a daily or weekly or monthly timescale (depending on the topic).

          My favorite example is The Digital Provide: https://web.stanford.edu/class/comm1a/readings/jensen-digital-divide.pdf

          It shows that from the time mobile phones were available to time of full steady state adoption was ~ 20 weeks, and a similar time scale for the massive reductions in fish price volatility that were observed.

          This was a REALLY FAST change that had clear significant benefits to the market participants. But it still took people observing that other people around them were using cell phones, and deciding based on what they see that the cell phone is a worthwhile new thing to have, and some people are early adopters and some people are late adopters etc. If you had a discrete time model that worked on daily data… it would be a dynamic model far more detailed than the average “here’s the new equilibrium” type solution. Of course if you’re interested in a decade long timeseries then you could ignore the dynamics. But this is the kind of stuff I mean by dynamics. How long from the introduction of Facebook ads to the full development of the Bangladeshi click farm industry?

          “Solve for a new equilibrium price” doesn’t do that. But you’re right that calculus per se is not necessarily needed.

  3. Too easy to get buried in the weeds here – I think the bigger (50,000 foot level) issue concerns methodology, and it is the “as if” methodology that matters most. While the debates about utility, etc. go back further, my familiarity with “as if” dates to Milton Friedman. In response to criticism of some of his early work on the household consumption function, he responded that people don’t need to do any of the calculations the models assume, they only need to act “as if” they did. I always bristled at that broad open-ended statement, but it is more subtle than it first appears. After all, all models are wrong, but they may be useful. And many of those models are useful.

    What I think the rational expectations models raise is mostly the issue of how far we want to push this “as if” methodology. The more we look at situations with imperfect, and especially asymmetric, information, the more strained the as if methodology becomes. The behavioral economics school and much of psychology shows us that people behave strangely when they face uncertainty, and particularly when they have to transact with entities that have better information than they have. In such cases, models that specify optimizing behavior on the part of economic agents become increasingly suspect (at least to me).

    Static vs dynamic behavior is yet another branch of this – the more imperfect and asymmetric the information, the larger the potential gap between static and dynamic modeling (conjecture on my part). Since rational expectations is a methodology specifically dealing with imperfect information, I think that is an area where the as if methodology deserves to be critically assessed. I have not done so, given that I view macroeconomics as the black arts and try to avoid it as much as possible.

    Reply ↓

    • IMHO each economic actor acts with a very imperfect view of the world. This also plays into Tom Passins point about buying houses. One of the things about imperfect information is you can’t assign value to a thing you don’t know is even possible.

      As economic agents move around through life they learn about the world about the goods available and about the other people in society. There can be no static solution for the individual, just as there is no static equilibrium for molecules in a fluid. But larger scale behavior can still be fairly slowly changing on average. Like salt in a saturated solution. Some molecules release from the crystals while other molecules attach… On average the salt concentration stays constant or changes slowly with the changing room temperature.

      Economics is like that except new goods and services are invented every day, new restaurants and yoga studios and automotive custom shops open all the time and older businesses like shoe repair or CRT TV repair close down… Dynamics is everything at the small scale, dynamics is also everything at the large scale, but the time scales are very different. The crust of the earth flows like a liquid on timescales of 100M years, but it’s hard rock when you fall down and skin your knee.

      Economics seems to miss this point entirely.

      Reply ↓

  4. “People do choose jobs to maximize their economic well-being and choose houses to maximize some intuitive basket of housing characteristics.”

    As a very small illustration of the difficulty of this area, I disagree with this statement – unless you allow the goals of “maximization” to vary *during the act that is supposed to maximize it*.

    I’m speaking as one who has had a number of jobs and owned a number of houses, as most readers of the blog probably are too. For a concrete example, when my partner and I bought our current house, we had certain criteria, several of which we thought were non-negotiable. The house we ended up buying did not meet two of them but had other features which it turned out we liked a great deal, like a lovely dry-stacked stone wall and woods behind the house among others. Discovering that we liked some features we hadn’t considered developed during the course of viewing a few dozen houses during our search.

