poly: Re: idea markets (was GMU vs. UCB)

In a message dated 99-02-17 15:04:38 EST, hanson@econ.berkeley.edu (Robin
Hanson) writes:

> My job talk at UCB info had this theme: the potential for using markets to
> better aggregate information has been neglected by economists because they
> don't do "prototypes," i.e., designs lacking specific theoreticial models.
> It's all right to construct models of specific institutions, and all right
> to test such models in lab experiments, but its not all right to do lab
> tests of institution designs w/o such models, even if these designs are made
> using rules of thumb derived from more rigorous tests.
 
> In other more engineering oriented disciplines, one is allowed to build
> designs using accepted rules of thumb and then do lab tests of them.
> In economics only a few senior people can get away with this. Thus the
> long empirically-demonstrated remarkable ability of speculative markets
> to aggregate information has not been followed up by attempts to
> better design markets specifically for this purpose. Such markets are
> too complex to do detailed models of with current tools. I pitched this
> idea to UCB, so if they accept me, that will be an endorsement of my plan
> to do such prototype testing in their department.

Very intriguing concepts. I'm thinking of this more from the science
information
point of view, and of course we have to have systems for the manipulation
and evaluation of speculative data in science. In my decision on where to
go, I have to evaluate things like what kind of system would be superior for
generating and evaluating ideas - will I do better at a smaller school where
everybody knows each other or at a larger school where individuals are more
isolated? It's a pity we have little real data on this kind of thing and that
my decision will consequently be ill-informed on this matter.

To rephrase my ideas, the engineering *does* get done on these systems in
the sense that we do create speculative markets and other information
evaluation
systems, but the decisions tend to be ill-informed since the empirical data
is scanty. So it seems like your idea should be an easy go. I'm surprised
UCB wasn't interested but I wasn't there.

Biologists often address the issue of how to test unmodelable systems by
making extreme simplifying assumptions. Often, these work even when
the assumptions are seriously bogus. An example would be quantitative
genetics. Quantitative genetics basically assumes that a) each gene
has a small effect on the outcome and b) effects of different genes can
be added without considering complex interaction. Both these assumptions
have recently come up very wrong. Nonetheless, quantitative genetics
has been spectacularly successful in breeding and in gene mapping.
In addition, testing a theory, even one you know is wrong, can provide "cover"
for developing techniques.
Received on Wed Feb 17 18:18:30 1999