SFI External Professor Cosma Shalizi is leading a research team that has received a grant from the Institute for New Economic Thinking to extend proven techniques in statistical learning theory so they cover the kind of models and data of most interest to macroeconomic forecasting.

Their effort is the second SFI-affiliated project to gain financial backing in INET’s Inaugural Grant Program. A team led by SFI Professor Doyne Farmer has been selected to receive an INET grant to develop agent-based models of the economy.

Traditional economic models failed to predict or make sense of the global economic crisis that began in 2007. INET seeks to address this failure, and the weaknesses of currently accepted economic theory, by promoting changes in theory and practice through conferences, grants, and education initiatives. INET was founded in October 2009 with a $50 million pledge by George Soros. Starting in 2011, INET will conduct two grant cycles annually.

Shalizi, Mark Schervish, and Daniel McDonald, all of Carnegie Mellon University’s Department of Statistics, plan to use the INET funding to resolve macroeconomic disputes and determine the reliability of whatever new models emerge for macroeconomic time series. The team has deep experience in the statistical analysis of complex systems, including time series prediction, self organization, and network analysis.

“Over the last three decades statisticians and computer scientists have developed sophisticated methods of model selection and forecast evaluation, under the rubric of statistical learning theory,” said INET Executive Director Robert Johnson. “Applying the methods that have revolutionized the modern industry of data mining is an approach that we believe holds great promise in improving the quality of economic forecasts and predictions.”

“Economists have to use models for forecasting and policy-making, which means comparing competing models, which means needing to know how well they will generalize to new data,” said Shalizi, an assistant professor in Carnegie Mellon’s Department of Statistics. “INET is giving us the opportunity to bring that model evaluation up to the same rational level everyone uses in data mining and statistical learning. In the end, we'll show economists how they can reliably select the best models, and control prediction error.”

INET’s Inaugural Grant Program received more than 500 applications from around the world, and INET selected 31 initiatives to be awarded grants totaling $7 million.

INET’s announcement of the award

SFI news article about INET’s previous award to an SFI team for agent-based models of the economy