    The way in which one comes to a decision about things like this is mostly not conscious, and the way one’s weights change is only partially conscious. Emotions play a large part. Even now, we regret giving up on one or two of our original “must have” features of our house, though we really like it overall.

    “Rational maximization”, whatever that is, does not play a large role in every day life of most of us. Yes, there is a balancing act, but the factors aren’t all that rational, and the desired balance shifts all the time.

    Reply ↓

    • I think there are two issues here. One, which you emphasize, is discovery. There is a big aspect of search in learning, followed by learning by doing. Economists have actually been on this case, at least in some contexts — not real estate AFAIK. The other is commensurability. Utility theory is predicated on the view that there is a common currency in which all attributes can be measured: your house is scalar-, not vector-valued. A few philosophers, like Martha Nussbaum, have challenged this, and rightfully so. That was also an aspect (but just one) of my critique of the VSL literature. (Also in the context of climate change.)

      As a practical matter, I think the assumption of commensurability has heuristic value in some situations; I wouldn’t chuck it altogether. But when it fails, it really fails.

      Reply ↓

    • Tom: it doesn’t matter to a “rational expectation” model that your emotions ruled your choice to have a forest in your backyard. The question is: how much did that irrational choice cost you vs. some other choice that may – or may not – have been rational? How much did it cost vs. the “truly rational” choice? What is the “truly rational” choice?

      I’m not anything close to an expert in this type of modeling but it seems obvious that when you or others make “irrational” or “emotional” choices about economic decisions, you do so within the confines of some rational bandwidth, and that the variation in that bandwidth is small enough to be irrelevant to a general model.

      Reply ↓

      • “it seems obvious that when you or others make “irrational” or “emotional” choices about economic decisions, you do so within the confines of some rational bandwidth”

        I don’t understand what you mean here. If people’s choices are based on imperfect information, and emotional or unconscious factors play a large role, then it seems to me that one can’t treat the outcome as based on “rational maximizing”. It’s possible that one is maximizing something, but what that something consists of can be very unclear.

        Reply ↓

        • ” and emotional or unconscious factors play a large role”

          You’re saying that, not me. The fact that you bought a different house based on an emotional whim doesn’t mean that house is significantly different in terms of economic optimization.

          I can drive to the airport on I-5 or I-405. By I-5 there is horrible downtown traffic and no diamond lane. By 405 there is a pay diamond lane half the way but the route is much longer and traffic is still heavy. Maybe I have a thing against pay lanes, so on that irrational basis I go I-5. Or maybe 10 miles of stop and go traffic drives me nuts and on that irrational basis I go I-405. Regardless of what choice I make, rational or irrational, whimsical or carefully planned, the time to the airport is hardly any different. Even if there if I discover there is a major accident on one route or the other, traffic shifts so quickly it doesn’t matter. Regardless of the reason, whether its rational or not, all my choices have the same level of optimization.

          That’s your house decision. It doesn’t matter because the alternative has an equal level of optimization.

  5. We propose that solving this computationally demanding problem, which we refer to as “joint utility improvement”, was one important selection pressure for encephalization in the human lineage (i.e., the tripling of brain size in the last 2-3 million years). We propose that, in small scale societies, those who are chosen as leaders tend to be those who are particularly good at solving this problem.

    We also propose that selection to solve this problem was particular intense in mothers, who had to continually make “good” decisions for their multiple cognitively immature offspring of different ages (ranging from infants to adolescents), and who therefore had very different needs. Moms, in other words, were typically leaders of their families, a cognitively demanding role.

    Preprint here: https://osf.io/9bcdk

    Reply ↓

    • I will admit that I only skimmed the paper and don’t intend to read it – probably due to being allergic to anthropologist’s writing (which, admittedly, is better than economists). I generally find this sort of biological or evolutionary determinism unappealing – no doubt such forces are real and perhaps quite strong – but humans seem to have the ability to act differently despite evolutionary forces that might be channeling them a different direction (such as the relative success of family leaders that have multiple children, or their relative brain sizes).

