Skip to content

🇬🇧 English · 🇪🇸 Español

puremacro

Macroeconometrics and structural modeling on the core scientific-Python stack.

PyPI Version Python 3.11+ JupyterLite Playground License: MIT

What is puremacro?

Research workflows: independent evidence and structural estimation.

puremacro brings econometric estimators and quantitative macroeconomic models together using NumPy, SciPy, pandas and Matplotlib, with no compiled extension of its own. A local workstation installation is the supported target. The numerical core also runs under Pyodide / WebAssembly; browser and tablet workflows remain best effort, with limits on file formats, network access and computation. See the runtime guide and the structural validation status for the scope of the evidence.

Key Subsystems

  1. Structural DSGE & Dynare Parity (3.0 – 3.2):
  2. Native Dynare .mod Parser & puremacro-dynare CLI: Parse and solve standard .mod files directly in pure Python.
  3. Exact Analytic Score Recursions: Symbolic AST differentiation and generalized Sylvester solver evaluating \(\nabla_\theta \ln L\) in a single forward pass without finite differences.
  4. Pure-Python HMC & NUTS: Hamiltonian Monte Carlo with dual-averaging step-size adaptation, diagonal mass adaptation, and rank-normalized split-\(\hat{R}\) diagnostics.
  5. Optimal Discretionary Policy vs Commitment: Dennis (2007) Riccati matrix iterations for Markov-perfect discretionary policy vs timeless-perspective commitment, decomposing inflation and stabilization biases.
  6. DSGE-VAR Hybrid Modeling: Del Negro & Schorfheide (2004) DSGE-VAR(\(\lambda\)) prior optimization and structural identification.
  7. Anticipated News Shocks: State-space companion augmentation with nilpotent shift operators preserving Blanchard-Kahn determinacy.
  8. Nonlinear Particle Filtering & MS-DSGE: Bootstrap and auxiliary particle filters with stochastic volatility, Differentiable OccBin for NUTS, and Foerster et al. (2016) Markov-switching rational expectations solutions with closed-form analytical GIRFs.

  9. Second-Order Pruned Perturbation & OccBin: Schmitt-Grohé & Uribe (2004) cross-derivatives with Kim et al. (2008) pruning, and Guerrieri & Iacoviello (2015) piecewise-linear Zero Lower Bound (ZLB) regimes.

  10. Heterogeneous-Agent Models (HANK & VFI):

  11. Sequence-Space HANK (Auclert, Bardóczy, Rognlie & Straub 2021, Econometrica): General equilibrium incomplete-markets models solved in \(\mathcal{O}(T^3)\) sequence space.
  12. Fake News Algorithm: Fast \(\mathcal{O}(T^2)\) computation of sequence-space consumption Jacobians via expectation vectors and cumulation identities.
  13. HANK Sequence-Space Bridge: Declare hetagent_block directly inside standard Dynare .mod files.
  14. Targeted Fiscal Transfers: Dynamic consumption IRFs and cumulative fiscal multipliers across wealth deciles.

  15. Continuous-Time HJB: Achdou et al. (2022) implicit upwind finite-difference M-matrix schemes, the adjoint Kolmogorov forward equation for the stationary density, and continuous-time Aiyagari general equilibrium.

  16. Econometric Engines & Local Projections:

  17. Unified LPResult: Standardized Local Projections (lp_hac, lp_iv, lp_state_dep, panel_lp) with Newey-West HAC, fixed-\(b\), and Driscoll-Kraay standard errors.
  18. Structural VARs & FAVAR: Cholesky, Blanchard-Quah, Sign Restrictions, Narrative Sign Restrictions, Proxy/External Instruments, Max-Share/News, and Factor-Augmented VAR.
  19. Modern Difference-in-Differences: Staggered adoption estimators robust to treatment heterogeneity (Callaway & Sant'Anna, Sun & Abraham, Borusyak-Jaravel-Spiess, Synthetic DiD, Honest DiD).

