Title: Real-time density nowcasts of U.S. inflation: A model-combination approach
Authors: Edward Knotek - Federal Reserve Bank of Cleveland (United States) [presenting]
Saeed Zaman - Federal Reserve Bank of Cleveland (United States)
Abstract: A model combination framework is developed to produce density nowcasts for U.S. headline and core inflation at a trading-day frequency. We apply a flexible aggregation strategy and several weighting schemes to combine individual density nowcasts coming from three classes of mixed-frequency inflation nowcasting models. We assess the properties of the density nowcasts generated from the individual classes of models and from the grand combination of the three model classes using high-frequency, real-time data over the period 2000-2015. As information accumulates over a month or quarter, density nowcast accuracy steadily improves as judged by predictive scores, but only the grand combination is always among the most accurate and comes closest to passing all the necessary tests to be correctly calibrated. Both point and density nowcasts from the grand combination outperform survey benchmarks in the case of headline inflation.