      But since the paper seems to discuss “leaders” somewhat more generally, it is hard for me to reconcile any of the claims about brain sizes or competency with what we observe in political leaders. And, I’m not sure why what is true of family leaders wouldn’t extend (at least somewhat) to leaders of a community or country. If anything, those leaders strike me as among the people least interested in the utility of others.

      Perhaps I completely misunderstand this work. If so, as Emily Litella would say, “never mind.”

      Reply ↓

      • Thanks for taking a skim. When I first started working in these small-scale societies, I also expected folks to be skeptical about the motives of their leaders. What I found instead seemed to be genuine respect for them. We’ve since done extensive surveys of the ethnographic record, as well as followup studies, and although, yes, there is plenty of evidence for exploitative leaders, there is equal if not more evidence for respected leaders. One key is that in small-scale societies, leadership is usually informal — leaders typically do not have any formal authority, and there might not even be a formal leader role. There is no obligation to do anything the “leader” says. Instead, their words carry “weight” — in community discussions, when this person talks, people listen. A second key is that these societies are frequently “fission-fusion”: people can switch groups pretty easily, and often do. Exploitative leaders soon find their “followers” voting with their feet and moving somewhere else.

        Here is one of our reviews of the ethnographic literature: https://anthro.vancouver.wsu.edu/documents/584/garfield_et_al_2019.pdf

        Reply ↓

        • I would have said the opposite – politicians seem to be more competent at rational reasoning than social reasoning. Of course, neither of those terms have clear definitions and I’m not aware of any research regarding brain sizes in relation to either of these ill-defined terms (there may be some, but I’m not aware of them). But, given Ed’s clarification above, it seems to me that whatever evolutionary forces tend to leaders having larger brains and/or more competency breaks down with scale. Families and small tribes may exhibit that phenomenon, but national leaders?

  6. Bentham could reply (and may have replied somewhere) that utilitarianism as a moral philosophy is for people who care somewhat about increasing other people’s utility, about behaving in a moral way, or about getting other people to behave in a moral way or in ways taht increase utility. That is, advancing these goals define part of their personal utilities, so that bad behavior has less utility than good behavior (for example), other things being equal. A few people do not have such goals or concerns and therefore should not bother to read Bentham, Mill, Sidgwick, Hare, Singer, et al.

    Reply ↓

  7. Let me start in nondefensive mode by agreeing with Daniel Lakeland that economics has traditionally given very short shrift to dynamics over equilibria. Even that was hard enough for a long time that mere existence proofs for equilibria were good enough to win Nobel Prizes.

    In addition, the pronounced preference for closed-form solutions led to a world (other than a few outliers like Herb Scarf) where the assumptions of what sort of preferences people were allowed to have were sharply constrained to allow characterizations of “the answer.”

    The next step of adding some dynamics were so called representative agent problems, in which what people thought was “solved” by assuming everyone thought the same thing.

    Rational expectations was the next step. people were allowed to think different things, but they knew what everyone else thought. This is, to say the least, counterfactual. The fact that the model is hard to solve even then is the problem Asher Meir is highlighting here.

    But now I’ll get a little bit defensive. A variant of the as-if argument can still work here, even as I grant that nothing as simple as Milton Firedman’s original billiard player analogy will work. His original argument is that a billiard player might know no math or physics, but asks as if he is solving extremely complicated physics equations to run the table. The logic still follows with a n-player billiard game; it’s just that there are two problems. The first one (which is an important critique of Friedman’s original thought) is that his analogy is fine for expert billiard players, but we need to worry about novice billiard players and billiard players who don’t understand the rules and billiard players who weren’t issued cue sticks. It is no defense of the actual interactions in the world to ignore all these people if they form a non-negligible percentage of the economy. And the second one is that the ability to get caught in dynamic stable suboptimal equilibria is very easy.

    Economists created game theory in which all this is much easier to see. Nash equilibria are what people talk about; sometimes they talk about the stability of those equilibria; often, they talk about cooperation and the changes that co-operation make towards those equilibria. Almost never is there a real exploration of the dynamics to get there. The problem is just too damn hard.

