Finance, economics, and policy: paper-to-code routes

On this page

All 60 tools · How to contribute

These are construction routes researched by agents. The execution line reports the bounded work actually performed; the comparison section defines what remains before claiming a published result. Human scientific review is pending.

quantlib#

QuantLib — Can a European call computed by a library and Monte Carlo agree with the original pricing equation?

Route: Paper model adaptation. Execution: Not run. Source confidence: high.

Paper for quantlib#

Fischer Black; Myron Scholes (1973). The Pricing of Options and Corporate Liabilities. 10.1086/260062.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Publisher metadata checked live; user-supplied restricted article previously inspected; no PDF redistribution.

Target in the paper: Assumptions p. 640; Eq. 13 p. 644. Eq. 13 is the target, using explicitly synthetic inputs rather than historical observations.

Code and inputs for quantlib#

Artifact or upstream source. Current official implementation and numerical-engine comparison, not Black/Scholes author code. Existing OSRI Python adaptation implements the equation.

Source inspected: yes. Ref: master. Commit: 9ba30347b4ea179a5f4daec446cd221d21b8721f.

Files: Examples/EquityOption/EquityOption.cpp

Terms: BSD-3-Clause

Environment and compute: Existing lab Python 3.12 environment: QuantLib 1.44 and pinned numerical dependencies; alternatively compile the upstream C++ example in a separate CMake build. CPU; seconds to minutes for the teaching case.

Inputs: Existing OSRI adaptation uses synthetic spot=100, strike=100, rate=0.05, volatility=0.2, maturity=1 year, 50000 paths and seed 42. The C++ upstream example has different put-option inputs.

Data access: Synthetic inputs; no market-data subscription.

Construct and compare quantlib#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. # Existing lab command, not a new execution in this research pass
  2. python simulation-lab/examples/option_monte_carlo.py --help
  3. # Follow the existing CPU walkthrough with its pinned requirements and explicit output directory. Do not mix the C++ example parameters with the Python target.

Expected outputs: Native options CSV, parameters JSON, Monte Carlo standard error and confidence interval.

Comparison with the paper: Compute Eq. 13 independently; compare QuantLib analytic value to double-precision formula and assess Monte Carlo error using its reported standard error. Existing run analytic value was 10.450583572185565; matching it verifies the synthetic equation example, not an empirical published table.

Execution result: Source inspection only; no new execution for this route.

Execution evidence: None recorded.

Blockers: No original 1973 source-code artifact; the existing example is a modern adaptation.

Next action: Retain current formula test, add a grid over maturity/moneyness/volatility and convergence versus path count before proposing a numerical-method study.

Gaps: Equation adaptation already executed in prior lab release; no new paper-result reproduction.

Sources for quantlib#

abides#

ABIDES — How does a large buy order alter a simulated order book and the subsequent price path?

Route: Paper model adaptation. Execution: Not run. Source confidence: high.

Paper for abides#

David Byrd; Maria Hybinette; Tucker Hybinette Balch (2019). ABIDES: Towards High-Fidelity Market Simulation for AI Research. arXiv:1904.12066v1; conference version DOI 10.1145/3384441.3395986.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Open HTML inspected; Crossref confirms 2020 conference ABIDES record. The detailed target here is explicitly the 2019 preprint.

Target in the paper: Section 6, Figures 6–7: greed-controlled impact event and aligned event-study paths; Section 5 Figure 4 uses historical IBM/MSFT trades.

Code and inputs for abides#

Artifact or upstream source. Repository explicitly linked by the preprint. Current impact.py uses a generated mean-reverting oracle, unlike the historical-data oracle described in the preprint.

Source inspected: yes. Ref: master. Commit: c4bf157678928934417aba6073eb0651aeaf6d15.

Files: config/impact.py; config/rmsc03.py; requirements.txt; README.md

Terms: BSD-3-Clause

Environment and compute: Isolated historical Python environment; inspect old NumPy/pandas/SciPy requirements before solving dependencies. Do not install into the modern CPU environment. CPU; many-agent event simulation; benchmark one session before scheduling seed batches.

