Google Ad Auctions // 2014 - Present // Google Ad Auctions
Two different basic auction formats are currently used in online ad auctions. Facebook uses Vickrey-Clarke-Groves (VCG), while Google and others have used mainly Generalized Second Price (GSP). Which format is more efficient in creating value? How is surplus split between buyers and sellers? Do bidders converge to equilibrium behavior? Answers are of real interest to economists and are potentially very valuable to organizations that sell online ads, yet so far there is remarkably little knowledge, at least in the public domain. We use laboratory experiments to compare the formats’ efficiency and revenue capture, using proprietary software that enables real-time bid adjustment.
Edgy - Revealed Preferences and General Equilibrium in the Laboratory // 2014 - Present // SES-1357867
General Equilibrium (GE) theory is the mathematical framework that economists use to understand how markets generate prices and distribute resources within a society. However, GE is impossible to directly test in the field because a primary input, the preferences of individuals, is private information not known to outside analysts. Working in the laboratory over several decades, some economists have instead tested aspects of GE theory by inducing artificial preferences. In parallel, another group of economists has studied how to measure the actual preferences of individuals, using a different body of theory called “Revealed Preferences” that infers individual preferences from choices made in the face of varying prices and income levels.
Our first project begins with a slight variation on a prominent revealed preference approach to portfolio choice. Subsequent projects incrementally transform simple preference elicitation into state-of-the-art market trading institutions. Individual decision-making tasks are gradually transformed into full-blown interactive markets, but the mechanics of making choices remains nearly identical throughout, allowing us to reliably compare behavior across settings. A suite of new software tools, called Edgeworks, will implement the entire sequence in a unified visual framework. This research will give insight into the efficiency and stability of interdependent (multi-good) markets, especially important aspects of financial markets. It will also help reveal the robustness of elicited preferences in the face of market experience, including recently controversial aspects such as small stakes risk aversion and reference dependence.
Economic Analysis of Recommender Systems // 2011 - Present // CCF-1101741
A recommender system takes on-line user histories and demographic information, and automatically creates personalized recommendations for goods and services. Already common for small items such as books and movies, such systems are destined to play an increasing role in the economy as new applications emerge such as sourcing or job search. This project will shed broader light on the economic benefits and costs that recommender systems bring to the market.
Boiling the Frog Optimally: A Laboratory Experiment // 2011 - Present // HP Labs Open Innovation 2011
Internet content providers confront a tricky dilemma. Building original content is costly, but generating revenue to cover these costs is generally inconvenient for visitors (fees or distracting ads) & will reduce their numbers. Even if the proper tradeoff is identified, the question remains of how best to introduce the chosen inconvenience: all at once, or in small increments.
One view is that website visitors are like Quinn’s proverbial frogs, and that very few of them will leave if the inconvenience is introduced sufficiently gradually. Another view notes the absence of scientific evidence for the frog proverb, and asserts that it is best to introduce inconvenience all at once. LEEPS Lab is investigating this question.
Continuous Games // 2009 - 2013 // SES-0925039
We build a new computer platform, called Continuous Games (ConG), for studying strategic interaction in the laboratory. ConG focuses on continuous games - especially continuous time, but also continuous action sets and "large" player populations. Such continuous settings are often more realistic and easier to learn, and have novel theoretical properties that have never been tested empirically.
We use ConG to study games of conflict (e.g., prisoner's dilemma and social dilemmas) and games of coordination in which theory predicts far more efficient behavior in continuous time than in discrete time, as well as to study games in which the unique mixed strategy equilibrium has proven elusive in discrete laboratory environments. We also examine a set of applied games-- such as strategic pricing, quantity setting and location choice -- for which continuous time settings are simply more realistic.
The current version of ConG software is downloadable by request. Later it will be accessible via EconPort (a National Digital Library facility). Researchers can run many variants on our games merely by downloading the software and resetting the parameters. With a little programming, they also can run new and quite different continuous games.
Gradient Dynamics as PDEs // 2007 - 2014
Unfunded research. The payoff or fitness function in a population game defines a ``landscape,'' and each player adjusts her action to move up the payoff gradient. Consequently the landscape changes over time. We extend techniques from nonlinear partial differential equations to analyze the resulting dynamics, and present applications in biology, behavioral ecology, politics, economics, and finance.