Atlanta Fed GDPNow

GDPNow is a statistical model from the Federal Reserve Bank of Atlanta that tracks growth for each measured quarter. A nowcast means a rolling estimate that updates as source data arrives for that quarter. It estimates what the Bureau of Economic Analysis advance estimate for that quarter is heading toward, not growth in later quarters.

The model covers the United States only and aggregates 13 spending parts with chain weighting that mimics official method. It draws on 124 monthly series from public sources such as trade, retail, income and production reports. There is no survey panel and no human judgment, only mathematical results from the model.

Each cycle covers one quarter and starts about 90 days before the advance estimate for that quarter. The cycle opens on the weekday after the prior advance and ends on the closing business day with a release before the advance. Starts can shift by about a week after major annual revisions when source detail tables are delayed.

Expect 6 or 7 weekday updates in a month, usually within hours after morning source releases. Updates follow 7 reports covering factory activity, trade, wholesale, retail, housing, durables and income, with other reports folded into the next update day. Each update revises tracked growth and the parts behind it, while model code stays fixed until after the advance.

Growth reflects domestic spending plus government, with trade and inventories often moving the total. Contributions shift with each update as new monthly clues arrive.

• Consumer goods and services: household spending on goods plus everyday services, the largest domestic weight.
• Business equipment and software: tools, machines, vehicles, software and research type investment by firms.
• Structures and housing: offices, factories, homebuilding, improvements and sales commissions.
• Government: federal plus state and local consumption and investment.
• Net exports: goods and services exports add growth while matching imports subtract, measured without gold flows.
• Inventories: the quarterly change in unsold stock, the most volatile tracked part and hard to nowcast.

The publisher reports an average absolute error of 0.77% and a root mean squared error of 1.17% for final estimates. Use that record to gauge uncertainty, not as a signal to buy or sell.

The base uses a multi lag Bayesian model for each of the 13 parts and their price measures over a long window. Nominal monthly values are split with price measures, and missing price measures come from a multi variable monthly price model. One shared activity factor comes from 124 monthly series and moves forward on its own motion, smoothed to remove noise.

Factor guided models project each missing month, then actual values plus forecasts combine into quarterly moves. Consumption uses monthly retail maps and vehicle sales directly without a bridge. Investment and government parts link monthly clues to quarterly growth and blend that link with the base estimate. The blend uses weights that sum to a full total.

Trade divides nominal values by price indexes for real flows, while stocks combine book values with price based valuation. The 13 parts then aggregate with updated weights, imports enter with a minus sign, and the sum prints as seasonally adjusted annual growth.