    Finally, I’ve always been hopeful that cellular automata models could provide answers to these questions, but I’m not sufficiently familiar with the actual results (as opposed to the concepts) to reliably comment.

    Reply ↓

    • To your last point, there is a growing literature in agent-based modeling, which, whatever its shortcomings, is an honest way to address the question of “how do people with limited knowledge and information processing ability actually get there?” Wasn’t it originally von Neumann’s hope that economics would evolve into ABM?

      That said, from what I’ve seen — admittedly not a whole lot — ABM has only a tenuous relationship to empiricism. I’ve seen a number of studies that begin by pointing to an empirical regularity or pattern and then tweak an ABM to generate it. (My diss chair, Herb Gintis, who died recently, was doing that back in the 80s when I was working with him, mostly for his own entertainment.)

      Reply ↓

      • Agent Based models are I think the best hope for useful alternatives. Heterogeneous agents each acting according to simple rules may not give us particularly accurate descriptions of real world individuals, but if we want to understand systems dynamics, such as stocks and flows of individuals through different states, it’s pretty much all we have. Humans are born into families with initial wealth, consume health enhancing and education goods, get jobs, move from place to place, earn income, invest and consume heterogeneously, pair off, do or do not have children, retire, age, die, and pass on inheritances…

        At any given time how many people who are at a given level of income, education, age, health, and in a given industry living in a given location in the country are taking what kinds of actions, and changing their state to a different kind of state (job, industry, location, parental status, whatever).

        If we infuse cash into the bank accounts of each household like the multi-trillion jump in M1 money supply in 2020, and we also infuse new remote work job opportunities, what changes in individual behaviors do we expect, among which populations, and how does that alter the flow of resources among all the related industries such as banking, manufacturing, computing, real estate, transportation, health care, etc?

        Very large scale dynamic agent based models at least can be compared to survey data from the consumer expenditure survey and the ACS and BLS surveys etc… And tuned until they produce something approximating observed results on aggregate. And then could be run in counterfactual scenarios and give estimates of causality for such interventions. The models constantly being compared to empirical observation has got to be better than an alternative where that’s not the case or only the case for a couple large aggregated stats.

        Reply ↓

  8. This thread is sort of up my alley, raising questions I’ve been mulling on for decades. There’s a lot to say, but I’ll keep it brief.

    Right from the beginning, when I first encountered Muth and his appropriation by macroeconomists, I doubted the rational expectation approach. It has this elegant recursivity, which is probably a large part of its appeal, but I was troubled by the same dilemma as raised in the OP. On my own, I worked out a ratex version of an intertemporal neo-Ricardian model, and it was simply a disaster. There was no possible algorithm that could get agents from here to there. (How could intertemporal profit rate equalization work, really? And why would you want it to?) Even the stripped down gen eq models I learned at the time (Arrow et al.) were understood as demonstrations of how implausible such an equilibrium was — e.g. all those contingent markets. I have honestly never understood how they could be seen in any other way — and that’s before you endogenize expectations, which ups the dimensionality even more.

    Meanwhile, I sort of backed into an appreciation for the dependence of mono-equilibrium on the assumption of no or limited interaction effects between agents and choices. This has to do with the routine invocation of quasi-convexity in modeling, which has immense social theoretic significance. A few name economists (e.g. Durlauf and Brock) have also written on this, but it’s difficult to go beyond a general discussion. Without getting too into the weeds here, my own view is that it’s easier to make progress on micro applications of nonconvexity analysis, for instance the theory of the firm.

    I hope I haven’t chased away all the noneconomists; this is really an area where economics encounters social science more broadly.

    Reply ↓

  9. Try to put it into an equation to maximize! Our consumption is an argument to our utility function, OK. Other people’s consumption is an argument, nu. Now make other people’s UTILITY FUNCTIONS an argument. Sound easy? Now make the dependency of B’s utility function on A’s utility functions an argument in A’s utility function.

    Reminds me of basic newtonian solar system simulations. Each object has a mass, along with (x, y, z) position and velocity. During each simulation step you loop over the objects, calculating the distance between each pair and resulting force. Then the sum of these forces is used to update each object’s position/velocity for the next step.

