On October 7, 2015, the Pontifical Catholic University of Rio de Janeiro hosted a conference, “Big Data in Economics,” organized by Marcelo C. Medeiros in PUC-Rio’s Department of Economics.
As one of the keynote speakers, I gave a talk on “Partial Identification via Shrinkage,” a joint project with Anders Bredahl Kock at Aarhus University, Federico Bugni at Duke University, and Soumen Lahiri at North Carolina State University. Big data issues come up everywhere now, such as the difficulties inherent in identifying the main causal effects or contributing factors to economic decision making, and in making good decisions where the environment does not provide clear guidance regarding your decision-making variables. Our project tries to develop a new approach to decision making when the possibilities are many and the data sets are big. An example of a basic application of the paper might be to understand how airline companies enter certain markets; how can we identify important factors in their entry decision into certain U.S. cities? This type of analysis is related to industrial organization issues in economics.
There were several other contributors at the conference, with related techniques for finance and Internet searches: Esfandiar Maasoumi (Department of Economics, Emory University), Anders B. Kock (CREATES and Department of Economics, Aarhus University); and Eduardo F. Mendes, Pedro C.L. Souza, and Medeiros from PUC-Rio. One of the most interesting talks was on Globo.com, a major media company in Brazil that is learning consumer preferences by using big data techniques to analyze “clicks.” The talk was linked to also show how YouTube and Vase are related to what Globo.com is doing. One of their main issues is getting data, then of course storage and the usage of the data to analyze consumer preferences.
Another presentation was on spatial analysis related to social networks. The group of researchers from Brazil and United Kingdom try to understand how the social networks like Facebook evolve. Each social media user is related to a group of friends, and trying to understand how these groups are formed in a very large data environment is a big challenge. They proposed a solution of tools that they newly developed and seem to be important.
Challenges exist, and all of us who were at the conference are trying to develop new tools to tackle them.