Panel: Applications of Predictive Modeling and Machine Learning to Improve Policy Implementation
(Tools of Analysis: Methods, Data, Informatics and Research Design)

Saturday, November 4, 2017: 10:15 AM-11:45 AM
McCormick (Hyatt Regency Chicago)

*Names in bold indicate Presenter

Panel Organizers:  Stephan D Whitaker, Federal Reserve Bank of Cleveland
Panel Chairs:  Zubin Jelveh, University of Chicago Crime Lab
Discussants:  Hal Martin, Federal Reserve Bank of Cleveland and John Wilkerson, University of Washington

Predictive Modeling of Surveyed Property Conditions and Vacancy
Stephan Whitaker1, Hal Martin1, Isaac Oduro2, Francisca G.-C. Richter2 and April Urban2, (1)Federal Reserve Bank of Cleveland, (2)Case Western Reserve University

Measuring News Sentiment
Adam Shapiro1, Moritz Sudhof2 and Daniel Wilson1, (1)Federal Reserve Bank of San Francisco, (2)Kanjoya

The use of machine learning, computationally intensive predictive models, and text analysis is rapidly expanding to additional policy applications.  This session showcases a set of related methods applied to a variety of tasks.  Text analysis is applied to news articles and then paired with predictive models to forecast economic activity.  Random forest models are used to predict bank failures, and gradient boosted methods predict surveyor's assessments of the condition of residential properties.  The papers specifically contrast the accuracy of the new methods with the accuracy of long-established methods to demonstrate where value is added.  As off-the-shelf software continues to be developed, practitioners will be able to apply these methods to new issues and improve policy management.