How GDPcast works

The current quarter and next (h=0–1) are driven by bridge equations — monthly source data, largely the same series BEA uses to build the NIPA tables, mapped directly into quarterly component growth. As each government release lands, the affected component updates and the change flows straight to the headline. This is what makes the forecast “live,” GDPNow-style, and what makes attributing a forecast change to a specific data release possible.

Quarters 2 through 8 are driven by one error-correction equation per component, estimated on quarterly history back to 1995 and driven by assumption paths for rates, oil, the dollar, wages, home prices, corporate profits, foreign growth, tariffs, and federal fiscal policy. The two engines blend smoothly around h=1 rather than switching abruptly.

Real GDP components are chain-weighted, so they don't sum arithmetically. GDPcast aggregates the same way BEA does in Table 1.1.2 — each component's growth is weighted by its nominal share of GDP to produce a contribution in percentage points, and the headline is the sum of those contributions. That arithmetic is what powers the subcomponent contribution chart and the forecast-change log on the home page.

A handful of estimated linkages tie the components together: disposable income feeds consumer spending, domestic final sales feeds imports, and the gap between final sales and inventories pulls the inventory forecast back toward trend. A consistency dashboard checks every run against historical ranges — the implied saving rate, import share of final sales, and investment shares of GDP — and flags anything an assumption path implies that history hasn't seen.

Nominal GDP is published alongside the real forecast: nominal growth is built from real growth plus GDP-deflator inflation, itself derived from the PCE inflation assumption plus a small estimated wedge for import prices and government/investment deflator trends. In the current quarter, actual published price data takes precedence over the assumption.

Before publishing, the model was backtested pseudo-real-time over 2016–2025, replaying the government release calendar week by week using as-first-published (ALFRED vintage) data where available. Results are compared against a simple per-component AR benchmark, the Atlanta Fed's GDPNow archive at matched points in the data cycle, and the Survey of Professional Forecasters median for the outer quarters. 2020 is reported separately rather than folded into the averages.

Coefficients are re-estimated annually, after BEA's fall NIPA revision, with a before/after report — the published forecast moves only because of new data or an assumption change, never silent coefficient drift in between.

All inputs are free, public data: the Bureau of Economic Analysis, the Census Bureau, the Bureau of Labor Statistics, the Federal Reserve (FRED), and the U.S. Treasury. GDPcast is built and maintained independently by Economic Impact Group and is not affiliated with or endorsed by any of these agencies.

The home page shows the current-quarter nowcast and how it has evolved as data has landed this quarter. The full 8-quarter forecast, scenario comparisons, and component-level detail are published four times a year for subscribers to the Weekly Economic Update at alfiemeek.com.