Friday, January 15, 2016

Complexity without purpose: evolution as a framework for economics

Mobile phone evolution (source)


While I don't have access to the numbers, this article at Evonomics by David Sloan Wilson appears to be doing well by an estimate of the number of times it has appeared in my Twitter feed. It's an interesting read; check it out. I especially liked this dig:
This places [economist Dani] Rodrik’s understanding of evolution just a shade above creationists.
I also completely agree with this:
For an evolutionist such as myself, the idea that big theory is dead or that big data can be analyzed without a big theory is wacky. ... Einstein observed that “It is the theory that decides what we can observe”. The world is so complex that we cannot possibly attend to everything, so a theory—broadly defined as a way of interpreting the world around us—is required to tell us what to pay attention to and what to ignore. Without a theory, we are literally blind.
I talk about this in the context of machine learning -- a particular way data scientists go about looking at "big data". A lack of an explicit big theory means you're using implicit theories.



However, this really needs some data to back it up:
I got excited when Richard Thaler and Cass Sunstein called for an economics based on Homo sapiens ... Then I searched my kindle edition for the word “evolution” and came up empty. How can economics be based on Homo sapiens without any discussion whatsoever of Homo sapiens as a product of genetic and cultural evolution?
How does Prof. Wilson know the genetic and cultural evolution of H. sapiens is a necessary ingredient (such that its inclusion excites him)? What specific stylized economic facts that are presently not understood by economists, but well enough understood by biologists to warrant a more complex evolutionary agent based model above and beyond the normal agent based models?


What, really, are any stylized facts of economics at all?

Equilibrium?

Economists actually have very concrete definitions of equilibrium. However, they don't have any handle on how one achieves equilibrium or travels from one equilibrium to another (see here and links therein). One could easily see evolution as guide to understanding the process for finding equilibria, but the word "equilibrium" doesn't appear in Prof. Wilson's article or the cited paper. 


This is the stylized fact of economics the information transfer model focuses on, attempting to answer the question: what values of macroeconomic aggregates should we expect to see?

Money?

Economists haven't really figured out what money actually is or what it does. I've proposed a cultural evolutionary mechanism (which is admittedly highly speculative), but it stands on an economic model of what money is and how money works. The economic content doesn't really depend on the specific evolutionary pathway, so evolution isn't really adding to economic understanding, but rather historical understanding.

Does Prof. Wilson think evolution will help solve the question of why money has value? That would be interesting!

Business cycles?

Economists haven't figured out what the business cycle is or does, or even if there is a "cycle" at all. There is no universal definition of "recession" or what is happening during one. Adding evolution isn't going to help understand things like the business cycle until we have an operational definition of what a business cycle is. Which macro aggregates are relevant? What distinguishes recession from non-recession? Compounding the problem, answering these questions requires a theory to support identifying the features of the data itself.

For example, suppose you propose a model where evolution tells us why humans are prone to panic when resources (e.g. stocks) become threatened, leading to an empirically accurate (after fitting a few parameters) description of financial crises. You propose this as an explanation of recessions. But are recessions caused by financial crises? We don't know the answer to that: some economists think so, but others think financial crises are a symptom of recessions. In the latter case, macroeconomics (e.g. central bank targets) is triggering humans' evolutionary behavior. Macro causes behavior.

Behavioral economics?

Richard Thaler says
The field of behavioral economics has been around for more than three decades, but the application of its findings to societal problems has only recently been catching on.
There would be two possible reasons for behavioral economics to take three decades (!) to make some headway: 1) there is a conspiracy blocking the research from getting into the mainstream, or 2) it doesn't work demonstrably better than run-of-the-mill utility maximization by H. economicus. Basically, it turns out the reason is mostly #2 for macroeconomics.


And since macro is the issue Prof. Wilson is thinking about (e.g. the reference to the 2008 financial crisis at the beginning of his article), if the H. sapiens behavioral econ Thaler is talking about doesn't add much understanding beyond the H. economicus understanding, why would even more complex behavioral econ based on evolution add more? Another way to say it is that econ has tried to add a bit of behavior, but econ data is uninformative given such complex models -- and therefore we can't tell the difference. It may be more theoretically satisfying, but if a behavioral model doesn't improve empirical results, all it's doing is raising your AIC.

Economic data?

Expanding on the previous part to make the point more forcefully: macro data is uninformative. Empirical data can't reject one theory in favor of another -- and that's when you have simple H. economicus models! Adding evolution to the theory would only make it more uninformative.

