CMStatistics 2016: Start Registration
View Submission - CMStatistics
B0880
Title: Aggregating the panel of daily textual sentiment for sparse forecasting of economic growth Authors:  Keven Bluteau - Institute of Financial Analysis, University of Neuchatel (Switzerland) [presenting]
Kris Boudt - Vrije Universiteit Brussel and VU Amsterdam (Belgium)
David Ardia - HEC Montréal (Canada)
Abstract: Textual sentiment analysis leads to a high-dimensional panel of high-frequency sentiment measures per text. We show how sparse forecasting methods can be used to extract a predictive signal for forecasting economic growth from the large number of sentiment indices that can be constructed by cross-sectional pooling and time series aggregation of the panel of sentiment by text. We apply the methodology to forecast 1- to 12-month growth in German industrial production using the texts published on LexisNexis. We show that the proposed textual analysis approach yields significant accuracy gains in forecasting, compared to the traditional use of standard time series methods or the use of economic sentiment based on questionnaires.