AI Economist Skill Adds ML Calibration to GDP Nowcasting and Policy Analysis

Summary

The GitHub project introduces an installable agent skill for macroeconomic nowcasting, central-bank policy diagnostics, and economist-style interpretation in the United States and Canada. Its main GDP forecast is a structural bridge-equation model with SVD factor extraction and ragged-edge filling for recent missing monthly data. Machine learning is deliberately secondary: it calibrates bounded residuals and explains where the structural model may be missing information, while baseline and calibrated estimates are reported side by side. The policy module provides Taylor 1993, Taylor 1999, and nonlinear inflation-response variants, using labor, output, capacity, inflation, financial, external, and other macroeconomic indicators to produce base and data-enhanced signals. The skill includes expanding-window out-of-sample backtesting for GDP nowcasts, release-lag-aware pseudo-real-time feature filtering, and comparable baseline-versus-calibrated R2 and RMSE reporting. It discloses that the tests use revised data rather than a full vintage-data evaluation, and states that causal claims require a separate identification design. Reports are expected to include the target period, data-through date, source coverage, factor decomposition, validation window, leakage controls, and limitations. The installable bundle is the economics-ml/ directory; the repository dashboard, tests, and assets are supporting project files. Live data pulls and backtests require Python dependencies and a local FRED API key.