Sunday, August 27, 2017

A random physicist takes on economics

Jason Smith, a random physicist, has a new book out where he takes aim at some of the core foundations of microeconomics. I encourage every economist out there to open their mind, read it, and genuinely consider the implications of this new approach.

Go get it now. It only costs a few bucks.

So what do I think? His approach is exactly what economics needs - a set of fresh eyes on the basics.[1]

The book is, fundamentally, an introduction to Smith's new view of what I would call ‘microeconomics as the emergent characteristic of random agents in constrained situations’. Or, more simply put, why you don’t need rational decision-makers for a useful economic theory that makes good predictions.

To get some sense of why this is important, economists are often criticised for getting the big picture stuff of macroeconomics wrong, like missing the financial crisis. But in reality, the economic micro-level stuff, about responding to relative prices, making choices based on incomes and preferences, is also a failure built on an elaborate, but highly questionable, theoretical structure.

Smith says that this theoretical structure is unnecessary. In fact, he says, people acting randomly within their budget will have the emergent property of behaving ‘as if’ they comply with the rational economic model. That is, humans are irrational at the individual level, but the fact that our choices are constrained by our incomes means that in aggregate, the average behaviour responds to external changes in a similar manner to that of the mythical rational person.

To get a feeling for this, consider the graph below from one of Smith's favourite economics paper by Gary Becker, which made similar points. For those who haven't guessed what it shows, the X and Y axes are the amount of consumption of two goods, and the lines C-D and A-B are the budget constraints (i.e. how much of each good, or combination of goods, can be bought) at two different sets of relative prices. The line C-D has a lower price of good Y, and a higher price of good X, than the line A-B.


The point C is the centre of the triangle A-B-0. So if people consume randomly within that budget constraint at those prices, the average person will consume a combination of goods near point C. When the budget constraint changes to C-D, the new centre point is C'. This shift, from C to C', is very similar to the shift p to p', which is what would be expected under the standard theory utility maximisation.

Just taking the average of random choices within the budget constraint predicts the same patterns as utility maximisation, without requiring any knowledge of individual behaviour. When the price of good Y declines, people consume more of it, and vice-versa for good X, whether people act randomly or as utility maximisers.

Smith's (and Becker's) simpler approach reframes traditional micro-level topics as the emergent behaviour they are, and leaves the quirky patterns within individual choices to the realm of psychology. This approach makes perfect sense to me.

However, I highly doubt that this idea will be picked up in a hurry by the economics profession for a couple of main reasons. First, it gives away the big prize of economics-- utility and social welfare. If people just behave randomly at an individual level, it breaks the important link between individual choices, higher utility, and social welfare, which forms the backbone of economic thinking, and gives the profession the claim to power in political debates. No longer will economists be the only ones to proclaim that they know the secrets to a better (or higher utility/welfare) society. In fact, they will have to admit that they don’t.

Second, it removes the ability to blame bad choices by individuals as the cause of their economic destiny. If our theory of microeconomics is that people behave randomly within constraints, then to improve outcomes for certain groups of people, we need to change the nature of their constraints, not their decisions. These constraints could be income, wealth, social status and relationships, etc. Solutions to inequality, to homelessness, and other social problems are immediately redirected by this theory back to society at large, and the rules and systems we put in place to create constraints on individuals.

I highly recommend the read. When I finished the book, however, my mind was racing with more ways in which this approach could be applied in more ways across economic topics.

Finally, if you want to read more, you can read Smith's more detailed and technical attempt at piecing together this new approach here.

fn.[1] As a brief disclaimer, I have followed the author's terrific blog for quite a while. I make a point of keeping an eye on original thinkers in general, and he certainly is one, though I’ve never met Smith in person.

10 comments:

  1. Didn't Armen Alchion make the same point back in the 60s?

    It probably wasn't new even back then.