As George E. P. Box said (what he really meant by "all models are wrong" -- see here):
Since all models are wrong the scientist cannot obtain a "correct" one by excessive elaboration. On the contrary following William of Occam he should seek an economical description of natural phenomena. Just as the ability to devise simple but evocative models is the signature of the great scientist so overelaboration and overparameterization is often the mark of mediocrity.
Supply and demand?

I am planning on giving a talk this summer about this one. If the cultural and genetic evolution of H. sapiens matters, then 1) why do we get the same results as the utility maximizing framework with random behavior and 2) why do E. coli and C. capucinus exhibit the same results.

My view is that #1 and #2 together tell us humans don't really matter that much. In fact, that's also what the most successful empirical economic result of all time tells us (kidding, but Noah Smith seems to think so). It's based on random utility and has some overlap with the partition function approach this blog is into.

Evolution would answer #2 with either a) utility maximizing behavior was a property of the last common ancestor of E. coli and H. sapiens, or b) it is the result of convergent evolution. 

And both of these could be true in some general sense. The requirement for entropy maximization to be a good model is simply that agents fully explore their state space ... something that biological organisms have been doing ever since they evolved. And convergent evolution is sometimes a result of physics or other mathematical properties. So maybe that is what is happening. 


...

The thing is that in order to know if evolution is necessary to economics, we need to know what questions we are trying to resolve with evolution and how much of (or even if) economics can be understood without it. Which theoretical economic problems does evolution address?


I am of the opinion that humans' sense of their own agency has gotten in the way of understanding economics. I wrote a piece about biologists saying we needed really complex agents to understand economics based on no reason whatsoever other than gut feelings that economics is a "complex" system. It has to be complex, right? I mean, I'm a complex human with all kinds of thoughts and feelings about inflation and government debt, so that just has to matter ... right?

Maybe it does, but the information equilibrium model does pretty well with its predictions and it's based on humans as mindless atoms -- they're not even utility maximizing drones. We have H. atomicus, an even simpler being than H. economicus. Information equilibrium also addresses most of the questions above (it relegates the business cycle to a question of scope).

...

At the end of the article, Prof. Wilson cites a paper he wrote with economists John M. Gowdy and J. Barkley Rosser proposing evolution as a theoretical framework for economics. I love the idea of a framework for economics! It's one of my biggest complaints about the field and my own paper on information equilibrium proposes to be its own theoretical framework for economics.

In the abstract, they present a questionable argument for using evolution as a framework:
Then we consider four reasons why evolution might not need to be consulted for human-related subjects such as economics and public policy. We conclude that these reasons can be valid in particular cases, but they fail for any sizeable human-related subject area. Hence evolution can and should become a general theoretical framework for economics and public policy.
This is kind of a logic fail. Let's set up some variables:

e = evolution is a good framework for economics
p = reasons for evolution not being a good framework

The argument is then

p → ¬ e
¬ p
∴ e
(invalid)

This is incorrect application of modus tollens/modus ponens. The valid arguments are

p → ¬ e
e
∴ ¬ p


p → ¬ e
p
∴ ¬ e

The only way the argument works is if p ⇔ ¬ e, which is only true if p represents every possible reason for evolution not being a good theoretical framework for economics. So what is the list of reasons? My paraphrasing:
Reason #1: Economists are smart and have been studying economics for a long time, therefore an evolution framework will come to the same conclusions. 
Reason #2: If markets are efficient from a theoretical standpoint, then the process matter by which they reached efficiency does not matter. 
Reason #3: The process by which something reached its extant state does not matter to the study of that extant state [Ed. note: I'm not entirely sure how this is different from reason #2.
Reason #4: Not all branches of knowledge are necessarily relevant other branches.
It seems fairly obvious this does not exhaust the list of reasons why evolution would not be a good theoretical framework for economics or not need to be consulted. I have one for macroeconomics that I mentioned above:
Reason #5: Macroeconomic data is already uninformative for simple models using H. economicus, and empirical data would only make it harder to reject competing theories using H. sapiens.
Macroeconomics is already too complex to sort out its own theories! What's needed is something radically simpler (or something with much broader scope). There are probably loads of others, including one big one that I got from the Wikipedia page on unsolved problems in biology:

At present there is no theoretical model for how adaptation occurs that is close to being complete. 

This appears to be the same problem economics has with equilibrium (listed above). We know one when we see one, but we don't know how new ones arise or how we get there. How does evolution propose to be a theoretical framework for economics when it's not even a complete theoretical framework for itself?

In any case, we need convincing reasons to use fully evolved H. sapiens in economics, not arguments against reasons not to use H. sapiens.

Actually, with the information transfer model (1) we do come to many of the same conclusions as economics, (2 and 3) if maximum entropy is a good approximation, then it doesn't matter how it came to that state (quite literally: maximum entropy means losing all possible information about the initial state), and (5) since it represents a much simpler theory, macro data becomes informative and can reject models and mechanisms.

