So What Have You Been Maximizing Lately?

A while back, Brad DeLong referred to Ezra Klein’s review of Tyler Cowen’s book Discover Your Inner Economist. (Which I own but haven’t yet read; if it’s as interesting as the blog, I’m sure it will be great.) The question involves rational action in the face of substantial mark-ups on the price of wine in nice restaurants:

I did once try to convince Bob Hall at a restaurant in Palo Alto not to order wine: the fact that the wine would cost four times retail would, I said, depress me and lower my utility. Even though I wasn’t paying for it, I would still feel as though I was being cheated, and as I drank the wine that would depress me more than the wine would please me.

He had two responses: (i) “You really are crazy.” (ii) “Think, instead, that it’s coming straight out of the Hoover Institution endowment, and order two bottles.”

He is crazy, of course — crazy like an economist. I left a searingly brilliant riposte in the comment section of the post, which mysteriously never appeared. He will probably claim it was a software glitch or that I hit “Preview” instead of hitting “Post,” but I know better. What are you afraid of, Brad DeLong!?

Economists have a certain way of looking at the world, in which (to simplify quite a bit) people act rationally to maximize their utility. That sort of talk pushes physicists’ buttons, because maximizing functions is something we do all the time. I’m not deeply familiar with economics in any sense; everything I know about the subject comes from reading blogs. Any social science is much harder than physics, in the sense that constructing quantitative models that usefully describe the behavior of realistic systems is made enormously difficult by the inherent nonlinearities of human interactions. (“Ignoring friction” is the basis of 98% of physics, but nearly impossible in social sciences.) But I can’t help speculating, in a completely uninformed way, how economists could improve their modeling of human behavior. Anyone who actually knows something about economics is welcome to chime in to explain why all this is crazy (very possible), or perfectly well-known to all working economists (more likely), or good stuff that they will steal for their next paper (least likely). The freedom to speculate is what blogs are all about.

Utility is a map from the space of goods (or some space of outcomes) to the real numbers:

U: {goods} -> R

The utility function encapsulates preferences by measuring how happy I would be if I had those goods. If a set of goods A brings me greater utility than a set B, and I have to choose between them, it would be rational for me to choose A. Seems reasonable. But a number of issues arise when we put this kind of philosophy into practice. So here are those that occur to me, over the course of one plane ride across a couple of time zones.

  • Utility is non-linear.

This one is so perfectly obvious that I’m sure everyone knows it; nevertheless, it’s what immediately popped into mind upon reading the wine story. We need to distinguish between two different senses of linear. One is that increasing the amount of goods leads to a proportional increase in utility: U(ax) = aU(x), where x is some collection of goods and a is a real number. Everyone really does know better than that; the notion of marginal utility captures the fact that eating five deep-fried sliders does not bring you five times the happiness that eating just one would bring you. (Likely it brings you less.)

But the other, closely related, sense of linearity is the ability to simply add together the utility associated with different kinds of goods: U(x+y) = U(x) + U(y), where x and y are different goods. In the real world, utility isn’t anything like that. It’s highly nonlinear; the presence of one good can dramatically affect the value placed on another one. I’m also pretty sure that absolutely every economist in the world must know this, and surely they use interesting non-linear utility functions when they write their microeconomics papers. But the temptation to approximate things as linear can lead, I suspect, to the kind of faulty reasoning that dissuades you from ordering wine in nice restaurants. Of course, you could have water with your meal, and then go home and have a glass of wine you bought yourself, thereby saving some money and presumably increasing your net utility. But having wine with dinner is simply a different experience than having the wine later, after you’ve returned home. There is, a physicist would say, strong coupling between the food, the wine, the atmosphere, and other aspects of the dining experience. And paying for that coupling might very well be worth it.

Physicists deal with this by working hard at isolating the correct set of variables which are (relatively) weakly-coupled, and dealing with the dynamics of those variables. It would be silly, for example, to worry about protons and neutrons if you were trying to understand chemistry — atoms and electrons are all you need. So the question is, is there an economic equivalent to the idea of an effective field theory?

  • Utility is not a function of goods.

