{"slug": "ai-economist-an-installable-agent-skill-for-macro-nowcasting", "title": "AI Economist, an installable agent skill for macro nowcasting", "summary": "A new installable agent skill called AI Economist, hosted at github.com/garroshub/ai-economist-skill, provides macroeconomic nowcasting, central-bank policy diagnostics and economist-style interpretation, with machine learning used only as an auxiliary calibration layer rather than the primary forecaster. The skill installs from the economics-ml/ folder, which contains SKILL.md, a Python runtime and requirements, and covers GDP nowcasts via a structural bridge model with SVD factor extraction, Base and Data-Enhanced Taylor rule readouts, and baseline-versus-calibrated out-of-sample backtesting with release-lag filtering. The repository's dashboard/, tests/, src/ and assets/ folders are project-side files not required for the installed skill.", "body_md": "An installable agent skill for macroeconomic nowcasting, central-bank policy\n\ndiagnostics, and economist-style interpretation. The installable skill lives in\n[`economics-ml/`](https://github.com/garroshub/ai-economist-skill/blob/main/economics-ml); the dashboard and tests are project-side\n\nsupporting files.\n\n| Layer | Main Output | Role of ML | \n|---|---|---|\n| GDP nowcast | Bridge-equation baseline by country and quarter | Bounded residual calibration, reported separately | \n| Policy diagnostics | Base Taylor and Data-Enhanced Taylor readouts | Historical residual calibration against observable macro factors | \n| Backtesting | Baseline vs calibrated OOS comparison | Prior-window calibration only, release-lag filtered | \n| Agent report | Driven factors, data-through dates, validation notes | Structured interpretation, not a hidden predictor | \n\nML is not the primary forecaster. It is an auxiliary calibration and measurement layer used to explain where the structural model may be missing information.\n\nInstall the `economics-ml/` folder for normal use. It contains `SKILL.md`, the\nPython runtime, and requirements. The repository-level `dashboard/`, `tests/`,\nand `assets/` folders are not part of the installed skill.\n\nmacOS/Linux:\n\n```\ntmpdir=\"$(mktemp -d)\"\ngit clone --depth 1 https://github.com/garroshub/ai-economist-skill.git \"$tmpdir/ai-economist-skill\"\nrm -rf ~/.codex/skills/economics-ml\ncp -R \"$tmpdir/ai-economist-skill/economics-ml\" ~/.codex/skills/economics-ml\nrm -rf \"$tmpdir\"\ncd ~/.codex/skills/economics-ml\npip install -r requirements.txt\n```\n\nWindows PowerShell:\n\n``` php\n$skillDir = \"$env:USERPROFILE\\.codex\\skills\\economics-ml\"\n$tmpDir = Join-Path $env:TEMP \"ai-economist-skill\"\nRemove-Item -LiteralPath $tmpDir -Recurse -Force -ErrorAction SilentlyContinue\ngit clone --depth 1 https://github.com/garroshub/ai-economist-skill.git $tmpDir\nRemove-Item -LiteralPath $skillDir -Recurse -Force -ErrorAction SilentlyContinue\nCopy-Item -Recurse -LiteralPath \"$tmpDir\\economics-ml\" -Destination $skillDir\nSet-Location $skillDir\npip install -r requirements.txt\n```\n\nAfter installation, ask the agent for tasks such as:\n\n```\nUse the Economics ML skill to explain the latest US and Canada GDP nowcast drivers.\nUse the Economics ML skill to compare Base Taylor and Data-Enhanced Taylor signals.\nUse the Economics ML skill to audit whether the backtest has forward-information leakage.\n```\n\nThe skill should answer with target period, data-through date, baseline estimate, data-enhanced estimate, driven factors, validation checks, and limitations. It should not merely point the user to a Python command.\n\nIf you only want the agent to produce structured interpretation from\nuser-provided values or existing reports, `SKILL.md` alone is enough. Live data\npulls and backtests require the runtime bundle above.