Inputs: For a synthetic adaptation, config/impact.py, same random seed in control and treatment, documented greed parameter. Exact paper replay additionally requires original historical ticks and configuration revision.

Data access: Synthetic oracle needs no ticks; original historical-trade days may require licensed tick access.

Construct and compare abides#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. # Proposed command from inspected CLI; clone and checkout the pinned artifact first
  2. python -u abides.py -c impact -s 1234 -g 0.3 -l impact_g03
  3. python -u abides.py -c impact -s 1234 -g 1.9 -l impact_g19
  4. # Repeat paired control/treatment experiments over independently selected seeds; inspect -i semantics (it disables impact).

Expected outputs: Exchange/order-book logs, aligned trade/midprice series, per-seed price-impact curves.

Comparison with the paper: Predeclare event time and signed price-change metric; compare treatment-control impact over time with Figure 6/7 after recovering historical configuration and inputs. Current synthetic-oracle run is an adaptation; paper curve values and tolerance are not yet extracted.

Execution result: Source inspection only; no new execution for this route.

Execution evidence: None recorded.

Blockers: Current config uses different background agents/oracle.; Historical paper configuration and tick data not recovered.; Historical environment not executed.

Next action: Recover the paper-era commit/configuration and lawful tick inputs; begin with paired synthetic impact experiments while explicitly labeling the adaptation.

Gaps: A runnable RMSC03 tutorial alone would not reproduce these paper figures.

Sources for abides#

finrl#

FinRL — Does an ensemble of PPO, A2C and DDPG improve risk-adjusted stock-trading performance over the paper’s baselines?

Route: Paper model adaptation. Execution: Not run. Source confidence: medium.

Paper for finrl#

Hongyang Yang; Xiao-Yang Liu; Shan Zhong; Anwar Walid (2020). Deep reinforcement learning for automated stock trading: an ensemble strategy. 10.1145/3383455.3422540.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Crossref title/authors/year checked; author university-hosted paper text available.

Target in the paper: Empirical ensemble-versus-individual-agent and DJIA/minimum-variance backtests; target is the paper performance comparison, with table/figure numbers and numeric tolerances still to extract from full text.

Code and inputs for finrl#

Artifact or upstream source. Author-maintained notebook explicitly named for the ICAIF 2020 experiment, but subsequently updated. Exact paper repository linked by a later arXiv record returns HTTP 404.

Source inspected: yes. Ref: master. Commit: 286fef6c828a02a2ccf807c05b292d8cb28f8e95.

Files: examples/FinRL_Ensemble_StockTrading_ICAIF_2020.ipynb

Terms: MIT; market-data reuse rights separate

Environment and compute: Isolated pinned FinRL/Stable-Baselines3/PyTorch/Jupyter environment. Notebook installs FinRL from mutable Git HEAD; replace with a tested immutable version before execution. CPU possible for small diagnostics; full rolling ensemble training likely hours, GPU optional.

Inputs: DJIA constituents, adjusted OHLCV and indicators, transaction costs, split dates, turbulence threshold, training budgets and seeds. Inspected notebook currently uses 2010-01-01..2021-10-01 train and 2021-10-01..2023-03-01 test, so it is not the original 2020 evaluation.

Data access: Notebook calls YahooDownloader; freeze exact historical inputs and confirm provider terms and adjusted-price treatment. No live trading needed.

Construct and compare finrl#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. # Proposed preparation: clone FinRL-Meta and checkout the recorded commit
  2. # Inspect every notebook cell; pin FinRL dependency and archive lawful input data before training
  3. jupyter nbconvert --execute --to notebook --ExecutePreprocessor.timeout=-1 examples/FinRL_Ensemble_StockTrading_ICAIF_2020.ipynb --output ensemble.executed.ipynb

Expected outputs: Executed notebook, models, account-value CSVs, ensemble-selection trace and backtest-statistic tables.

