Goldcorp, one of the world's top gold producers, was on the brink of
bankruptcy as a result of its inability to reliably estimate the
value and location of gold on its property. As a final maneuver, the
company published its geological data and asked people to submit
their estimates and estimation methods in exchange for prize money
that totalled a little over half a million dollars. The crowd
identified 110 targets that led to the extraction of a staggering $8
million ounces of gold, transforming the $100 million company into a
$9 billion powerhouse.
This is an example of crowdsourcing, a technology where innovation and production are sourced out to the public. At its core is a resource allocation problem: there is an abundance of workers but a scarcity of high skills, and an easy task assigned to a skilled worker is a waste of resources. The standard solution to this problem comes in the form of knowledge hierarchies, where
tasks are first attempted by low-skilled labor and high-skilled workers only engage with a task if workers with lesser skills are unable to finish it. Organizing these hierarchies in crowdsourcing is difficult because firms have little or no information about the skill
of the workers they can hire and the firm-worker relationship is
fleeting and temporary, providing an incentive for workers to
misrepresent their skills. To complicate matters, the exact
difficulties of the constituent tasks may not be known in advance
(for example, a manager may want a piece of code that accomplishes
something, but does not know how difficult it is to write such code),
so tasks that require skilled labor cannot be identified and priced
ahead of time. I design a dynamic pricing mechanism for tasks that
induces crowdsourced workers to endogenously sort themselves into the
desired optimal hierarchy, so that the outcome of the mechanism
corresponds to the outcome obtained if the firm knows the workers'
skills and the difficulties of the tasks.
Mrs. Gertrude Walton, an 83 year old grandmother from West Virginia, was sued in 2005 by the Recording Industry Association of America (RIAA) for illegally sharing 700 pop and rap songs online. The case made the news because of two conspicuous aspects: Mrs.Walton did not even own a computer. She was also dead.
Why would the RIAA do such a foolish thing? The answer may lie in the observation that online piracy displays a strong social component; people who have friends who illegally procure digital goods with no repercussions are more inclined to attempt the same behavior compared to people who know someone who got punished for similar actions. Starting from this simple model, I derive the optimal policy to control the rate of piracy. Remarkably, the resulting policy corresponds exactly to the one employed by the RIAA from the period 2003 onwards. To the best of my knowledge, this is the only game-theoretic model whose predictions match the actual observed policy.
Network formation and opinion dynamics have usually been studied in isolation. Yet, the dependence of one on the other is undeniable: people form connections with those who hold similar opinions or interests and can sever ties with those whose opinions are very different from theirs. At the same time, network structure can accentuate some interactions over others and thus skew the opinions of society in a certain way, especially with the presence of popular agents (political analysts, celebrities, famous researchers, etc.). What happens when these two processes simultaneously evolve? What are the resulting opinions and the nature of clusters formed? These questions have bearing on a wide range of issues, from what makes certain products trendy to how extremism can develop in a population. I provide a characterization of the evolution of this dynamical system and discuss how it can be controlled.
Computational advertising is probably the fastest moving inventory system in existence. Goods (opportunities to display an ad) arrive randomly at an extremely fast rate and have to be allocated immediately to fulfill advertisers' demand. I model the problem of display advertising, where advertisers pay publishers to place ads alongside content on websites, as a multi-period stochastic inventory control problem where supply is uncertain (since the opportunity to display a particular ad depends on users' browsing patterns) and advertisers have penalties associated with shortage or other contractual terms (for example, penalties associated with uneven spread in display of ads). Since it is impossible for the publisher to allocate each display opportunity individually, the decision variables in each period are the fraction of the still unrealized supply to be allocated to each advertiser. I derive the structure of the publisher's optimal allocation policy and show how it can be approximated so that the publisher obtains the most cost-effective policies given their computation budget.