Title: Advances in nowcasting economic activity
Authors: Juan Antolin Diaz - Fulcrum Asset Management (United Kingdom) [presenting]
Thomas Drechsel - LSE (United Kingdom)
Ivan Petrella - Warwick Business School (United Kingdom)
Abstract: Dynamic factor models (DFM) have become the workhorse model for nowcasting economic activity. Exploiting recent advances in Bayesian computational methods, we extend the DFM framework along four dimensions. First, we model low-frequency movements in the growth rate and the volatility of the variables. Second, we allow for heterogeneous lead-lag patterns in the responses of the variables to the common factor. Third, we introduce automatic outlier detection by modeling fat tailed observations in the variables. Fourth, we endogenously model seasonal fluctuations, which is particularly useful whenever there is suspicion that ``residual seasonality'' is present. We then put our modeling innovations to the test in a comprehensive out-of-sample evaluation exercise using fully real-time unrevised data for seven countries. As the model is re-estimated each time new information arrives, the sheer scale of the exercise requires massive computational power and is made possible thanks to the use of cloud computing. Paying special attention to the production of well-calibrated density forecasts, we show how low frequency movements, dynamic heterogeneity, outliers and seasonality are pervasive features of macroeconomic data and their modeling advances our understanding of the real-time assessment of macroeconomic conditions.