Comparison with the paper: Recreate the original evaluation universe and splits; compare annual return, drawdown and Sharpe statistics over repeated seeds. Predeclare tolerance/distributional comparison. Current notebook dates and dependencies require changes and must be described as an adaptation until matched.

Execution result: Source inspection only; no new execution for this route.

Execution evidence: None recorded.

Blockers: Exact original repository URL currently 404.; Notebook data splits are newer than paper.; Mutable dependency installs and random seeds not fully pinned.

Next action: Recover archived paper-era notebook/data, then run a frozen backtest with repeated seeds; use author notebook as a construction scaffold meanwhile.

Gaps: Do not describe contemporary notebook returns as reproduced ICAIF-2020 results.

Sources for finrl#

engine#

Open Source Risk Engine — How does central clearing change systemic exposure and counterparty credit risk?

Route: Paper result candidate. Execution: Attempted; blocked. Source confidence: high.

Paper for engine#

Nikolai Nowaczyk; Sharyn O’Halloran (2024). Computing the impact of central clearing on systemic risk. 10.3389/frai.2024.1138611.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Open full HTML inspected, including methods, validation, results and source footnotes.

Target in the paper: Sections 7.3–7.6 ORE risk calculations and notebook visualization; Section 8.4 Figure 15: 50% of exposure concentrated in CCPs; Figure 14 relative CVA impacts. Footnotes 13, 16–17 link code and ORE v1.8.4.0 inputs.

Code and inputs for engine#

Artifact or upstream source. Paper-linked author notebook and precomputed simulation outputs. Repository tree contains dashboard assets, not the upstream generator that invokes ORE. ORE v1.8.4.0 is separately referenced by the article.

Source inspected: yes. Ref: master. Commit: 677f5e7b639d4b8de18767a0f2b8db4844ceaf1a.

Files: dashboards/central_clearing/simulation_cva/central_clearing_sim_cva.ipynb; dashboards/central_clearing/simulation_cva/environment.yml; dashboards/central_clearing/simulation_cva/df_risk_systemic.csv; dashboards/central_clearing/simulation_cva/df_risk_cp.csv; dashboards/central_clearing/simulation_cva/risk_graphs_nx.json

Terms: No repository license found; inspect permission before redistributing author assets. ORE itself BSD-3-Clause.

Environment and compute: Historical dashboard uses Windows-specific Conda Python 3.7.5, pandas 0.23.4, networkx 2.2; porting to Linux needs a new recorded environment. Fresh simulation additionally needs ORE v1.8.4.0 plus missing generator. Replot small enough for CPU; full ORE portfolio/network batch must be profiled after generator recovery.

Inputs: Bundled risk CSVs/graphs for a figure replot; original generated portfolios, market/calibration files, random graph settings and ORE inputs needed for a fresh run.

Data access: Public dashboard includes synthetic-output tables; reuse license unspecified. Bundled ORE market inputs referenced in paper.

Construct and compare engine#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. # Proposed replot route, after constructing a compatible isolated environment
  2. cd dashboards/central_clearing/simulation_cva
  3. jupyter notebook central_clearing_sim_cva.ipynb
  4. # Fresh simulation cannot be specified completely from this repository: retrieve portfolio-generation and ORE orchestration code first.

Expected outputs: Dashboard exposure/CVA plots and parsed risk tables; fresh ORE exposure CSVs require missing orchestration.

Comparison with the paper: No new ORE numerical outputs. Prior saved-output shares 50.339%–50.927% still differ from exact paper 50% statement and remain unadjudicated.

Execution result: Fresh simulation cannot run from retrieved author artifact: no portfolio generator or ORE orchestration in full source/history. Prior saved-output analysis is retained separately and not counted as a fresh run.

Execution evidence: Author-code attempt; Run record

Blockers: Author portfolio generator, ORE orchestration/configurations and original seed specification missing.; Paper-linked saved-output version correspondence/discrepancy unresolved.; Dashboard repository has no license found.

Next action: Retrieve author generator/configuration artifact and versioned ORE v1.8.4.0 inputs; then declare a numerical target and tolerance before running.

Gaps: Author portfolio generator, ORE orchestration/configurations and original seed specification missing.; Paper-linked saved-output version correspondence/discrepancy unresolved.; Dashboard repository has no license found.

Sources for engine#

hark#

HARK / Econ-ARK — Can the author-associated code regenerate Carroll’s buffer-stock saving figures and tables?

Route: Paper result candidate. Execution: Example executed. Source confidence: high.

Paper for hark#

Christopher D. Carroll (1997). Buffer-Stock Saving and the Life Cycle/Permanent Income Hypothesis. 10.1162/003355397555109.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: User-supplied restricted full text previously inspected; live Crossref and Econ-ARK material record checked.

Target in the paper: Figures I, V, VI and VII and Tables I–II correspond to named Figure1a/1b, Figure5a/b/c, Figure6/7 and table1/2 output code. Original baseline pp. 5–8 already annotated by the lab.

Code and inputs for hark#

Artifact or upstream source. Econ-ARK identifies this as a replication by Carroll, Yusuf and Son. Source explicitly saves paper-associated figures/tables.

Source inspected: yes. Ref: master. Commit: 0a278360b77eeac1df1e91374c168b3091365fbf.

Files: reproduce.sh; Code/Python/BufferStock-LifeCycle.py; binder/environment.yml; Paper/Figures; Paper/Tables

Terms: Inspect repository license and HARK Apache-2.0 separately; restricted article excluded.

Environment and compute: Historical Econ-ARK notebook runtime; README documents econark/econ-ark-notebook:0.10.4; Python script metadata says 3.7.6. Modern HARK 0.17.2 has changed APIs. CPU; policy-function solutions and household lifecycle simulation, followed by plotting.

Inputs: Author calibration and lifecycle profiles from the pinned script; preserve shock distribution details, unemployment probability, discount factor and exact growth variants.

Data access: Code/input profiles publicly available; original PDF remains restricted and is not redistributed.

Construct and compare hark#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. git clone https://github.com/econ-ark/BufferStock-LifeCycle.git
  2. cd BufferStock-LifeCycle
  3. git checkout 0a278360b77eeac1df1e91374c168b3091365fbf
  4. # Prepare its historical environment; inspected reproduce.sh runs the Python script using IPython
  5. bash reproduce.sh

Expected outputs: Paper/Figures/Figure1a.png, Figure1b.png, Figure5a.png, Figure5b.png, Figure5c.png, Figure6.png, Figure7.png; Paper/Tables/table1.tex and table2.tex.

Comparison with the paper: Actual script execution restored; baseline target net wealth rounds to 0.31 rather than printed 0.32. Author implementation methods differ from original paper. No paper reproduction claim.

Execution result: Seven figure PNGs and two LaTeX tables generated from exact source. Strict original-paper baseline target comparison failed; SciPy non-progress warning retained.

Execution evidence: Author-code attempt; Run record

Blockers: Strict printed table comparison fails.; Author discretization/simulation and calibration conventions differ from original paper.; SciPy non-progress warning elsewhere in full script; convergence of all quantities not established.

Next action: Audit numerical settings/shock distribution and deterministic target residual; resolve mismatch before declaring paper result reproduced.

Gaps: Strict printed table comparison fails.; Author discretization/simulation and calibration conventions differ from original paper.; SciPy non-progress warning elsewhere in full script; convergence of all quantities not established.

Sources for hark#

quantecon#

QuantEcon.py — How does uninsured idiosyncratic earnings risk affect precautionary savings and equilibrium capital?

Route: Paper model adaptation. Execution: Not run. Source confidence: high.

Paper for quantecon#

S. Rao Aiyagari (1994). Uninsured Idiosyncratic Risk and Aggregate Saving. 10.2307/2118417.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Crossref identity checked; full original paper not read in this pass. Maintainers’ full model lecture/code inspected.

Target in the paper: QuantEcon lecture The Aiyagari Model, equilibrium implementation and capital-supply/demand plot; original article quantitative table target not yet inspected.

Code and inputs for quantecon#

Artifact or upstream source. Maintainer-authored teaching implementation of Aiyagari’s model, not the 1994 author program. The live lecture uses a two-state simplified income process.

Source inspected: yes. Ref: main. Commit: 007bbd592a4c1f43931ebb0eeb0e64ba36fa8a47.

Files: lectures/aiyagari.md

Terms: Lecture/book license and QuantEcon.py MIT separately; inspect before copying code.

Environment and compute: Pinned Python + quantecon, JAX/JAXLIB, SciPy and matplotlib; enable JAX 64-bit as the lecture does. CPU supported although lecture was built using GPU. CPU supported; first JAX compilation adds overhead.

Inputs: Lecture model: log utility, beta=0.96, 100 asset points, two productivity states [0.1,1.0], transition [[0.9,0.1],[0.1,0.9]], alpha=0.33, delta=0.05. These are teaching defaults, not asserted to match the paper calibration.

Data access: Synthetic model inputs, no household microdata needed.

Construct and compare quantecon#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. git clone https://github.com/QuantEcon/lecture-python.myst.git
  2. cd lecture-python.myst
  3. git checkout 007bbd592a4c1f43931ebb0eeb0e64ba36fa8a47
  4. # Proposed: execute lectures/aiyagari.md as a MyST/Jupyter notebook in its pinned book environment; preserve all cells and outputs.

Expected outputs: Savings policy, stationary distribution, capital-supply curve, demand curve and equilibrium interest rate/capital.

Comparison with the paper: For the teaching case verify stationarity, probability normalization, resource feasibility and capital-market residual under grid refinement. Recover the original paper’s discretization/calibration and named table cells before any 1994 numerical comparison.

Execution result: Source inspection only; no new execution for this route.

Execution evidence: None recorded.

Blockers: Original paper result/table and calibration not audited.; Book environment and JAX versions need a tested lock.

Next action: Execute the pinned lecture and export numerical arrays/residuals, then construct a separate paper-matched calibration and quantitative comparison.

Gaps: Do not label the simplified lecture as an original-paper table reproduction.

Sources for quantecon#

og-core#

OG-Core — How do alternative tax-function representations change predicted macroeconomic responses to tax policy?

Route: Paper model adaptation. Execution: Not run. Source confidence: medium.

Paper for og-core#

Jason DeBacker; Richard W. Evans; Kerk L. Phillips (2019). Integrating Microsimulation Models of Tax Policy into a DGE Macroeconomic Model. 10.1177/1091142118816744.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Publisher/Crossref metadata and author replication README inspected; full-text draft available from authors/conference.

Target in the paper: Tax function alternatives and baseline/reform general-equilibrium results; exact published table/figure numbers and cells still require final-paper inspection. Author repository provides TFI_tables_figures.py.

Code and inputs for og-core#

Artifact or upstream source. Current OG-Core is a modern framework connected to country models such as OG-USA. Exact 2019 paper code lives in rickecon/TaxFuncIntegr and uses historical OG-USA; switching to OG-Core is a port, not a same-code rerun.

Source inspected: yes. Ref: master. Commit: 7c31fbb5e3cb0fc49de5e99da444af9c7c0bb469.

Files: examples/run_ogcore_example.py; examples/example-diffs; ogcore/SS.py; ogcore/TPI.py

Terms: CC0-1.0

Environment and compute: Isolated current OG-Core Python environment; separate historical TaxFuncIntegr environment if reproducing the 2019 result. Note README points to run_examples but inspected snapshot actually uses examples/. CPU; upstream README warns full baseline/reform paths may exceed two hours.

Inputs: Current official example sets baseline corporate rate 0.21 and reform 0.35 with its own spending paths, not the exact historical paper experiment. Paper route requires preserved tax-function fits and historical demographics.

Data access: Current example ships synthetic/default calibration; paper fits available but PUF microdata are restricted.

Construct and compare og-core#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. git clone https://github.com/PSLmodels/OG-Core.git
  2. cd OG-Core
  3. git checkout 7c31fbb5e3cb0fc49de5e99da444af9c7c0bb469
  4. # Proposed environment install from the pinned pyproject, then run the actual inspected entry point
  5. python examples/run_ogcore_example.py

Expected outputs: examples/OG-Core-Example/BASELINE and REFORM outputs, model_params.pkl, SS/SS_vars.pkl, TPI/TPI_vars.pkl, example_output.csv and example_plots.

Comparison with the paper: First reproduce the current official example with market-clearing/Euler residuals and its own regression reference, after resolving outdated example-diffs names. Then port a named TaxFuncIntegr paper case and compare aggregate transition paths under the same calibration.

Execution result: Source inspection only; no new execution for this route.

Execution evidence: None recorded.

Blockers: Exact historical model uses OG-USA, not current OG-Core.; README path/output names and examples/example-diffs are stale relative to current script.; Paper-matched port and full output comparison not performed.

Next action: Freeze a compatible OG-Core release and execute its example; separately define a paper-calibration port using the author archive and explicit differences.

Gaps: Current OG-Core tutorial cannot be presented as the historical paper replication.

Sources for og-core#

tax-calculator#

Tax-Calculator — Can the paper’s micro tax-rate estimates and macro tax-function comparisons be regenerated?

Route: Paper result candidate. Execution: Example executed. Source confidence: high.

Paper for tax-calculator#

Jason DeBacker; Richard W. Evans; Kerk L. Phillips (2019). Integrating Microsimulation Models of Tax Policy into a DGE Macroeconomic Model. 10.1177/1091142118816744.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Publisher/Crossref metadata and author replication README inspected; full-text draft available from authors/conference.

Target in the paper: May 2018 author manuscript DEP_TaxFuncs.pdf, Tables 7 and 8, preferred specification steady-state GDP change +0.07%; final 2019 journal numerical tables not independently matched.

Code and inputs for tax-calculator#

Artifact or upstream source. Author replication code for the paper/draft; calls Tax-Calculator and stores fitted tax-function outputs and solved model outputs.

Source inspected: yes. Ref: master. Commit: 715cc76e3305c00dd64d79521c504bb388c6d87d.

Files: run_TaxFuncInt.py; TFI_tables_figures.py; environment.yml; 2017_law.json; OUTPUT_BASELINE; OUTPUT_REFORM

Terms: CC0-1.0 dedication; LICENSE.md inspected. Tax-Calculator itself also CC0.

Environment and compute: Historical Python 3.6 environment, taxcalc>=0.19.0 in a loose Conda environment.yml, historical OG-USA local package. Exact resolved historical dependency versions not preserved by this file. README estimates about 10 hours with 10+ processors or 100+ hours serial for full 12-case analysis.

Inputs: Original IRS PUF for fresh tax-function estimation; precomputed fitted-function pickle files provided as a partial open route. CPS data can replace PUF but changes the experiment.

Data access: IRS PUF not published; existing estimated functions are available. Never upload PUF/restricted tax records.

Construct and compare tax-calculator#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. git clone https://github.com/rickecon/TaxFuncIntegr.git
  2. cd TaxFuncIntegr
  3. git checkout 715cc76e3305c00dd64d79521c504bb388c6d87d
  4. # Historical environment must be reconstructed; README contains a pyton typo and outdated Python/ folder path. Inspect files before use.
  5. # Proposed: set run_micro=False to reuse published fitted functions; do not claim this reruns the microsimulation stage
  6. python run_TaxFuncInt.py
  7. python TFI_tables_figures.py

Expected outputs: Baseline/reform fitted tax functions, SS/TPI outputs, publication tables and figures from the plotting script.

Comparison with the paper: Numerical +0.069545535% matches May-2018 draft +0.07% at declared tolerance; six baseline outputs match archived author SS. Scope is fresh macro solve using supplied fitted inputs. Negative G in both cases is retained as failed economic feasibility, so neither this match nor small Euler residuals establish a valid policy prediction.

