Jorge Jesus Alonso Ortiz

Jorge Jesus Alonso Ortiz

Research, teaching, coding… and dogs

ITAM · Mexico City

Gallery

Overview · Gallery · From Stata & Dynare · Run it anywhere

Every figure, with the code that made it

Every figure below was made by code that ships with puremacro: one of its 92 example scripts, one of its showcase notebooks or one of its guides. Click a figure to enlarge it; the link under it opens the code. The figures are restyled for this site, and the scripts draw them in their own style.

New in 5.1 · New in 4.8 · New in 4.2 · New in 4.0 · New in 3.3 · Impulse responses & identification · Replications on real data · Structural models · Time variation, regimes & volatility · Data, nowcasting & text

New in 5.1

Three panels: the savings rule a-prime of a for the lowest and highest productivity states around the 45 degree line; the Lorenz curve of wealth from the stationary distribution, with a Gini of 0.3494; and the Gini of a simulated panel of 5,000 households that all start at a = 0.1, rising to 0.4449 at period 10 and settling at 0.3496 by period 199, next to the exact value
Notebook 21’s precautionary-savings model, declared with dp: mean assets 6.5733 and a wealth Gini of 0.3494, both exact. A simulated panel that starts everyone at a = 0.1 overshoots to 0.4449 at t = 10 and settles at 0.3496 by t = 199. · Notebook 21
Three panels: mean assets and consumption by age in a 60-period life cycle, with assets peaking at the retirement age of 45; the probability of working by assets for the lowest, median and highest productivity states, with an overall participation rate of 63.3 percent; and simulated aggregate capital in a Krusell-Smith economy rising from 14 to fluctuate around a mean of 34.5
Three models from the dp guide, solved by EGM: a life cycle whose mean assets peak at retirement (age 45), a work-or-not choice with taste shocks (participation 63.3%), and a Krusell–Smith economy whose forecast rule converges in 8 rounds with R² of 0.99996. · Code
Two panels: Lorenz curves of household net worth in the SCF 2025 and in an Aiyagari model calibrated to the median wealth-to-income ratio, the data curve far below the model's; and five moments, data against model: Gini 0.824 against 0.590, top 10 percent share 0.721 against 0.379, top 1 percent share 0.313 against 0.063, bottom 50 percent share 0.021 against 0.088, and share with net worth at or below zero 0.086 against 0.091
SCF 2025 against an Aiyagari economy with β calibrated to the median wealth-to-income ratio (β = 0.9521, s.e. 0.0023). The model’s Gini is 0.590 against 0.824 in the data, and its top 1% holds 6.3% of net worth against 31.3%: one asset and earnings risk miss the top, as the literature expects. · Code

New in 4.8

Ten rows, one per panel builder, each showing how many economies have data in each year from 1900 to 2025, with peaks of 169 economies for the Maddison Project, 62 for BIS consumer prices, 188 for IMF consumer prices, 45 for OECD monthly consumer prices, 147 for ILOSTAT employment, 185 for the Penn World Table, 64 for OECD annual GDP, 41 for 10-year yields, 214 for World Bank GDP and 41 for Eurostat GDP
Economies with data in each year, by one-call panel builder, rebuilt offline from the cached live pulls of 7 October 2026. Consumer prices reach back to 1913 (BIS) and 1914 (IMF, OECD); the Maddison Project goes back to 1 AD. · Code
Two panels: Mexico's real GDP per capita from 1800 to 2025 on a log scale, drawn dotted where it comes from the Maddison Project, dashed from the Penn World Table between 1950 and 1959 and solid from the World Bank from 1960, with the 1950 and 1960 seams marked; and the 16 seams of eight economies ordered by how much their overlap ratio drifts, four of them above the 2 percent flag and three with a change of units
Mexico’s real GDP per capita from the World Bank, extended back to 1950 with the Penn World Table and to 1800 with the Maddison Project. Of the 16 seams for eight economies, four drift by more than 2% across their overlap and are flagged; three carry a change of units. · Code

New in 4.2

Three panels: propensity score densities for treated and control groups with trimming bounds at 0.01 and 0.99; treatment effect estimates comparing naive OLS (13.72, upward biased) with DML-IRM ATE (10.89), ATT (11.13), and DML-IV (12.20, covering true effect 12.50); and first-stage effective F-statistic of 268 well above the Montiel Olea–Pflueger critical value of 23.1
Common support overlap trimming, unbiased treatment effects for DML-IRM (10.89) and DML-IV (12.20 vs true 12.50) compared to naive OLS (13.72), and effective F of 268. · Notebook 60
Three panels: dynamic common factors for Mexico and Brazil estimated from real-time vintage panels; horizontal waterfall chart of Bańbura–Modugno news attribution decomposing nowcast revisions by economic indicator; and calibrated predictive density fan chart with a Berkowitz uniformity test p-value of 0.96
Dynamic factor trajectories, exact Bańbura–Modugno news decomposition by indicator release, and calibrated predictive density fan chart (PIT p = 0.96). · Notebook 59

New in 4.0

Three panels: saving by assets for high and low income, positive for high earners and negative for low earners down to the borrowing limit; the stationary wealth density, with a spike at zero assets for low earners and a hump for high earners; and household capital supply rising and firm demand falling with the interest rate, crossing at r* = 1.89 percent
Saving and the stationary wealth density of a continuous-time household at r = 3%, and the Aiyagari equilibrium where capital supply meets demand at r* = 1.89%. · Notebook 56
Three panels over 15 quarters comparing OccBin with both constraints against the unconstrained linear model: the policy rate stays at zero for three quarters while the linear rate goes negative; borrowing is held at its limit on impact while the linear path overshoots it; output falls 14.6 percent instead of 9.4 percent. Grey shading and hatching mark the quarters when each constraint binds
A demand and credit shock with a zero lower bound and a borrowing limit: both bind on impact, the rate stays at zero for three quarters, and output falls 14.6% instead of 9.4% in the linear model. · Notebook 57
Left: estimates with 95 percent intervals for DML with Lasso (1.70) and ridge (1.66) learners and OLS on all 30 controls (1.69), all covering the true effect of 1.75, and a naive Lasso estimate of 1.58 with no valid interval. Right: cross-fitted outcome residuals plotted against treatment residuals, with the DML slope of 1.70
Double/debiased ML covers the true effect of 1.75 with its 95% interval, while a Lasso that also penalizes the treatment shrinks it to 1.58. With 30 controls and 500 observations, OLS is still valid too. · Notebook 57
Left: a triangle of simulated GDP growth rates by reference quarter and snapshot, darker where a figure was later revised more. Middle: first-release growth swinging between 0.33 and 0.78 percent around a latest-snapshot path near 0.6 percent. Right: revisions plotted against first releases, falling along a line with slope −0.89
On the notebook’s simulated panel: a revision triangle for GDP growth, first releases against the latest snapshot, and a Mankiw–Shapiro test that rejects both news and noise. · Notebook 58
Left: an impulse response estimated with a strong instrument, where the 90 percent Wald band and Anderson–Rubin set nearly coincide. Middle: with a weak instrument, finite Wald bands inside an Anderson–Rubin set that is the whole real line at 11 of 12 horizons. Right: effective F statistics near 73 for the strong instrument and below 1 for the weak one, against a critical value of 23.1
With a strong instrument, Wald bands and Anderson–Rubin sets agree. With a weak one, the Wald band still looks informative while the Anderson–Rubin set covers the whole line. · Code

New in 3.3

Two panels: a savings policy with a borrowing limit, and a zoom on the kink where Chebyshev interpolants oscillate while the finite-element solution follows the exact policy
Chebyshev interpolants ring around the kink; the finite-element solution sits on the exact policy. · Notebook 51
Three panels after a 5 percent TFP shock: the real interest rate jumps and then undershoots, household capital supply and firm demand move together, and the wealth Gini falls and then recovers
A +5% TFP shock in an Aiyagari economy: the real rate, capital-market clearing and wealth inequality along the transition. · Notebook 52
Left: sensitivities of the savings policy to four parameters across capital. Right: contours of the GMM loss with the BFGS path converging to the true parameters
Left: how the savings policy responds to each parameter. Right: the estimator’s path on the loss surface. · Notebook 53
Left: capital converging from 60 percent of its steady state. Right: the learned consumption policy lying below cash on hand and passing through the steady state
A ten-country growth model: convergence from a capital shortage, and one slice of the learned policy. · Notebook 54

Impulse responses & identification

Narrative sign restrictions around the October 1979 Volcker episode (Antolín-Díaz & Rubio-Ramírez, 2018), on a synthetic DGP
Narrative sign restrictions around the October 1979 Volcker episode (Antolín-Díaz & Rubio-Ramírez, 2018), on a synthetic DGP · Code
Lag-augmented vs Newey–West local projections (Plagborg-Møller & Wolf)
Lag-augmented vs Newey–West local projections (Plagborg-Møller & Wolf) · Code
A local projection on live FRED data
A local projection on live FRED data · Code
Cholesky vs non-Gaussian SVAR impulse responses
Cholesky vs non-Gaussian SVAR impulse responses · Code
Sign-restricted SVAR: standard vs Giacomini–Kitagawa robust bands
Sign-restricted SVAR: standard vs Giacomini–Kitagawa robust bands · Code
Proxy SVAR on a controlled synthetic DGP, Mertens–Ravn style

Proxy SVAR on a controlled synthetic DGP, Mertens–Ravn style · Code

Replications on real data

Galí (1999): hours after a technology shock, in differences and in levels
Galí (1999): hours after a technology shock, in differences and in levels · Code
Ramey & Zubairy (2018): state-dependent spending multipliers and weak-IV diagnostics
Ramey & Zubairy (2018): state-dependent spending multipliers and weak-IV diagnostics · Code
Galí (1999) and Mertens & Ravn (2013) from the built-in datasets
Galí (1999) and Mertens & Ravn (2013) from the built-in datasets · Code
Kilian (2009): oil-market shocks and the historical decomposition of the real oil price

Kilian (2009): oil-market shocks and the historical decomposition of the real oil price · Code

Structural models

Sequence-space HANK: wealth distribution, MPCs, responses to a rate hike and the Jacobian
Sequence-space HANK: wealth distribution, MPCs, responses to a rate hike and the Jacobian · Code
A Dynare-style HANK bridge: Fake News Jacobian and a −25 bp monetary shock
A Dynare-style HANK bridge: Fake News Jacobian and a −25 bp monetary shock · Code
Bayesian DSGE with NUTS: traces and posteriors
Bayesian DSGE with NUTS: traces and posteriors · Code
An RBC model simulated from its Klein QZ policy function
An RBC model simulated from its Klein QZ policy function · Code
Posterior distributions from the NUTS sampler
Posterior distributions from the NUTS sampler · Code
DICE climate-macro: temperature, emissions, damages and the social cost of carbon
DICE climate-macro: temperature, emissions, damages and the social cost of carbon · Code

Time variation, regimes & volatility

TVP-VAR with stochastic volatility (KSC mixture sampler)
TVP-VAR with stochastic volatility (KSC mixture sampler) · Code
Markov-switching VAR recession dating
Markov-switching VAR recession dating · Code
DCC time-varying correlation recovering a regime shift
DCC time-varying correlation recovering a regime shift · Code
GARCH-MIDAS: short-run volatility and a secular macro component
GARCH-MIDAS: short-run volatility and a secular macro component · Code
BVAR fan charts from the Gibbs sampler
BVAR fan charts from the Gibbs sampler · Code
HAR model of realized volatility

HAR model of realized volatility · Code

Data, nowcasting & text

Nowcasting GDP: latent factors, loadings and the news decomposition
Nowcasting GDP: latent factors, loadings and the news decomposition · Code
A hawkishness index from FOMC communication and its effect on rates and inflation
A hawkishness index from FOMC communication and its effect on rates and inflation · Code
Narrative tax shocks across G7 countries
Narrative tax shocks across G7 countries · Code
Real-time data: how revisions change a series and its forecast
Real-time data: how revisions change a series and its forecast · Code
Penalized forecasting with the elastic net and adaptive lasso
Penalized forecasting with the elastic net and adaptive lasso · Code
Wavelet variance decomposition across cycle lengths

Wavelet variance decomposition across cycle lengths · Code

Run them yourself

pip install puremacro
python -m puremacro.examples.gali_1999_hours

Most examples need no data and no API key. Browse all 92 scripts on GitHub.