22 comments:

  1. You cover a lot of ground here. Just a couple of comments now, before retiring for the night. :)

    Jason: "But are recessions caused by financial crises? We don't know the answer to that"

    Well, back in 2008 a 15 minute web search led me to some history of recessions and recoveries. The search was prompted by Sec. Paulson's defense of the bank bailout "so they can start lending again." (Sorry, I can't find the site again. Web searches ain't what they used to be.) Anyway, recessions and financial crises do not always go together. And recovery after a combination financial crisis/recession tends to take a long time. The banks were not going to resume lending any time soon.

    Jason: "There would be two possible reasons for behavioral economics to take three decades (!) to make some headway: 1) there is a conspiracy blocking the research from getting into the mainstream, or 2) it doesn't work demonstrably better than run-of-the-mill utility maximization by H. economicus. Basically, it turns out the reason is mostly #2 for macroeconomics."

    Well, in his latest post, http://mainlymacro.blogspot.co.uk/2016/01/heterodox-economists-and-mainstream.html , it seems to me that Wren-Lewis indicates the conspiracy pretty clearly. It lies in the requirement to provide micro-foundations for macroeconomic hypotheses. And the generally acceptable micro-foundations in current economics are implausible from the viewpoint of psychology, sociology, anthropology, and evolution. It is not accident that the great majority of so-called behavioral economists are actually psychologists and cognitive scientists. There is a turf battle going on, n'est ce pas?

    As for your second point, I am not sure that that is right, but let's say that it is. First, how much does it matter, given that macroeconomics is only weakly empirical? Second, is it not also the case that behavioral micro-foundations do no worse than the assumption of homo economici? Recognition of that should bolster the viewpoint of information transfer economics, as it is more parsimonious than either of the other two.

    ReplyDelete
    Replies
    1. Hi Bill,

      Regarding the first quote, yes recessions coupled to financial crises do seem to take longer to recover from as one of the few physical books I've read in awhile (Reinhart and Rogoff "This time is different") says.

      The ITM seems to say these kinds of things happen when monetary policy is ineffective (or the monetary authority is incompetent, as in the early days of central banks).

      Regarding the second quote, the idea of adding H. sapiens is microfoundations. It is true the typical (analytically tractable) microfoundations used in econ aren't very realistic. But the issue is that complex microfoundations and simple microfoundations both suck, empirically speaking.

      There is also the issue of dimensional reduction, or as I put it, emergent H. economicus. H. sapiens, C. capucinus or E. coli go in the model and H. economicus comes out.

      http://informationtransfereconomics.blogspot.com/2015/09/the-emergent-representative-agent-1.html

      This actually helps make sense of why simple and complex microfoundations both suck. If you put both simple and complex microfoundations in, you get (effective) simple agents out. No matter how many degrees of freedom you add to your micro agents, it's not going to be a better description of the macro aggregates because you've framed the problem incorrectly.

      That incorrect frame -- at least in my opinion -- is thinking macro is about the agents, when it's really about the state space.

      Delete
    2. "No matter how many degrees of freedom you add to your micro agents, it's not going to be a better description of the macro aggregates because you've framed the problem incorrectly."

      Amen, brother. Now to get Wren-Lewis on board. ;)

      Delete
    3. Actually, let me change my reply. Adding degrees of freedom to micro agents makes their aggregate behavior more like randomness, and hence, more like Information Transfer Economics. :)

      Delete
  2. Jason, nice post. On a slightly different subject, I've sometimes heard about "agent based models" where there are vast numbers of agents and each agent follows a simple set of rules. I think I've seen you make similar criticisms of this approach (too many parameters, etc.).

    I don't know if I've characterized agent based models correctly. However, supposing I have, I wonder if it would be possible to use the ITM approach to predict aggregate behavior of an agent based model (perhaps given a few measurements to find ITM parameters with). That would be quite a coup, wouldn't it? A kind of secondary validation for pursuing the ITM approach?

    ReplyDelete
    Replies
    1. See my reply to Bill above for a bit more on this.

      Regarding your challenge, that is what happens with atoms in the ideal gas. The ITM approach gets to the ideal gas law just like a simulation with atoms in a container would. I see no reason why the same wouldn't happen for some agent based model.

      Delete
    2. Your gas simulation vs ITM example: Yes! I wrote almost the same thing before erasing it (to make my comment shorter).