  20. Spatial Econometrics & Global VAR: Conley spatial HAC, spatial autoregressive models (sar, sem, sdm), spatial local projections, and the Pesaran Global VAR (var.gvar).

  21. Applied Macroeconomic Policy Suites (puremacro 3.2):

  22. Central Bank Policy Stance & Fan Charts: Real-time monetary policy stance evaluation against counterfactual Taylor rules, multi-period fan chart projections, and statement sentiment extraction (Guide).
  23. Real-Time Nowcasting & News Decomposition: Giannone-Reichlin-Small mixed-frequency dynamic factor models (DFM), handling asynchronous ragged-edge data vintages and release-day news attribution (Guide).
  24. Macroprudential Risk & Financial Stability: Adrian-Boyarchenko-Giannone Growth-at-Risk (GaR) conditional skew-\(t\) densities, and Diebold-Yilmaz systemic connectedness networks (Guide).

  25. Fiscal Policy & Sovereign DSA: Multi-method fiscal multipliers (SVAR vs LP vs Narrative LP-IV) and stochastic sovereign Debt Sustainability Analysis fan charts under macro-climate stress scenarios.

  26. Global Macro Data Ecosystem & Modular Panels (puremacro 3.2):

  27. Emissions & Climate Data: World Bank WDI and OECD SDMX greenhouse gas accounts across countries and economic sectors (Guide).
  28. Energy Balances & Commodities: Primary energy consumption, clean energy transition shares, and expanded World Bank Pink Sheet commodity benchmark suites (Guide).
  29. International Financial Stability: Sovereign yield curves (10Y/2Y), central bank policy rates, BIS credit-to-GDP gaps, and real property prices (Guide).

  30. Modular Panel Builders: One-call constructors build_climate_panel and build_financial_panel with automated frequency harmonization (M\(\to\)Q, A\(\to\)Q).

  31. Causal ML, Continuous-Time Heterogeneous Agents & Regional Real Time (puremacro 3.4):

  32. Double / Debiased Machine Learning: Chernozhukov et al. (2018) partially linear regression with Neyman-orthogonal Robinson scores, \(K\)-fold cross-fitting, and pure-NumPy penalized learners — lasso and ridge only, no elastic net.
  33. Implicit HJB, Adjoint KFE & Continuous Aiyagari GE: Sparse M-matrix policy iteration, a stationary density normalized with trapezoid quadrature weights so it integrates to one on non-uniform grids, and a general-equilibrium loop whose converged flag means \(|K^s - K^d| \le\) tol_ge.
  34. Multi-Constraint OccBin: \(M \ge 2\) simultaneous occasionally binding constraints, with a regime bitmask (bit \(k\) set when constraint \(k\) binds) and non-convergence reported as converged=False plus a warning naming each reason (Guide).
  35. Montiel Olea-Pflueger Weak-IV Inference: Effective \(F\)-statistic critical values from the closed form \(\chi^2_{K,\,1-\alpha}(K/\tau)/K\), and multi-instrument Anderson-Rubin sets inverted exactly rather than on a grid (Guide).

  36. Latin American Real-Time Connectors: Banco de México, INEGI, Banco Central do Brasil and Banco Central de Chile. These four sources publish only the current edition, so their vintage date is the local snapshot date and revision history accumulates only across repeated captures on different days.

  37. Publication-Grade Reporting:

  38. Zero-dependency, camera-ready table export directly to LaTeX (.to_latex()), Typst (.to_typst()), and Markdown (.to_markdown()), complete with standard errors and significance stars (Guide).

  39. Running Anywhere:

  40. A Pyodide-compatible numerical core for browser notebooks, with best-effort tablet and browser support (Guide), Google Colab offloading helpers (runtime.colab), chunked long-run execution (longrun), and portable .pmz data cartridges (pocket).

Installation

pip install puremacro

The parquet and Excel readers (the course's ENOE lessons, build_all, the shock atlas) need the optional file-format engines:

pip install "puremacro[io]"

Or install with full notebook tools:

pip install "puremacro[notebooks]"