Title: Improving accuracy of early quarterly estimates of GDP components in the U.S. national accounts
Authors: Baoline Chen - Bureau of Economic Analysis (United States) [presenting]
Abstract: Differences in the source data used to compile the three quarterly GDP releases underlie a substantial part of the revisions between the first two releases and the third. Such revisions can be sizeable for some GDP components, especially during the periods of rapid changing macroeconomic conditions. Revisions to the early releases could also be due to a lack of sufficient relevant monthly information and a lack of information on the longer term dynamics of the GDP component series. The aim to develop a structured approach for compiling early estimates of GDP components in the national accounts, combining information on the longer-term characteristics of the component series and short-term movements in the series resulting from the changing economic conditions. The proposed estimation approach is primarily based on the bridge-equation and bridging with factor model in the now-casting literature, using a set of relevant current and lagged monthly indicators to provide more accurate information on the component series in the current quarter while employing lagged quarterly information to provide longer term dynamics for the component series. We apply this approach to the detailed component series in the PCE of services with the objective of minimizing the root mean squared revision (RMSR) statistic.