Another in the category of “surely all the economists in the world know this, but they don’t always act that way.” A classic (if tongue-in-cheek) example is provided by this proposal to cure the economic inefficiency of Halloween by giving out money instead of candy. After all, chances are small that the candy you collect will align perfectly with the candy you would most like to have. The logical conclusion of such reasoning is that nobody should ever buy a gift for anyone else; the recipient, knowing their own preferences, could always purchase equal or greater utility if they were just given the money directly.

But there is an intrinsic utility in gift-giving; we value a certain object for having received it on a special occasion from a loved one (or from a stranger while trick-or-treating), in addition to its inherent value. Now, one can try to account for this effect by introducing “having been given as a gift” as a kind of good in its own right, but that’s clearly a stopgap. Instead, it makes sense to expand the domain set on which the utility function is defined. For example, in addition to a set of goods, we include information about the path by which those goods came to us. Path-dependent utility could easily account for the difference between being given a meaningful gift and being handed the money to buy the same item ourselves. Best of all, there are clearly a number of fascinating technical problems to be solved concerning strategies for maximizing path-dependent utility. (Could we, for example, usefully approximate the space of paths by restricting attention to the tangent bundle of the space of goods?) Full employment for mathematical economists! Other interesting variables that could be added to the domain set on which utility is defined are left as exercises for the reader.

  • People do not behave rationally.

This is the first objection everyone thinks of when they hear about rational-choice theory — rational behavior is a rare, precious subset of all human activity, not the norm that we should simply expect. And again, economists are perfectly aware of this, and incorporating “irrationality” into their models seems to be a growth business.

But I’d like to argue something a bit different — not simply that people don’t behave rationally, but that “rational” and “irrational” aren’t necessarily useful terms in which to think about behavior. After all, any kind of deterministic behavior — faced with equivalent circumstances, a certain person will always act the same way — can be modeled as the maximization of some function. But it might not be helpful to think of that function as utility, or as the act of maximizing it as the manifestation of rationality. If the job of science is to describe what happens in the world, then there is an empirical question about what function people go around maximizing, and figuring out that function is the beginning and end of our job. Slipping words like “rational” in there creates an impression, intentional or not, that maximizing utility is what we should be doing — a prescriptive claim rather than a descriptive one. It may, as a conceptually distinct issue, be a good thing to act in this particular way; but that’s a question of moral philosophy, not of economics.

  • People don’t even behave deterministically.

If, given a set of goods (or circumstances more generally), a certain person will always act in a certain way, we can always describe such behavior as maximizing a function. But real people don’t act that way. At least, I know I don’t — when faced with a tough choice, I might go a certain way, but I can’t guarantee that I would always do the same thing if I were faced with the identical choice another hundred times. It may be that I would be a lot more deterministic if I knew everything about my microstate — the exact configuration of every neuron and chemical transmitter in my brain, if not every atom and photon — but I certainly don’t. There is an inherent randomness in decision-making, which we can choose to ascribe to the coarse-grained description that we necessarily use in talking about realistic situations, but is there one way or the other.

The upshot of which is, a full description of behavior needs to be cast not simply in terms of the most function-maximizing choice, but in a probability distribution over different choices. The evolution of such a distribution would be essentially governed by the same function (utility or whatever) that purportedly governs deterministic behavior, in the same way that the dynamics in Boltzmann’s equation is ultimately governed by Newton’s laws. The fun part is, you’d be making better use of the whole utility function, not just those special points at which it is maximized — just like the Feynman path integral established a way to make use of the entire classical action, not just those extremal points. I have no idea whether thinking in this way would be useful for addressing any real-world problems, but at the very least it should provide full employment for mathematical economists.

Okay, I bet that’s at least three or four Sveriges Riksbank Prizes in Economic Sciences in Memory of Alfred Nobel lurking in there somewhere. Get working, people!

58 Comments

58 thoughts on “So What Have You Been Maximizing Lately?”

