Thursday, September 8, 2011

Instructions are part of a market's design

Whenever my colleagues and I help design a new marketplace, we're very aware that a part of the market mechanism are the instructions that accompany it. There's no reason to assume that the benefits of a strategy-proof mechanism, for example, will be realized if the participants aren't made fully aware that it is strategy-proof, so that it is safe for them to reveal their true preferences.

That is why I was glad when the HBS MBA program invited me to explain the modified serial dictatorship mechanism that Clayton Featherstone and I designed for the first year of operation of a 2nd year MBA field experience module, in which Harvard MBA students will choose countries in which to spend time at a company.  (We felt it was particularly important to start with a strategy-proof mechanism, for reasons we hope to write about in the not too distant future.)  Here's the video of my explanation (which can also be found at http://video.hbs.edu/videotools/play?clip=aroth_field2_algorithm or, if that is gated, http://stream.hbs.edu/remediated/cd/aroth_field2_algorithm.mp4)




While I think of it, let me mention that Clayton is an unusually talented and versatile market designer, theorist and experimenter who will be on the econ job market this year.

1 comment:

dWj said...

"Strategy-proof" implies a "direct mechanism", in which the action set is the type space, of which I think the salient characteristics are "easy to understand" and "robust to beliefs". Typically "easy to use" requires a curtailment of the action set; rather than ask for a full specification of preferences over lotteries, we ask for ordered preference over outcomes, for example, and may assume away (in various contexts) complementarities that we have reason to believe are unimportant.

It seems likely to me that there is some context in which a stronger reduction in the action set will be easier to use, even if it no longer looks like a direct mechanism -- it requires that the agents have an intuitive sense of a best-response function on the types they are likely to have. I'm not coming up with an example at the moment, but if I look for one, it will be in a situation that closely resembles a problem that people are used to solving. (Actually, this is one reason "approval voting" is popular among certain less ambitious voting-reform types; the optimal strategy resembles that for first-past-the-post.)