    ReplyDelete
    Replies
    1. There are some connections to Alchian's "Uncertainty, Evolution, and Economic Theory" and he would probably see a lot to like in the book. However there is a broader argument towards separating "mathematical economics" as a result of the opportunity set/possible state space (in terms of information theory and effective theory), and a more social/psychology theory for the deviations from mathematical econ that I haven't seen before (especially the information theory and effective theory references).

      There is also an argument about the properties of symmetrical/high dimensional opportunity sets that I haven't seen anywhere as well as a re-interpretation of Hayek's price mechanism as something that is consistent with information theory but allows failure modes ... failure modes that are discussed throughout that I never found any precedent for in this context.

      Delete
  2. But its unlikely people do in fact act randomly.
    It's hard to see how this is anything but wrong in different ways than RE.
    It also simply pushes the question around by begging the obvious follow up - where do constraints come from?

    This could be a useful result if it happened. It could lead to interesting analysis of the current issue of productivity growth among other things for example.

    And maybe it would force people to go back to a split analysis of the so called political economy comprised of both market goods that can respond to utility price and individual choice AND political goods that don't.

    I don't understand why there is seemingly no attention paid to how political goods are supplied.

    Better analysis of individual agents will probably have to wait for more processing power for the ability to reconcile the math with the social/psychological.

    ReplyDelete
    Replies
    1. It is true that people do not act randomly. When I talk about random agents, I am actually referring to the fact that the upper limit of algorithmic complexity is algorithmic randomness (see e.g. here) -- effectively assuming agents are so complex, they look random. We can consider this a "scope condition" in the sense that where humans appear to be predictable, this assumption will fail. One such example I point out in the book is panic in financial markets: humans predictably sell off instead of "rationally" holding on to assets in a financial panic. We would look at this as a sequence of buy sell that goes from being (effectively) algorithmically random to appearing (effectively) predictable for a long sequence of "sells" correlated among many agents.

      In Gary Becker's diagram above, what happens is that agents bunch up in one corner of the triangle instead of being uniformly distributed throughout.

      Delete
    2. This comment has been removed by the author.

      Delete
  3. The point C is the centre of the triangle A-B-0. So if people consume randomly within that budget constraint at those prices, the average person will consume a combination of goods near point C. When the budget constraint changes to C-D, the new centre point is C'. This shift, from C to C', is very similar to the shift p to p', which is what would be expected under the standard theory utility maximisation.

    That's not obvious to me and there are several difficulties.

    The first one is with "point C": which of the two points C are we talking about? The one on the vertical axis or the one inside the triangle A-B-0?

    Okay, from the context one guesses it's the point C inside the triangle A-B-0.

    Fine. But that point C does not maximise utility: it's inside the triangle, the feasibility set, not on the straight line segment A-B. The consumer is not spending all its money endowment of fiat money. The consumer, in other words, keeps some of that money, which by being fiat has no utility. If "the average person will consume a combination of goods near point C", then it means that the average person does not maximise its utility, but chooses a suboptimal bundle. How does that prove that optimality is an emergent property?

    ReplyDelete
    Replies
    1. "that point C does not maximise utility: it's inside the triangle, the feasibility set, not on the straight line segment A-B. The consumer is not spending all its money endowment of fiat money."

      Yes that's right. That's Jason's point. If consumers spend randomly within their feasibility set, changes to the set will generate behaviour that on average match the changes you would see if they maximised utility.

      He actually explains how that when you extend the problem to many more goods that just two, that you do end up getting very close saturating the budget. This is just the simplest example that fits on a two-axis diagram!

      Delete
  4. Yes that's right. That's Jason's point. If consumers spend randomly within their feasibility set, changes to the set will generate behaviour that on average match the changes you would see if they maximised utility.

    Frankly, I can't see that.

    No matter how much prices change, those households are choosing bundles inside the changing feasibility sets resulting from those price changes. How can the aggregation of points inside feasibility seats result in a point in the budget line?

    ReplyDelete
  5. How does this model accommodate risk aversion? Are people purchasing insurance just randomly, because it exists?

    ReplyDelete