Wednesday, February 24, 2010

Market design at Yahoo!

Here's a project description from Yahoo! Research: Microeconomics and Market Design
By Michael Schwarz and David Reiley


What are the effects of complexity in mechanism design? Participating in an auction mechanism (or any market institution) requires time and effort on the side of the buyers. That in turn may lead some buyers to bid very conservatively, or not to participate at all, in order to economize on effort. We lack a theory for quantifying the complexity of a mechanism.


Auction theory and matching theory offer models and algorithms for allocating goods or matching market participants. However, the majority of markets do not use structured mechanisms (such as auctions and matching algorithms). Why are some markets (such the medical residency match) using centralized matching procedure while other markets (such as the purchases of automobiles with various options packages) do not? Answering this question may help to design structured market mechanisms for the markets where chaotic negotiations rule the day.

Understanding when and how reputations emerge, and how to improve the designs of structured reputation systems. Reputations of individuals and businesses play an important role in economic and personal life. With some exceptions, most notably in electronic commerce, the mechanisms for accumulating reputation are informal rather than structured.

How should a firm set posted prices for a menu of products? Although economics offers a theoretical framework for how to set prices, most theory assumes omniscience about demand. To build a practical pricing engine, one needs a system that generates exogenous price variation and then uses it to calculate appropriate prices going forward. Yahoo! offers thousands, even millions, of different advertising products, so estimating a demand system suffers from the curse of dimensionality as well as practical constraints (perceived fairness, etc.) in generating price variation. We are looking for better theory and econometrics for the problem of building a pricing engine.

No comments: