Reproducible Python replications of landmark papers in monetary economics and the New Keynesian Phillips Curve. Each replication reproduces the original paper end to end — its theory (the model and estimating equations), its data (downloaded from public sources), and its empirical results — in a single runnable script, with modern identification-robust inference. By Eric Vansteenberghe.
⚠️ Independent replications for study purposes. Where modern or annual cross-country data are used in place of the original samples, results approximate the spirit of the original rather than reproducing its exact numbers.
Replications in this repository
| Original paper | Method | Python replication |
|---|---|---|
| Sims (1980) — Macroeconomics and Reality | VAR / SVAR | code/sims1980_var.py |
| Galí & Gertler (1999) — Inflation Dynamics: A Structural Econometric Analysis | NKPC, GMM | code/gali1999_replication.py |
| Galí, Gertler & López-Salido (2001) — European Inflation Dynamics | NKPC, GMM (euro area) | code/gali1999_euro_replication.py |
| Barnichon & Mesters (2020) — Identifying Modern Macro Equations with Old Shocks | NKPC, shock-IV + Anderson–Rubin | code/barnichon2020_nkpc.py, code/barnichon2020_euro_nkpc.py |
The through-line. The four replications tell one story — about the slope of the New Keynesian Phillips curve and how hard it is to identify. Galí–Gertler (1999) and its euro-area sequel GGL (2001) estimate the NKPC by GMM with lagged instruments; on current-vintage data those moment conditions turn out to be weakly identified, so the textbook standard errors are illusory and the Anderson–Rubin set for the slope is unbounded. Barnichon–Mesters (2020) supply the remedy — identify the curve with a sequence of external monetary shocks — recovering a bounded, larger slope for the US. The same shock-IV applied to the euro area delivers the opposite verdict: monetary shocks move marginal cost but barely move inflation, so the flat post-EMU curve reappears as an identification failure rather than a small coefficient.
Original paper. Christopher A. Sims (1980), "Macroeconomics and Reality," Econometrica 48(1), 1–48. DOI: 10.2307/1912017
Python replication → code/sims1980_var.py
A self-contained replication of Sims's reduced-form VAR and recursive structural VAR (SVAR) analysis of money, output, and prices.
- 📐 Theory — reduced-form vector autoregression; recursive (Cholesky) identification of structural shocks; orthogonalized impulse–response functions; forecast-error variance decomposition (FEVD); Granger / block-exogeneity tests (e.g. money → unemployment, money → inflation); residual-whiteness and stability diagnostics.
- 📊 Data — annual JST Macrohistory Database (Release 6), auto-downloaded:
narrowm(M1 proxy),rgdpbarro(output),unemp,wage,cpi,imports(import-price proxy). Baseline recursive orderingnarrowm, rgdpbarro, unemp, wage, cpi, imports. - 🐍 Code — one command-line script with information-criterion lag selection and a robustness check across alternative recursive orderings.
python code/sims1980_var.py --country USA --maxlags 8 --ic aic --prefer dta --horizon 12
# --no-cache forces a fresh JST downloadCaveat. A reduced-form VAR is a statistical forecasting system, not a structural model on its own; recursive identification makes the variable ordering part of the maintained hypothesis — hence the ordering-robustness check. This is an approximation with annual data, not the exact frequency/sample of Sims (1980).
Original paper. Jordi Galí and Mark Gertler (1999), "Inflation dynamics: A structural econometric analysis," Journal of Monetary Economics 44(2), 195–222. DOI: 10.1016/S0304-3932(99)00023-9
Python replication → code/gali1999_replication.py
A full replication of the estimation of the New Keynesian Phillips Curve (NKPC) with real marginal cost measured by the labor income share.
- 📐 Theory — the forward-looking NKPC and the hybrid NKPC with rule-of-thumb price setters; structural parameters (Calvo price stickiness θ, discount factor β, backward-looking share ω) recovered from the rational-expectations orthogonality conditions; reduced-form vs. structural coefficients (λ, γ_f, γ_b).
- 📊 Data — quarterly FRED series: GDP deflator (
GDPDEF) for inflation, the non-farm-business labor share (PRS85006173) for marginal cost, plus instruments (IPDNBS,GS10,TB3MS,COMPNFB,PPIACO,GDPC1). Original 1960Q1–1997Q4 sample and an extended sample to the latest data. - 🐍 Code — two-step GMM with a Newey–West HAC weighting matrix; both moment
normalizations; restricted-β and non-farm-deflator robustness variants; an
Anderson–Rubin weak-identification-robust confidence set for the marginal-cost
slope; a
--nlagsswitch for instrument-count robustness; and a side-by-side comparison of the published coefficients with the replication.
python code/gali1999_replication.py --sample both
# --refresh re-downloads the FRED series; --nlags 2 uses fewer instrumentsCaveat. With current-vintage data the labor-share–inflation link is weaker than in the paper's 1998 vintage and the curve is weakly identified, so λ comes out smaller, θ higher, and β is driven toward 1 — though the qualitative ranking γ_f > γ_b (forward-looking pricing dominates) still holds. See Weak identification below.
Reference. Jordi Galí, Mark Gertler & J. David López-Salido (2001), "European inflation dynamics," European Economic Review 45(7), 1237–1270 (erratum EER 47(4), 759–760, 2003).
Python replication → code/gali1999_euro_replication.py
GGL (2001) apply the same NKPC/GMM framework to the euro area. This standalone script replicates it on modern data: inflation from the euro-area GDP deflator (FRED), real marginal cost from the euro-area labour income share (Eurostat national accounts, compensation of employees over GDP, adjusted for the self-employed), instruments analogous to the US set, sample 1995Q1–latest. It reproduces GGL's qualitative findings — a positive marginal-cost slope, high price stickiness (θ ≈ 0.83), and an almost purely forward-looking hybrid (ω ≈ 0, γ_f ≫ γ_b) — under the constant-returns (ξ = 1) normalization.
python code/gali1999_euro_replication.py
# --refresh re-downloads FRED + Eurostat series; --nlags sets instrument depthGMM estimation of the New Keynesian Phillips Curve is weakly identified: the moment conditions are nearly uninformative about the structural parameters, so conventional GMM standard errors and the over-identification (J) test are unreliable. This is now the central methodological issue for NKPC estimation. Two diagnostics in this repository make it concrete:
- Degenerate J-test. With the paper's 25 instruments — a 25×25 HAC weight
matrix estimated from only ~120–150 quarters — the J-test has almost no power and
its p-values pin near 1.00. Shrinking the instrument set (
--nlags 2or--nlags 1) restores power, and the model is still not rejected. - Anderson–Rubin robust confidence set. The scripts compute the weak-instrument-robust Anderson–Rubin / Stock–Wright S confidence set for the marginal-cost slope λ (profiling out β on the continuously-updated objective). For both the US and the euro area, and for every instrument set tried (7–25 instruments), this set is unbounded — once identification-robust inference is used the data are essentially uninformative about λ, and the tight conventional standard errors are illusory.
How to do it properly. Do not rely on Wald t-statistics / standard errors under weak identification. Use identification-robust inference: the Anderson–Rubin / Stock–Wright S-statistic and the Kleibergen K-statistic, evaluated on the continuously-updated (CUE) GMM objective, and report confidence sets (which may be unbounded or empty) rather than point estimates ± standard errors; keep the instrument set parsimonious to limit many-instrument bias. See Kleibergen & Mavroeidis (2009) and Mavroeidis, Plagborg-Møller & Stock (2014) in the References — and the shock-IV remedy of Barnichon & Mesters (2020) below.
Original paper. Régis Barnichon and Geert Mesters (2020), "Identifying Modern Macro Equations with Old Shocks," Quarterly Journal of Economics 135(4), 2255–2298. DOI: 10.1093/qje/qjaa022
Python replication → code/barnichon2020_nkpc.py
(US, Table I) · code/barnichon2020_euro_nkpc.py
(euro-area application)
Identifies the New Keynesian Phillips Curve with a sequence of identified monetary shocks as instruments — an IV regression "in impulse-response space" — using weak-instrument-robust Anderson–Rubin inference. The headline: conventional lagged-instrument methods substantially underestimate the slope of the Phillips curve.
- 📐 Theory — the hybrid NKPC instrumented by a sequence of structural (monetary) shocks; the exogeneity/relevance conditions; the regression "in impulse-response space"; the Almon / quadratic-polynomial reduction to three instruments; the subset Anderson–Rubin (min-eigenvalue) statistic and the continuously-updated estimator under γ_f + γ_b = 1.
- 📊 Data — US: the authors' QJE replication file (inflation, output /
unemployment gap, and the Romer–Romer (2004) narrative monetary shocks).
Euro area: GDP-deflator inflation and the labour share (FRED + Eurostat) with the
Jarociński–Karadi pure-monetary shock (
MP_median, median-rotation identification), taken directly from M. Jarociński'sjkshocks_update_ecbrepository. - 🐍 Code — a faithful Python port of the authors' MATLAB reproducing Table I exactly (point estimates and Anderson–Rubin confidence intervals), plus a euro-area application contrasting shock-IV against lagged-instrument IV on the same sample.
python code/barnichon2020_nkpc.py # US, reproduces Table I (needs the QJE replication data)
python code/barnichon2020_euro_nkpc.py # euro-area shock-IV vs lagged-IV (needs the JK shock series)Findings. For the US (1969–2007), monetary-shock instruments deliver a bounded Anderson–Rubin confidence set for the Phillips-curve slope — positive, significant, and larger than conventional lagged-instrument estimates. For the euro area (2004–2024), the same external shocks move real marginal cost but barely move inflation (first-stage F ≈ 16 for the labour share vs. ≈ 4 for inflation), so the slope is not identified (unbounded AR set), while lagged instruments return a spurious wrong-signed estimate — the flat, anchored euro-area Phillips curve appearing as an identification failure rather than a small coefficient.
Data note. The US script reads
Data_QJE.xlsxfrom the authors' QJE replication package (git-ignored here — download it from the QJE replication materials). The euro-area script needs the monthly Jarociński–Karadi euro-area shock seriesshocks_ecb_mpd_me_m.csv, taken directly from M. Jarociński'sjkshocks_update_ecbrepository (git-ignored here).
| Replication | What it reproduces | Headline result |
|---|---|---|
| Sims (1980), US | reduced-form VAR + recursive SVAR; IRFs, FEVD, Granger tests | money–output–price dynamics; conclusions hinge on the recursive ordering (ordering-robustness check included) |
| Galí–Gertler (1999), US | NKPC by two-step GMM; structural θ, β, ω | forward-looking pricing dominates (γ_f > γ_b); marginal-cost slope weakly identified on current-vintage data |
| GGL (2001), euro area | the same NKPC/GMM on the euro-area labour share | positive slope, high stickiness (θ ≈ 0.83), near-purely-forward-looking; Anderson–Rubin set unbounded |
| Barnichon–Mesters (2020), US | shock-IV NKPC — reproduces Table I exactly | bounded AR set; slope larger than the lagged-instrument estimates that underestimate it |
| Barnichon–Mesters, euro area | Jarociński–Karadi shock-IV vs. lagged-IV (one sample) | shocks move marginal cost (F ≈ 16) but not inflation (F ≈ 4) → slope unidentified; the flat euro-area curve |
.
├── README.md
├── LICENSE
└── code/
├── sims1980_var.py # Sims (1980) VAR / SVAR (CLI)
├── gali1999_replication.py # Galí & Gertler (1999) NKPC (GMM, US)
├── gali1999_euro_replication.py # Galí–Gertler–López-Salido (2001) NKPC (GMM, euro area)
├── barnichon2020_nkpc.py # Barnichon & Mesters (2020) shock-IV NKPC (US, Table I)
└── barnichon2020_euro_nkpc.py # Barnichon–Mesters method applied to the euro area
data/ (downloaded datasets and caches) and third-party replication packages are
created on first run / supplied by the user and are git-ignored.
- JST Macrohistory Database, Release 6 — auto-downloaded from macrohistory.net (Sims replication).
- FRED (Federal Reserve Bank of St. Louis) — US series
GDPDEF,PRS85006173(non-farm labor share),IPDNBS,GS10,TB3MS,COMPNFB,PPIACO,GDPC1, and euro-area aggregatesCLVMNACSCAB1GQEA19,CPMNACSCAB1GQEA19,IRLTLT01EZM156N,IR3TIB01EZQ156N. - Eurostat — euro-area national accounts (
namq_10_gdp,namq_10_a10_e) for the labour income share. - Romer & Romer (2004) narrative monetary shocks — from the Barnichon–Mesters QJE replication package (US shock-IV).
- Jarociński–Karadi euro-area monetary-policy shocks — the
MP_median(median-rotation) pure-monetary shock, from M. Jarociński'sjkshocks_update_ecbrepository (euro-area shock-IV).
Python 3.10+.
pip install numpy pandas scipy statsmodels requests matplotlib openpyxlWorking-directory note. The replication scripts set a local working directory via
os.chdir(...)near the top; edit it to your own checkout (or remove it and run from the repo root) before executing.
Econometrics implemented from scratch in NumPy/SciPy, cross-checked against
statsmodels / linearmodels where an independent implementation exists:
- Vector autoregression (VAR) and recursive structural VAR (SVAR) — Cholesky identification, orthogonalized impulse–response functions, forecast-error variance decomposition (FEVD), Granger / block-exogeneity tests, information-criterion lag selection, stability and residual-whiteness diagnostics.
- GMM estimation of the NKPC — two-step efficient GMM, the continuously-updated estimator (CUE), Newey–West HAC weighting, both moment normalizations, and structural recovery of Calvo price stickiness θ, the discount factor β, and the backward-looking share ω.
- Weak-identification-robust inference — Anderson–Rubin / Stock–Wright S and subset-AR (min-eigenvalue) confidence sets, the Kleibergen K statistic, and reporting confidence sets (possibly unbounded or empty) instead of Wald standard errors; instrument-count robustness to limit many-instrument bias.
- Shock-IV / external-instrument identification — IV "in impulse-response space," a sequence of identified monetary shocks (Romer–Romer; Jarociński–Karadi) as instruments, Almon / quadratic-polynomial instrument reduction, a quadratic-spectral HAC kernel, and first-stage relevance diagnostics.
- Sims, Christopher A. (1980). "Macroeconomics and Reality." Econometrica 48(1), 1–48. DOI: 10.2307/1912017
- Galí, Jordi & Mark Gertler (1999). "Inflation dynamics: A structural econometric analysis." Journal of Monetary Economics 44(2), 195–222. DOI: 10.1016/S0304-3932(99)00023-9
- Galí, Jordi, Mark Gertler & J. David López-Salido (2001). "European inflation dynamics." European Economic Review 45(7), 1237–1270. RePEc · NBER w8218 DOI: 10.3386/w8218. Erratum: EER 47(4), 759–760 (2003).
- Barnichon, Régis & Geert Mesters (2020). "Identifying modern macro equations with old shocks." Quarterly Journal of Economics 135(4), 2255–2298. DOI: 10.1093/qje/qjaa022
- Romer, Christina D. & David H. Romer (2004). "A new measure of monetary shocks: Derivation and implications." American Economic Review 94(4), 1055–1084.
- Jarociński, Marek & Peter Karadi (2020). "Deconstructing monetary policy
surprises — the role of information shocks." American Economic Journal:
Macroeconomics 12(2), 1–43. DOI: 10.1257/mac.20180090.
Euro-area shock series: M. Jarociński,
jkshocks_update_ecb. - Kleibergen, Frank & Sophocles Mavroeidis (2009). "Weak instrument robust tests in GMM and the new Keynesian Phillips curve." Journal of Business & Economic Statistics 27(3), 293–311. DOI: 10.1198/jbes.2009.08280
- Mavroeidis, Sophocles, Mikkel Plagborg-Møller & James H. Stock (2014). "Empirical evidence on inflation expectations in the New Keynesian Phillips curve." Journal of Economic Literature 52(1), 124–188. DOI: 10.1257/jel.52.1.124
Eric Vansteenberghe. If you use this code, please cite the original papers listed in the References.
Keywords: monetary policy · macroeconometrics · replication · Python · VAR · SVAR · structural VAR · impulse response · FEVD · New Keynesian Phillips curve · NKPC · hybrid Phillips curve · GMM · two-step GMM · continuously-updated GMM · CUE · Newey–West HAC · weak identification · Anderson–Rubin · Stock–Wright S-statistic · identification-robust inference · Kleibergen–Mavroeidis · shock-IV · external instruments · monetary policy shocks · impulse-response space · Romer–Romer · high-frequency identification · EA-MPD · Jarociński–Karadi · Calvo pricing · inflation dynamics · marginal cost · labor share · euro area · Galí–Gertler · Galí–Gertler–López-Salido · Barnichon–Mesters · FRED · Eurostat · JST Macrohistory.