    That is simple enough and gets very close to the correct orbits. But does anyone think the universe actually works this way? Ie, once per instant, every particle calculates its distance to all other particles then sums the influences. If so, where/how are these calculations being performed?

    I think not, just like each person is not really calculating the expected utility of all the others.

    Reply ↓

  10. It’s not much of a stretch. There are a ton of smart people in finance, and it’s reasonable to assume they can predict as well as academic econ researchers– in fact, I’m sure they’re even better a lot of the time. So it’s plausible that many of the most relevant actors in the economy are basically consistent with the model.

    And of course, the whole point of solving the model is to compare the results with empirical data, so if an assumption is way off base in an important way, economists can figure that out.

    And that last point, I’m not at all convinced that North Koreans would be better off not knowing about South Korea. That seems very contrarian to me.

    Reply ↓

    • I do like Herbert Simon though, and I think there’s a lot of economics research remaining to be done on what happens when people diverge from rationality. And more importantly, work on *when* we should expect people to diverge from rationality.

      Reply ↓

    • The point of the Meir quotes above is that even *if* real-world economic actors could “predict as well as academic econ researchers”, they would not be able to predict as well as the fictitious agents in rational expectations models. Academic econ researchers find it hard to calculate equilibrium within complicated models, ascertain the true parameters of a given model, and choose between models. The agents in rational expectations models face no such difficulties: they know the objectively correct joint distribution of all variables relevant to their decisions.

      Reply ↓

      • The idea that the details about joint distributions etc. is important to my response is kind of humorous.

        Have you done simulations showing that any of that makes an important difference? Or are you just offended at the idea of simplifying assumptions

        Reply ↓

        • Let’s try an example. If you had asked academic economists for their inflation forecasts over the last few years, I expect that each individual economist would have professed considerable uncertainty about their point forecast and even what kind of statistical model is appropriate for forecasting inflation (how important is the random walk component, which measure of slack should be included in a Phillips curve-type model, whether there are nonlinearities, etc.). In the cross-section of economists there would also be substantial disagreement. “Smart people in finance” presumably have a similar level of uncertainty and disagreement. In a rational expectations model, by contrast, households and firms are 100% certain what the true statistical process for inflation is, given a monetary policy rule etc. Suppose that in the model, inflation over 2021-22 was caused by transitory shocks. Then agents in the model are 100% certain that inflation was caused by transitory shocks and is going to mean-revert in 2023. In reality, whether or not inflation was *actually* caused by transitory shocks, for sure households, firms, economists and smart people in finance are not all *certain* that it was caused by transitory shocks. So the mapping between shocks and people’s inflation expectations must be different in the REE model vs reality, even if everyone forecasts inflation like an economist.

  11. I have never believed that people, myself included, behave all that rationally, so I have never taken economics seriously. For those who do, let me ask: how do economists explain the failure of people to behave in economically rational ways even when there is pretty good information at hand? Flood risk is a good example. Geographers have long known that people don’t take proper account of the risk that (say) a home will flood, even when there is a long enough flow record for the relevant river that the risk can be estimated pretty well.

    Reply ↓

    • There are countless economic theories that can account for this type of behavior, but few have the tractability of RE. To answer your question with a question, how would *you* explain that behavior, and what are the implications of that explanation in terms of, say, interest rate policy?

      Reply ↓

  12. I largely agree with Meir that the assumptions are problematic, and I partially agree with Dale’s characterization of macroeconomics. This problem with assumptions was a source of intermittent, Nausea-like, periods for me during my Economics education. Particularly during graduate school, where I sometimes felt like I had wasted years of my life studying Alchemy. At the same time, whenever I read a critique of Economics along this line, I end up thinking “well, what else are we going to do”? It doesn’t seem to be enough to just say “build in these assumptions as well” or “take out this assumption” without actually showing that whatever alternative you’re proposing is more useful than the current standard. It also isn’t enough to dismiss the current standard based on the faulty assumptions alone. I agree that mainstream Economists are too comfortable drawing conclusions about the real world based on simplified models with unrealistic assumptions, but I’ve yet to come across something that is provably more useful.

