Information Aggregation through Price and Learning Social Networks with Online Convex Programming and Parametric Dynamics

The statement that price aggregates dispersed information has been a longstanding feature underscoring the importance of markets. Yet, formalizing how exactly price may incorporate individually held information has been a challenging task. I present one particular approach to information aggregation through price.

The framework is a double auction mechanism modeled as an incomplete information game. Traders hold individual information that informs them about the relevant values. `Value’ might capture many things from product quality to differences in tastes or resale opportunities. In the double auction, traders submit bid functions specifying quantities for every price. After bids are submitted, the market clearing price is centrally determined to be the price where quantities to be sold equal quantities to be bought.

The market clearing price is determined by the bid functions, therefore conveys an aggregate of all traders’ information. The two facts, that bids are expressed as functions of price, and that price aggregates other traders’ information; allows traders condition on other traders’ information right when submitting their bids. It is worth noticing, that learning is not explicitly modeled (there are no information channels or dynamic announcements), rather it is a byproduct of optimal trade.

In the presentation, I will focus on the project `Information Aggregation with Correlated Measurement Error’, also mention the ongoing projects on `Privacy Preserving Mechanism Design’, all joint work with Marzena Rostek.