Title: Monthly GDP growth estimates for the U.S. states
Authors: James Mitchell - Federal Reserve Bank of Cleveland (United States) [presenting]
Gary Koop - University of Strathclyde (United Kingdom)
Stuart McIntyre - University of Strathclyde (United Kingdom)
Aristeidis Raftapostolos - University of Strathclyde (United Kingdom)
Abstract: A mixed frequency vector autoregressive model (MF-VAR) is developed that exploits available state- and US-level data at the monthly, quarterly, and annual frequencies to produce timely nowcasts and historical monthly estimates of GDP growth for the 50 states of the US (plus Washington, DC) from 1964 through the present day. The model imposes temporal and cross-sectional constraints to ensure that the monthly estimates both 'add up' to published quarterly/annual data and that the GDP estimates for the 50 states (plus DC) sum to published GDP data for the US as a whole. We develop a computationally-fast approximate Bayesian Markov Chain Monte Carlo (MCMC) algorithm for estimating and nowcasting with this large-scale MF-VAR. The model is used to produce historical estimates of monthly GDP for the 50 (plus DC) US states back to the 1960s, and the utility of these estimates is illustrated by using them to understand business cycle dynamics and cross-state dependencies better. A nowcasting application, using real-time data, then shows how the model can be used to produce density estimates of state-level GDP two months ahead of the BEA's quarterly state-level estimates. We show the importance for nowcast accuracy of conditioning these state-level nowcasts on the latest estimates of US GDP from the BEA.