    Reply ↓

    • My grad macro prof was fond of saying that rational expectations is the worst assumption, except for all the other ones.

      There are a large number of well known alternatives to RE, such as adaptive expectations or rational inattention, and countless others that have been proposed in the literature in recent decades. The reason they are not more widely used is practical, not philosophical.

      Reply ↓

  13. The traditional answer to these objections is “Yes, many of the assumptions are unreasonable, but they give a baseline against which we can judge actual behavior. If we want to do a better job of capturing actual outcomes, we can always try to weaken the assumptions so as to capture a more grounded environment, but it’s going to be really hard and take a long time.”

    It isn’t just talk, either. You can find highly-cited models with incomplete markets. You can find highly-cited models with heterogeneous agents. You can find highly-cited models with myopic agents. You can find highly-cited models with irrational expectations (that can even nest rational expectations in the limit) and erroneous beliefs. Even back in the 80s the idea of an “expectationally-stable” criterion was proposed and became rather popular, which requires that an equilibrium *can* be learned by agents who make guesses and update their guesses over time.

    I’m not even saying all of these developments are great or will hold up over time, but I do find it a bit odd that all of these developments are being ignored.

    Reply ↓

    • If rational expectations models give “a baseline against which we can judge actual behavior”, then so does any model – that is no argument for selecting this particular set of models as the baseline. Nor is it plausible that economists write rational expectations models because doing anything else is really hard and takes a long time. In fact the opposite is true. As the Meir quotes above point out, solving for a rational expectations equilibrium in moderately complicated models is technically challenging – much harder than solving a model where agents have backward-looking expectations or follow decision rules which aren’t derived from an optimization problem, or simulating an agent-based model.

      Reply ↓

      • No, not any model is not a baseline. A model with a bunch of ad hoc assumptions about people acting irrationally is not baseline of anything. Rational actor models are central to economics and it’s very useful to see where and to what extent they match up with reality. That’s what makes it a “baseline.”

        Reply ↓

        • It’s true that in macroeconomics *as it exists now*, rational expectations equilibrium is used as a baseline model: REE is the default assumption, researchers prefer to write REE models if possible, if you want to depart from REE you generally need to provide a good reason. However, macroeconomics could be structured differently with a different set of models (ABMs, optimizing models where agents form expectations using econometrics, etc.) as the baseline. What is the argument for selecting REE as the baseline rather than some alternative? What is “ad hoc” about trying to learn how people form beliefs in reality and building models to approximate that, rather than assuming REE?

        • “What is “ad hoc” about trying to learn how people form beliefs in reality and building models to approximate that, rather than assuming REE?”

          The issue is that “people” are not a fixed target. Publicizing effectually any theory based on how people form beliefs “in reality” will inevitably change how people form beliefs, and thus automatically invalidate the theory. The benefit of REE as a baseline is that it’s in some sense “sticky”, that merely learning that people behave by REE will not cause agents to depart from rational behaviour. This is not true of most other formulations.

        • @Zhou Fang: Yes, how people form beliefs changes over time, sometimes in response to economic theory. So there will be no universal laws describing expectation formation – the best we can do is discover how people formed beliefs in particular periods, and engage in informed speculation about how belief formation might change in particular contingencies. Any particular non-REE model of belief formation will not be accurate everywhere and should be interpreted with care – but can still be used as a baseline.

          It’s also true of course that REE models have the special property that, if you convince an agent in the model that the model is true, that wouldn’t change their expectations or behavior. I do not see why this is an important property for a baseline model to have, especially since it comes at a cost in terms of realism.

        • I dunno, that seems like a highly important and desirable property in a baseline model, far more important than it being momentarily more realistic.

          In physics we start with stationary frictionless spheres in Newtonian motion. In statistics we work with the normal distribution. It’s similar here.

  14. When I was in graduate school at UCLA in the 1970s, I used to attend a seminar series that featured economists and theoretical ecologists. There was a lot of similarity in their modeling, but it struck me at the time that theoretical ecologists were constrained by field ecologists in a way that did not seem to apply in academic economics.

