Real-Time Forecasting with High-Frequency Seasonal Patterns

In this paper, we propose a novel, comprehensive approach to interpolate low-frequency of- ficial statistics from high-frequency data. Differently from standard mixed-frequency dynamic factor models, commonly used for nowcasting, we leverage on recent developments in nowcasting modeling to build a methodology that can easily deal with outlier detection, complex calendar patterns and temporal disaggregation. We deploy the new methodology to introduce a new weekly tracker for real activity in the United Kingdom based on the several new, high-frequency data provided by the Office for National Statistics (ONS). Results suggest that these new data sources, when properly managed via our model, introduce significant improvements in the predictive accuracy of traditional nowcasting models, generally based on lower-frequency data, in terms of both point and density forecasts.