Execution result: Fresh baseline and counterfactual macro steady states solved; author-manuscript SS GDP rounding target matches. Economic-feasibility diagnostic fails because both closures require negative government spending. Tax-Calculator imported but its microsimulation/PUF estimation NOT executed.

Execution evidence: Author-code attempt; Run record

Blockers: Original restricted IRS PUF not available for fresh Tax-Calculator micro-estimation.; Government spending is negative in both solved macro steady states.; Final 2019 journal table identity not checked; selected comparison is the May 2018 author manuscript.; Full tax-function fitting and transition paths not rerun.

Next action: Resolve fiscal closure feasibility and verify final journal table/version; use lawful PUF access only if fresh microsimulation estimation is required.

Gaps: Original restricted IRS PUF not available for fresh Tax-Calculator micro-estimation.; Government spending is negative in both solved macro steady states.; Final 2019 journal table identity not checked; selected comparison is the May 2018 author manuscript.; Full tax-function fitting and transition paths not rerun.

Sources for tax-calculator#

openfisca-core#

OpenFisca — How does a targeted cash benefit create a work-incentive cliff compared with a child allowance?

Route: Unresolved. Execution: Not run. Source confidence: low.

Paper for openfisca-core#

Max Ghenis (2022). How does targeted cash assistance affect incentives to work?. CGO Research in Focus, 2022-06-17 (research report; no DOI found).

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Open full report and publisher byline/date inspected. This is a published policy research report, not a journal article.

Target in the paper: Cambridge-family net-income cliff and Massachusetts poverty simulations. Report states statewide overall poverty reduction 15%, child poverty 42%; these are model predictions, not empirical intervention effects.

Code and inputs for openfisca-core#

Artifact or upstream source. Exact author notebook linked in the report, but repository currently returns 404. No immutable snapshot recovered. The report links historical OpenFisca-US; modern PolicyEngine is not assumed interchangeable.

Source inspected: no. Ref: Not established.. Commit: Not established..

Files: analysis.ipynb (named by report; unavailable for inspection)

Terms: Unavailable; do not assume source/data redistribution terms.

Environment and compute: Unresolved historical OpenFisca-US/Core environment until the notebook and dependency pin are recovered. Current OpenFisca core alone cannot recreate a historical country model or policy report. Likely CPU microsimulation; size unknown until data recovered.

Inputs: Original report notebook, household specification, 2022 law snapshot, survey microdata/weights and benefit assumptions; all still need recovery and terms inspection.

Data access: Report public; original notebook/data currently unavailable. Generic OpenFisca country-template is a separate tutorial and does not supply this research artifact.

Construct and compare openfisca-core#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. # Recovery steps only; no working execution command is established
  2. # Retrieve analysis.ipynb from the author or an authenticated archive; verify it matches the report
  3. # Reconstruct the historical OpenFisca-US and Core versions, law period, data and weights
  4. # Only then execute the recovered notebook in an isolated environment and compare the reported targets.

Expected outputs: Expected construction outputs: disposable-income-versus-earnings curve, poverty changes and budget-neutral child-allowance comparison. Filenames unknown until code recovered.

Comparison with the paper: After recovery, compare the specific report family’s benefit cliff and the stated poverty percentage changes using its original weights and definitions. Predeclare rounding tolerance; do not use current laws/country packages as a replacement and call it reproduction.

Execution result: Source inspection only; no new execution for this route.

Execution evidence: None recorded.

Blockers: Exact reproduction repo returns HTTP 404.; Historical model versions, input data rights and weighting not recovered.

Next action: Request the author notebook/archive and shareable input specification, or select an alternative paper with a surviving archive before constructing a public walkthrough.

Gaps: This tool does not yet have an accessible inspected research-code route from this candidate.

Sources for openfisca-core#

dynare#

Dynare — Can a published Bayesian DSGE model’s shock responses and posterior estimates be regenerated?

Route: Paper result candidate. Execution: Example executed. Source confidence: high.

