Gate-Aware Economic Forecasting Lab v0.1: First Historical Dry Run on Public Macro Data

This supplement reports the first historical dry run of the Gate-Aware Economic Forecasting Lab. It moves from protocol design toward an executable audit using public macroeconomic data, while deliberately stopping short of a live forecast.

The dry run uses the quarterly U.S. macroeconomic dataset bundled with statsmodels, covering 1959Q1–2009Q3. The dataset is revised-only and therefore does not test real-time or vintage performance. Two coarse directional targets are evaluated: two-quarter-ahead inflation direction and two-quarter-ahead output-growth direction. Four simple model variants are compared: persistence, moving average, a small linear macro model, and a preliminary gate-aware selector.

The result is intentionally modest and constraining: no model passes the gate-aware stability rule in this dry run. For output-growth direction, two models show acceptable average scores while failing critical gates, illustrating a weak false-robustness pattern. For inflation direction, all models fail the disinflation-turn gate.

The dry run therefore does not support any claim of economic forecasting superiority. Its value is methodological: it demonstrates that the registry structure, target definition, gate logic, evaluation code, sensitivity checks, and failure reporting can be executed on public macrodata before any live prediction is issued.

The central conclusion is that the first baseline model set is not gate-stable. Therefore, the lab should improve the pilot before issuing a live public prediction.

Main article DOI: 10.5281/zenodo.20826062
Supplement IV DOI: 10.5281/zenodo.20914535
Supplement V DOI: 10.5281/zenodo.20916725