    Reply ↓

  15. Interesting post. I’m not sure I’d equate the difficulty of calculating and characterizing multiple equilibria with what it means to best respond as an agent. But agree there should be more focus on the how. The normative/positive tension and fact that it’s difficult to study beliefs on beliefs over time outside of toy problems probably play some role.

    Reply ↓

  17. I would like to pose the following key question to participants in this intriguing forum, especially to Daniel Lakeland, Dale Lehman, and Peter Dorman:

    **What is the “right” mathematics for the science-with-practice study of real-world systems?**

    As a multidisciplinary researcher (history BA, economics PhD with a math minor, courtesy research professor of electrical and computer engineering) — increasingly immersed since 2001 in the study of many-to-one economic and physical measurement issues arising for the design and operation of grid-supported centrally-managed U.S. electric power markets — I would like to offer the following “resolution” for debate:

    *Resolved*: Completely Agent-Based Modeling (c-ABM) provides the “right mathematics” for the science-with-practice study of real-world systems.

    Evidence and arguments in support of this “resolution” are provided at the following “ACE Site”:

    ===============================
    Agent-Based Computational Economics (ACE): A Completely Agent-Based Modeling (c-ABM) Approach
    https://www2.econ.iastate.edu/tesfatsi/ace.htm

    ACE Site Overview:

    Scientists and engineers seek to understand how real-world systems work and could work better. Any modeling method devised for such purposes must simplify reality. Ideally, however, the modeling method should be flexible as well as logically rigorous; it should permit model simplifications to be appropriately tailored for the specific purpose at hand.

    Flexibility and logical rigor have been the two key goals motivating the development of Agent-based Computational Economics (ACE), a specialization to economics of a completely agent-based modeling (c-ABM) method characterized by seven specific modeling principles (MP1)–(MP7).

    The seven modeling principles (MP1)–(MP7) characterize a completely agent-based model as a computational laboratory permitting the exploration of a computationally-constructed world. This exploration process is analogous to biological experimentation with cultures in Petri dishes. The modeler configures and sets initial conditions for the world. The modeler then steps back, assuming the role of pure observer, as subsequent world events are driven solely by the interactions of the world’s constituent entities.

    The ACE site provides an overview of c-ABM and ACE, a brief history of their development, their role within a broader spectrum of modeling methods, and annotated pointers to extensive ACE/c-ABM research and teaching resources.

    =============================

    Reply ↓

    • Here is only a partial answer and one that reflects my very rudimentary understanding of agent-based modeling. What you are suggesting sounds a lot like systems modeling, a number of tools that I am aware of though I don’t use myself. I actually (and hopefully this won’t provoke too much negative reaction from others) use spreadsheet based simulation modeling in my teaching quite a lot. It depends a lot on the potential audience. If you are talking about PhD students, then I don’t have any recent experience with them, and your agent-based modeling might be appropriate. For my students – working professional business students – I think I prefer the spreadsheet approach due to more familiarity with the tools. For statistical analysis I use different tools, of course. But, regardless of the tools, the necessary education in my view is to understand how to decompose a practical problem into a series of logical steps – with attendant relationships – and then construct quantitative models of these steps that permit you to explore predictions, sensitivities, and limitations. I don’t think there is a unique tool that is “best” for achieving this. Rather, I think the tool is dependent on audience, researcher, and purpose of the analysis. I doubt that there is a standardized approach that would get universal support.

      If others respond to your question, I suspect we are not all in agreement. We can’t even seem to agree with it is essential that models be dynamic or not.

      Reply ↓

    • Interesting that you are at Iowa State. Were you there back in the early 90’s when I was?

      As for ABMs, yes I think these are *one* major appropriate tool for real world systems. I wouldn’t say it’s the only thing people should do, they tend to be computationally expensive… but I’d argue that in many cases the only way to get really meaningful results that aren’t over-simplified is through ABMs. Electricity markets are a good example where I think ABMs would help. Another might be something like water resources in CA, or the design of cooperative dynamic methods of dividing the electromagnetic spectrum or understanding how healthcare policy works to alter behavior and health outcomes.

