CS Colloquium - David Pennock
Refreshments available at 3:45
Hosts: Dragomir Radev and Joan Feigenbaum
Title: Expressive markets to crowdsource joint probabilities
A prediction market, like a financial market, elicits an accurate estimate of one expected value. I will discuss expressive prediction markets, or markets designed to elicit millions of probabilities structured in a continuous or joint distribution. First, I will present a market allowing traders to buy contracts that pay $1 if a continuous random variable, like the S&P 500’s opening price on January 17, 2020, falls within a specified interval. The market uses an automated maker maker that sells intervals with endpoints of arbitrary precision, for example “between 2,297.54 and 3,251.70”, at any time for some price. The market maker runs in time logarithmic in the number of partitions and has bounded loss, the first of its kind. Its prices imply an arbitrarily precise probability distribution over the random variable, improving on the discretization and lack of expressiveness in financial options markets. Second, I will present a combinatorial financial options exchange that accepts orders for call and put options with any strike price on any portfolio of stocks, for example the right to buy 2 shares of Google and 21 shares of Microsoft for $4481.67 on January 17, 2020. Orders are matched using linear programming. The exchange can recreate any standard option and endless more, including custom mutual funds, and combines liquidity across strikes, stocks, and funds. Prices reflect correlations among securities. If I have time, I will describe other combinatoria prediction market designs and the research prototypes that we have built and run live as online games.
David Pennock is a Principal Researcher at Microsoft Research New York City. His largest contributions are novel prediction markets and wagering mechanisms: financial markets harnessed to elicit probabilistic information from a crowd. He has over 70 publications cited 14,000 times and an h-index of 50. He has over twenty patent applications, over twenty press mentions, and has given more than fifty talks. His Ph.D. is in artificial intelligence and he has been an intellectual and organizational leader in the economics-and-computation subfield of AI for two decades. He co-founded two research areas, three workshops, and an ACM journal, and was a founding member of three corporate basic-research labs. He served as Assistant Managing Director of MSR NYC for six years. He was Chair of ACM SIGecom, Program co-Chair of ACM EC, and is co-Editor-in-Chief of ACM TEAC. In addition to his primary research area, he has published work in machine learning (including NeurIPS and ICML), theory (including STOC), information retrieval (including a Test of Time Award honorable mention in SIGIR), web science, sponsored search, Bayesian networks, constraint satisfaction, and recommender systems. He led the development of several popular online market games and blogged for Yahoo News. In 2005, he was named to MIT Technology Review’s list of 35 “top technology innovators under age 35” having the potential to profoundly impact the world.