Paper for dynare#

Frank Smets; Rafael Wouters (2007). Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach. 10.1257/aer.97.3.586.

Publication type: See primary metadata record; not classified here.. Primary metadata. Metadata verification: confirmed. Full-text access: Official AEA article record and replication-package landing page inspected; code inspection uses a maintained third-party implementation.

Target in the paper: Paper’s Bayesian estimation, shock decomposition and impulse responses; specific printed table cells/figure tolerances still require full-text comparison. Official replication archive DOI 10.3886/E116269V1.

Code and inputs for dynare#

Artifact or upstream source. Maintained replication by Johannes Pfeifer with explicit Smets/Wouters attribution and corrections. Prefer the AEA-linked original archive as reference and document differences.

Source inspected: yes. Ref: master. Commit: 26c21f79e4c6a6b19b18e6330edc5b23489907a4.

Files: Smets_Wouters_2007/Smets_Wouters_2007.mod; Smets_Wouters_2007/Smets_Wouters_2007_45.mod; Smets_Wouters_2007/usmodel_data.mat; Smets_Wouters_2007/usmodel_mode.mat

Terms: GPL-3.0-or-later for this Dynare implementation per header; original data/archive terms separate.

Environment and compute: Dynare with GNU Octave for open runtime; choose supported Dynare release and inspect the model’s version notes. Host has no Octave/Dynare identified; no system installation authorized. Inspected default model uses mh_replic=0, mode_compute=0 and the supplied usmodel_mode, then shock_decomposition y; it does not run a full MCMC chain by default. CPU; deterministic mode/IRF check smaller than full Bayesian MCMC estimation.

Inputs: Bundled seven-series usmodel_data.mat and posterior-mode starting file; original sample, transformations, priors, estimation options and seed must match selected target.

Data access: Original AEA-linked replication package available through OpenICPSR; bundled data exist in third-party repository. Check archive download/access terms.

Construct and compare dynare#

The sequence below contains upstream commands and proposed setup instructions, as labeled. Only commands identified in the execution evidence were run here; inspect the pinned source and prerequisites before running.

  1. git clone https://github.com/JohannesPfeifer/DSGE_mod.git
  2. cd DSGE_mod
  3. git checkout 26c21f79e4c6a6b19b18e6330edc5b23489907a4
  4. cd Smets_Wouters_2007
  5. # Proposed Octave command after installing an isolated compatible Dynare runtime
  6. octave --no-gui --eval "dynare Smets_Wouters_2007.mod"

Expected outputs: Dynare model/results files, evaluation at supplied posterior mode and shock decomposition for y under default settings. Full posterior/MCMC diagnostics require explicitly changing mh_replic and estimation options; default nograph/nodiagnostic suppress these outputs.

Comparison with the paper: Exploratory decomposition sum residual is 7.1e-15; Laplace density -1844.0985587 under provided mode/settings. Original published numerical target/sample/mode not matched, and runtime/model warnings remain.

Execution result: Fresh maintained model evaluation and finite 40×9×230 shock decomposition completed. No original-paper quantitative comparison or posterior estimation claimed.

Execution evidence: Author-code attempt; Run record

Blockers: Old mode file cannot be verified against current model.; gaminv prior-bound calculation did not converge for some values.; Negative array dimensions converted to zero in shock-decomposition setup.; 33 preprocessor warnings from unassigned steady-state block variables.; Official original replication archive requires account to download.; Maintained model uses supplied old mode and 230 observations; correspondence to selected paper target unverified.

Next action: Match original archive/sample/mode and resolve prior-bound/mode compatibility warnings, then define a named paper comparison before full posterior estimation.

Gaps: Old mode file cannot be verified against current model.; gaminv prior-bound calculation did not converge for some values.; Negative array dimensions converted to zero in shock-decomposition setup.; 33 preprocessor warnings from unassigned steady-state block variables.; Official original replication archive requires account to download.; Maintained model uses supplied old mode and 230 observations; correspondence to selected paper target unverified.

Sources for dynare#

View source on GitHub