\n\nClone the full repository only for development, dashboard work, or tests:\n\n```\ngit clone https://github.com/garroshub/ai-economist-skill.git\n```\n\nIn a full clone, `dashboard/`, `tests/`, `src/`, and the Python entrypoints are\nproject files. Only `economics-ml/` is required for installing the agent skill.\n\n- Structural bridge model with SVD factor extraction.\n- Ragged-edge filling for missing recent monthly observations.\n- US high-frequency auxiliary variables: initial claims, housing, durable goods, real disposable income, financial conditions, and yield curve.\n- Canada auxiliary variables: StatCan retail sales, CPI, CAD/USD, WTI oil, and US demand spillover measures.\n- Baseline and ML-calibrated nowcasts shown side by side.\n\n- Taylor 1993, Taylor 1999, and nonlinear inflation-response variants.\n- Output-gap proxies from labor slack, HP-filtered GDP, and capacity utilization.\n- Base Taylor result remains the main structural signal.\n- Data-Enhanced Taylor result learns historical residual adjustments from activity, inflation pressure, financial conditions, external pressure, and labor cooling.\n- Enhancement decomposition is reported in percentage-point contributions.\n\n- Expanding-window backtest for GDP nowcasts.\n- Release-lag-aware pseudo-real-time feature filtering.\n- Baseline R2/RMSE and ML-calibrated R2/RMSE shown on the same validation window.\n- Revised-data limitation is disclosed rather than treated as a vintage-data test.\n\nThe skill can be used for analysis without running local scripts. Install the runtime dependencies only when you want live data pulls, regenerated reports, backtests, or local dashboard builds.\n\n```\ncd economics-ml\npip install -r requirements.txt\nset FRED_API_KEY=your_key_here\n```\n\nRun GDP nowcasts:\n\n```\npython main.py gdp --country US\npython main.py gdp --country Canada\n```\n\nRun policy diagnostics:\n\n```\npython main.py policy --country US\npython main.py policy --country Canada\n```\n\nRun backtests:\n\n```\npython backtest_engine.py\n```\n\nBuild the dashboard:\n\n```\ncd dashboard\nnpm ci\nnpm run build\n```\n\n- FRED API for macro indicators.\n- BLS and Bank of Canada public data where available.\n- Statistics Canada public CSV tables for retail-sales auxiliary features.\n- Public news and outlook pages only when converted into structured measurement variables.\n\nNo private API key is stored in the repository. Set `FRED_API_KEY` locally before\nrunning live workflows.\n\nReports should include the forecast target period, data-through date, source coverage, structural baseline, calibrated result, driven-factor decomposition, backtest window, leakage controls, and limitations. Causal claims require a separate identification design.\n\n```\nai-economist-skill/\n|-- README.md\n|-- economics-ml/\n|   |-- SKILL.md\n|   |-- main.py\n|   |-- backtest_engine.py\n|   |-- requirements.txt\n|   `-- src/\n|-- dashboard/\n|-- assets/\n`-- tests/\n```\n\nFor skill installation, use only `economics-ml/`. The dashboard, tests, and\nassets support the public project page and validation workflow.", "url": "https://wpnews.pro/news/ai-economist-an-installable-agent-skill-for-macro-nowcasting", "canonical_source": "https://github.com/garroshub/ai-economist-skill", "published_at": "2026-10-09 03:26:55+00:00", "updated_at": "2026-10-09 03:46:52.351290+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "machine-learning", "developer-tools"], "entities": ["AI Economist", "garroshub/ai-economist-skill", "economics-ml", "Taylor 1993", "Taylor 1999", "StatCan", "GitHub"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/ai-economist-an-installable-agent-skill-for-macro-nowcasting", "markdown": "https://wpnews.pro/news/ai-economist-an-installable-agent-skill-for-macro-nowcasting.md", "text": "https://wpnews.pro/news/ai-economist-an-installable-agent-skill-for-macro-nowcasting.txt", "jsonld": "https://wpnews.pro/news/ai-economist-an-installable-agent-skill-for-macro-nowcasting.jsonld"}}