The Non-Nobel in Economics: Mistaking Games for Reality
On October 15, 2012, the non-Nobel Prize in Economics was awarded to two Americans Alvin Roth and Lloyd Shapley for the development of game theory and its application to resource allocation. Now at first blush this might seem an innocuous, if odd, choice, certainly not in line with the Swedish Central Bank's usual selection of neoliberal economists, but it really isn't. This approach to allocating resources or market design is really just a way to dress up and sell the efficient markets hypothesis. It gives a mechanism for how markets can arrive at optimal distributions of resources. The problem is that it is another neoclassical fairy tale. Or put more simply, life is not a game.
Roth who is the economist of the two has been credited with applying this model to student selection of schools in New York City, to kidney programs, even to speed dating.
But before we look at these, let's consider some of the requirements for such a model.
1. A clearly defined and understood preference
2. A small number of clearly defined variables which set the parameters of the preference
3. These variables must be easily hierarchized or weighted
4. All participants must have access to similar information and the ability to process it
Now in the example of the New York schools, it was reported that students might have a dozen different schools which they might select. So how do they go about prioritizing their selections? Well, that will depend on what they (and their parents) know about the schools and what they want. But on both counts the knowledge of the student and his/her parents could be highly variable. And this is where we run into our first problem. They may not know what they want and even if they do, they may not have sufficient or accurate enough information to make an informed choice. This immediately goes against the last point of the neoclassical idea of a population of similarly knowledgeable, hence interchangeable agents.
It is important to distinguish between a list and what someone wants. That is to say that not all preferences are the same. A student may really want to go to one school but feel the school might not choose them so they might leave it off their list or give it a lower place on their list. A more likely scenario is that out of a dozen choices a student might only like two or three and hate two or three and be indifferent about the rest. So if a student gets one of those top two or three, then they're happy. If they don't, they will get a lower preference in their list, but importantly, this preference will not reflect something they want.
The key here is that any system which matches a reasonable number of students with their top preferences will be seen as fair and effective. But as we proceed down the curve, this will be less and less the case. And you have to wonder about the students going to schools that no one wants to go. How does the model account for them?
In this case, the model has some usefulness, but is oversold. It will provide a solution to some, especially if they are flexible, but far from all.
One of the thought experiments for this game model is the selection of a best possible mate from a group. So you have say 30 males and 30 females and you match them up according to a set of variables. The game designers never seem to consider what happens if they have 60 incompatible people. Certainly, they can be matched up, but the matches are meaningless. This is the problem when you try to reduce a complex decision to a simple one, or an irrational/unconscious decision to a rational and conscious one. Many people may not know what they want in a mate, even when they think they do. Few would be able to come up with a weighted list, certainly nothing quantifiable, and they probably would be considered pretty strange if they did. Speed dating is an even worse example precisely because it tries to reduce this complex decision into an even more simplified form. Indeed beyond its novelty value, and that only for some, speed dating would appear more a means to filter out undesirable dates/mates than choosing desirable ones.
The one area where this model has some utility in uniformizing criteria and improving networking is in kidney transplant lists. That is precisely because pertinent variables can be identified and weighted with regard to patients on the list. Unlike the other examples where inputs come from those directly involved, here the inputs come from and are evaluated by a third party group, medical professionals. Further, there is no market here, but a matching process.
So where this model works best is in the case which has the least to do with actual markets and where the agents (the patients) are passive in the process.
In any real market situation, there is complexity, information asymmetry, variables which can not be quantified, trade offs, and let's be frank about current markets, fraud and looting. Stated preferences may have little or nothing to do with actual results. The process can be gamed. And the resource allocation may have more to do with steering resources into a CEO's compensation than in any productive business activity. That's the real world, the one we live in, and as I said at the start, it is not a game.