      But what I'm thinking here is a maybe a collaborative effort with an agent based modeler: or at least someone who's agent-based curious... someone who knows a thing or two about them, but who wouldn't be devastated if the results of the collaboration undercut their justification. Or perhaps all you need is access to a few publicly available and well known (?) agent based models. A diversion from you main interest I suppose, but if I were investing time in agent based modeling, I'd be VERY interested in such a comparison.

      Delete
    3. ... and who knows: you might find there's a surprise or two as well.

      Delete
    4. Here is a simple rule for human agents that has perceptible macroeconomic effects. Give lots of gifts twice a year at certain times. Macro results: Increased economic activity at those times. Increased prices at restaurants and bars at those times. Salary bonuses at those times. (The rule is to give lots of gifts.) It's called chugen in Japanese. When lived in Japan I had to stop eating out, except for lunch specials, during chugen. ;)

      Delete
    5. Oh, yes. It seems to me that chugen is not easy to model based upon homo economicus, who is not inclined to give gifts without an expected greater return. OC, bribery in the guise of chugen is part of what is happening. As a friend of mine explained, if the transportation minister does not give the prime minister a good enough gift, he is out of office. ;) Still, it is only part of what chugen is about. I guess if Wren-Lewis is your advisor, you should not write your thesis about chugen.

      Delete
    6. Hi, Tom,

      Chugen is Christmas is spades. I taught English to three classes of engineers. They told me that their chugen bonuses amounted to half their total salary. (Not that they spent it all on chugen, OC.)

      Delete
    7. Wow... Sounds like an event designed to spur economic growth.

      Delete
    8. Chugen certainly has economic effects, and combats the tendency to save, which is also quite strong there. It is, I think, part of reciprocally reinforcing webs of responsibility, called giri. When I was there, if you belonged to (note, not worked for, even though that's what you did) a sufficiently large company, you had a job for life, part of their giri to you. At the same time, you owed intense loyalty to the company. Hardly homo economicus behavior. ;) My friend who told me about the transportation minister and the prime minister hated chugen. He was in a position to receive more gifts than he gave, and it meant that he had the burden of doing favors for the people who gave him gifts for the rest of the year.

      Delete
  3. I think you (and Wilson) should look closer to home. Biology accepts physics as a constraint. In particular, the flow of energy is crucial to understanding biological systems at all levels. The flow of energy is also a binding constraint on economies. Notably, our modern world is defined by the exploitation of highly concentrated energy in the form of fossil fuels. Look there for your grounding.

    ReplyDelete
    Replies
    1. Hi Sam,

      While it is my own decidedly heterodox view, I think there is a misconception of "the economy" as an engine running on fossil fuels when it's really more of an algorithm where the energy constraints are far less relevant. Energy is just one input. If we run out of fuel, economics will still work the same (price of fuels will go up, and likely become the major cost of living as opposed to housing for most people today with our current most scarce/finite resource of land).

      I wrote a bit more about it here (part II, and links therein):

      http://informationtransfereconomics.blogspot.com/2014/10/entropy-in-three-pieces.html

      There will be huge issues when we run out of fossil fuels, but I think they will be more political than economic per se. Politics can re-direct our allocation of goods and services away from the market allocation, or reduce our needs for them. I'm not saying I'm optimistic this will happen (with or without disruption) -- I'm actually pretty pessimistic on this point. But the correct theory of economics will describe how the allocation problem will be solved if it is solved using the algorithm of "money" regardless of how much oil is left. This allocation won't be politically acceptable to most of us -- and the solution would probably be a non-economic allocation through regulation or rationing ... or in the most optimistic scenario, technology changes.

      Delete
    2. There is a constraint with nuclear energy though... Once the sun burns out, it'll be time to pack it in (if not a few billion years prior). ;D

      Delete
    3. We probably will if only because money has value today -- which implies a finite probability of humans existing at the end of time ... right? :)

      Delete
    4. Ha!... BTW, Simon Wren-Lewis and Lars Syll are squabbling about heterodox economics and Syll complains a lot about "mathematical-deductive analytical formalism" and other permuted subsets of those four words + "modeling." I don't know what that means.

      Delete
  4. O/T: Jason, maybe you can develop a model of the frequency of MMist CB criticism vs time.

    ReplyDelete
    Replies
    1. Pretty easy.

      Whenever inflation heads toward a de facto or de jure target, the central bank controls inflation. Whenever it doesn't, the central bank is incompetent at controlling inflation.

      Delete
    2. Yes, that's pretty easy. Can it be expanded to predict the ratio of the two kinds of posts regarding the Fed or the BoC say, in 2020?

      Delete

Comments are welcome. Please see the Moderation and comment policy.

Also, try to avoid the use of dollar signs as they interfere with my setup of mathjax. I left it set up that way because I think this is funny for an economics blog. You can use € or £ instead.