Consider two rationales for building models:
(a) Build a model in order to clarify the signal by filtering out the noise in a complex causal system. This is a knowledge-seeking endeavor.
(b) Build a model in order to be able to say, “In setting X, I can show how you get outcome Y.” This is just playing a game.
An example of playing a game is Akerlof’s Lemons model. In effect, it says, “In a setting where sellers know the quality of the product and buyers do not, sellers of high-quality products will have to settle for low prices, if they choose to sell at all.”
Some remarks:
1. I am pretty sure that economists are unique in their attachment to model-building as a game. My sense is that in other disciplines, including those that study human behavior and those that use non-mathematical models, researchers are more likely to be building models in order to try to separate the signal from the noise in a complex causal system.
2. Countless papers begin by describing a setting as having two factors of production, capital and labor, before adding further wrinkles to the setting. From the knowledge-seeking perspective, I fear that this is a dubious strategy. The two-factor model gets rid of a lot of signal and introduces a lot of noise. But for playing the game (and getting published) it works well.
3. Economists who work in business (think of Hal Varian at Google) do not have the luxury of playing games. If they want to use models to help the firm, they need to build them with the goal of separating signal from noise.
From what I can gather the firms are not keen on human-built models at all, and prefer to collect as much information and intelligence as possible from their operations and then give their sophisticated “Big Data” statistical and learning algorithms a go at it. The economist is there not to model but to help provide some expert judgment, interpretation, and iterative refinement. The market test for the models popular and prestigious in academic publications since to give them a failing grade, favoring analysis of empirical observations on a truly monumental scale.
Most importantly let’s not forget that the world changes, unlike scientific laws. If we look back and see that homeownership has been a good savings vehicle, then we push homeownership…it’s entirely possible that the resulting bump in prices means that homes will no longer be a good savings vehicle as future returns have been pulled forward. The market adjusts.
Arnold, I basically agree with your point but I think the Akerlof lemons piece is not a good example of “playing a game.” I found it, and still find it, profoundly insightful. I’m not saying this is true of you, but I have found that many free-market economists who dump on it (as well as less-free-market economists who apply it uncritically) don’t seem to have read the whole thing. It has been a long time since I’ve read it, but I do recall that after presenting the simple model, Akerlof goes on to talk about various ways people in markets handle, get around, and minimize the lemons problem.
I work in semiconductor and we use models that are so accurate that we can build billion transistor chips and know that they will work when manufactured. That is a different level of accuracy. But we have other models that are less accurate (but faster to use).
I like to point to modeling a plane. One model could be an Airfix-style plastic kit. This model can be used for some things like taking measurements. But it won’t fly.
On the other hand, you can build a paper airplane. This is a big abstraction, but it abstracts some key things, enough that you can fly it. But you can’t take useful measurements from the physical structure.
As George Box said “All models are wrong, but some models are useful.”
Yes, but can your models tell you what your customers will want those chips used in, when and where? Economic forecasters pretend to be able to do that with great accuracy, especially if the only variable they allow is the money supply…
Can you elaborate on #1? My prior is that every sociology paper falls into the “game” category, not the “discovery” category. Why do you believe otherwise?
Akerlof’s lemon paper is not “just” playing a game, it is an understanding-seeking endeavor.
This sort of model is found in some other fields. The Daisyworld model in climatology comes to mind.
David R. Henderson,
I have read the piece (repeatedly, in fact) and wrote a paper critical of it. Yes, Akerlof brings up the problem of information asymmetries impeding efficient exchanges in the neoclassical model of perfect competition.
So just like arguments with externalities, which Coase (1960) critiqued beautifully, Akerlof begins from some idealized position of equilibrium and explains how in a world where information is indeed not perfect, we deviate from that equilibrium.
Well, much like Coase argued, you can’t have it both ways. If we start from that equilibrium, then there are no information asymmetries. If we start from the status quo (i.e, the real world), then the important question is how entrepreneurs are able to bridge those asymmetries in order to bring about efficient exchanges.
Carmax. Carfax. Toyota Certified Pre-Owned. eBay. Amazon Marketplace. These are all just a few of the entrepreneurial endeavors that help us overcome the lemons problem that Akerlof argues keep markets from operating efficiently.
Exactly, Mark, and Akerlof discusses some of these.
As an analogy, Akerlof takes a bowling ball and a feather, puts them in a vacuum and drops them. They fall at the same rate. He then takes both the feather and the bowling ball outside and drops them. The bowling ball falls fast to the ground while the feather slowly drifts with the wind. Akeflof concludes that gravity fails, “but the feather did eventually make its way to the ground.”
He does, but he does it as an afterthought. The basis of his paper is that markets fail to efficiently allocate scarce resources when information asymmetries are present. But they’re never “efficiently” allocating scarce resources if you start from the unrealistic idealized position.
There is a legitimate use of B-style model building: constructing a counter-example against, or for illustrating an unstated assumption of, an A-style model
e.g.
-I construct an A-style model that predicts if X conditions hold, then Y should result
– empirically, if X then Y seems to hold most of the time, but not always
– someone constructs a B-style model showing the if X conditions hold but also novel Z conditions, then Y will NOT result
Here point of the B-style model is not to describe the general, normative case (Y does result most of the time), but exists in commentary to the normative model to expose its limitations.
Akerlof’s lemons are here a good example. Most of the time, higher quality goods command higher prices. But not always. The point of Akerlof’s model is to show that there is an assumed information level in the standard result of higher quality–>higher price. It would be (really, commonly is) a misuse of Akerlof’s model as the default, since empirically the higher quality–> higher price result holds more often than not. Another way to say that is that most of the time, information quality is the kind of noise a good descriptive model filters out. The proper use of the Akerlof model is that when one does encounter a situation where higher quality is not associated with higher prices, than one should probably look first toward the information and quality-signalling method to figure out why
There is another way in which economic modeling is not so unique. The models serve economists in the same way Latin did medieval priests. The laity suspects the clergy is talking bollocks, but can’t be sure.
When one does encounter a situation where higher quality is not associated with higher prices, the explanation seldom is asymmetric information. We live in a world full of easily acquired and cheap information. Specially about lemons.