      I have discussed in the past starting a nonprofit to build and investigate real world social dynamic ABM based models to inform various types of policies. I actually wrote a proposed mission statement at the beginning of this year. Anyone who knows of foundations interested in funding that sort of thing should let me know!

      Reply ↓

  18. To: Daniel Lakeland
    RE: Response/query on 31 Oct 2023

    Yes, I have been at Iowa State University (ISU) since 1990. I was a courtesy math professor at ISU from 1990 to 2017 — hence, while you were getting your BS in Math 1993-1997. Did you know/meet my friend and collaborator Dan Ashlock, a Professor in the ISU Math Department during your time at ISU? I’m tempted to say “how could you not” — Dan was a ebullient “force of nature.” For many years, Dan and I (plus others) ran a “Complex Adaptive Systems Seminar” at ISU.

    Another connection is that I was an economics professor at the University of Southern California from 1975–1990 and my main collaborator was Bob Kalaba, a great scholar and polymath whose home base was in the mathematics department but who also had courtesy appointments in biomedical engineering, electrical engineering and economics. Unfortunately, Bob died in 2004 — before your time at USC (PhD Civil Engineering, 2008–2013).

    I admit my “Debate Resolution” submitted to this forum — i.e., **Completely Agent-Based Modeling (c-ABM) is the “right” mathematics for the science-with-practice study of real-world systems** — is stated in overly strong terms. I did this deliberately to see if I could start an interesting discussion!

    However, I have increasingly come to believe that serious economic and physical measurement issues are negatively affecting the operations of current grid-supported U.S. electric power markets, issues that arise from: (i) a reliance on “measurements” expressed by means of classical mathematical representations (e.g., system states expressed as real-valued vectors); and (ii) manipulations of these measurements by means of classical mathematical operations that do not cohere well with complex underlying realities.

    As stressed in the following talk, I believe these that same issues (i) and (ii) arise more generally for the science-with-practice study of a variety of real-world systems consisting of intricately intertwined social and physical processes; and that c-ABM appears to offer many advantages for the study of such systems:

    “Agent-Based Modeling: The Right Mathematics for Social Science?”
    https://www2.econ.iastate.edu/tesfatsi/KeynoteAddress.LTesfatsion.SSC2021.22Sept2021.pdf

    Mainstream economists have rather recently discovered and accepted the economic importance of top-down (task-oriented) AI and generative AI (e.g., ChatGPT), which is great. However, they have yet to evince any interest at all in embodied AI methods, such as c-ABM. However, I am a long-term investor :)

    Reply ↓

    • Yes, I knew Dan reasonably well, took his course in Artificial Life optimization methods, and had him on my undergrad thesis committee. I looked him up about a year ago and discovered he had died recently, it was a bit of a shock, I think he was only about 60 ish?

      ABM is a great way to model systems with interacting components which have emergent properties caused by the complexity of those interactions. For example molecular dynamics can tell us lots of stuff about materials properties. That’s an ABM. My biggest concern is that people don’t WANT the truth. ABMs have the power to show that social systems as constructed today don’t serve the purposes that they ought to serve and instead serve the purposes that those who are in charge of them prefer, usually wealth and security for a few at the expense of others. Consider the Texas power grid failures a few years back. That wasn’t an accident. Prices going through the roof and producers charging families $10k for a weekend of electricity was the system *working as designed*

      I believe that there is rampant manipulation in our economy. Super wealthy people are super wealthy because they have friends who tell them secret information or manipulate the way things work to their friends benefit. Consider the outsized returns of Congress people and their families, and how many of the people in congress briefed early on COVID sold stocks immediately after (late 2019)

      The irregularities you see and bemoan in the energy market are there because they make some powerful people rich in all likelihood. ABM could improve efficiency, reliability, pollution, etc but would eliminate the special loopholes by which enormous quantities of cash are shoveled into a few peoples pockets… So there’s that.

      Basically I agree with you, but being right doesn’t always matter, ask Aristarchus of Samos…

      Reply ↓

Leave a Reply

Your email address will not be published. Required fields are marked *