  1. Hmm, some thoughts

    a) Economists are perfectly agreed that utility is non-linear. Indeed, the concept of marginal utility is of incredible importance in determining prices. Economists talk about the value of the marginal unit all the time, which doesn’t make sense if the marginal unit’s value is the same as the average value.

    b) I think your statement (b) is too strong. Utility is a function of goods. It’s also a function of other things, but goods definitely come into it. Imagine, for example, that you have hiked with your loved ones into the wilderness, two days away from the nearest road or shop or cafe. You have a wonderful camp spot, a beautiful view of a sunset over a local gate, and any other immaterial good you care to add in (I don’t know if that means you’ve just finished two hours of yoga or whatever). I’m willing to bet that, all other things being equal, your utility will be higher if you also have some material objects, like some food.
    There’s nothing in economics to stop one adding non-material utility in to a person’s utility function. For example, once my husband asked me what I meant when I said that I loved him, and I blinked in surprise and replied “That’s simple, I mean your utility is part of my utility function.” I would have gone on to add that my utility function increased with respect to his utility but he was laughing too hard for me to get a word in edgewise. I don’t know why.

    Economists do use linear equations in modelling budget constraints. For example, a person cannot spend more money than they have (taking into account their ability to borrow). So we can write that someone is maximising their utility with respect to goods available on the market subject to a budget constraint, where the sum of all the goods they buy multiplied by the price must be less than or equal to a set amount of money. Perhaps you have been confused by these budget constraints?

    c)I agree that the world “rational” can be confusing. But I also agree with Bad that the impulse to assume that people are rational unless proven otherwise has lead to some really fascinating discoveries.

    d) I’m not sure how much value would be added by including non-deterministic models of utility. Economists don’t try to model every single behaviour someone does, we don’t know anyone’s actual utility functions so even with a deterministic model we don’t know what any one person would actually choose. What we use utility models for is things like explaining how prices are calculated (price = marginal utility), or saying that we don’t know if labour supply increases or decreases in response to an increase in the hourly wage. (This is something taught in Econ 101. Basically, imagine a self-employed worker called Carol who can increase or decrease her hours at will. And for a moment just think of Carol as only being interested in money and leisure. Her hourly rate goes up. Now Carol can earn more each hour she works, which creates an incentive to increase her hours, the substitution effect. But, since she’s richer, she can also increase her leisure while maintaining the same money income by decreasing hours – the income effect. The conclusion is that we don’t know if the subsitutition effect or the income effect will dominate in Carol’s case so we don’t know if she will work more or less hours in response to an increase in her hourly rate). Now, if we can’t tell what Carol, or you, would do, with a deterministic model, what’s the value of going to a non-deterministic model?

  2. Tracy
    wrt d)
    What you are saying is that if the utility function of any individual is unknown to us, it doesn’t make it any less unknown to allow it be indeterminate. True enough. But if for instance we were to build an object oriented model of the world to similate individual behaviour in response to a particular environment it might indeed matter that an individual’s behaviour is not consistant over time (i.e. that observed behaviour in the past is not a foolproof guide to observed behaviour in the future). For a marketing firm for instance, that might be important (implying for instance they cannot be as narrowly focused on a particular subset of customers as they might think from past experience).

  3. Within the context of hte current conversation, astrology is not irrational. Astrology may be mistaken, but astrology is not irrational.

    The concept of “rationality” in economics deals with decisions made given your beliefs about the world. It doesn’t, at this level, say what those beliefs should be.

    Making costly (in the literal sense of cost > 0, not the figurative sense of cost = a lot) decisions on the basis of astrological advice when you don’t believe in astrology would be irrational. Not doing so when you do believe in astrology would also be irrational.

  4. Pingback: Healthcare Economist · Utility Function

  5. As a transport economist I know that the models that we use do not assume that people behave deterministically. In fact there is a whole field of behavioural modelling based on Random Utility, where utility is treated as a random variable with a specified probability distribution.

    It works best in analysing simpler systems – such as making a selection from a fixed set of discrete choices (such as “Will I drive, or catch a bus?”). Of course the analysis gets pretty ugly pretty quickly unless you make some pretty heroic assumptions. Any non-simulation based solution requires very simple models, with identically distributed terms, no covariance, and simple normal-like distributions.

    At least in my experience, there are still not too many models using the tools that I understand that computational physicists are using, such as Markov Chain Monte-Carlo models etc. But they are coming…

  6. Alex#38,
    You could make things path dependent in a flat space with a multiply connected topology. Then homotopic paths (assuming we are parallely propagating the goods from a source to a recipient throughout some path) would have path independent values amongst themselves but different values from a curve which is not of the same homotopy type. Cheers

  7. well….guys…..what do you think are some errors that occurs when solving or evaluating the #’s of utility..using the cardinal method?

Comments are closed.

Scroll to Top