CMStatistics 2017: Start Registration
View Submission - CFE
Title: Nowcasting with large, international data sets: On sparse priors Authors:  Philipp Hauber - Kiel Institute for the World Economy (Germany) [presenting]
Christian Schumacher - Deutsche Bundesbank (Germany)
Abstract: Factor models can summarize the comovements of a large number of variables and have proven useful in nowcasting and short-term forecasting of GDP growth. The main aim is to assess the importance of international variables for nowcasting national developments, an issue, which curiously has received relatively little attention in the academic literature so far. However, given the large number of potentially relevant variables at both the national and international level, the question arises whether all this information is useful for nowcasting or not. As such, we also contribute to the continuing debate on variable selection and the optimal size of factor models for forecasting applications. Rather than choosing variables ad hoc, we employ sparse priors on the factor model's loadings which can help to identify those business cycle indicators that essentially determine the factors, whereas irrelevant variables are sorted out. In an empirical exercise, we evaluate nowcasts of GDP for the Euro area and the United States from models that use only the respective national data as well as the combined data sets comprising of roughly 150 variables.