Rerun the macro stage of a tax-policy paper
This exercise uses the author replication archive for DeBacker, Evans and Phillips, Integrating Microsimulation Models of Tax Policy into a DGE Macroeconomic Model. The archive includes its May 2018 manuscript and public fitted tax-function/calibration inputs. The journal article appeared in 2019; agreement with its final numerical tables has not been checked.
We freshly solved the baseline and counterfactual macro steady states using the authors’ unchanged OG-USA solver and supplied fitted parameters. Tax-Calculator microsimulation/PUF estimation and transition paths were not rerun. Tax-Calculator 0.19.0 is installed because the legacy macro package imports it; that import does not count as micro-estimation execution.
sh get-source.sh
micromamba create -y -p ./legacy-env -f environment.yml
./legacy-env/bin/python -m pip install -r requirements-legacy.txt
./legacy-env/bin/python -m pip install --no-deps ./taxcalc-source
MPLBACKEND=Agg OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./legacy-env/bin/python run_fitted_pair.py
The harness reads the pinned archive’s model_params.pkl for _DEP_Age, invokes SS.create_steady_state_parameters and SS.run_SS, and stores fresh native results under output/ and pair-output/. It does not read PUF records or invoke get_micro_data. The source’s OUTPUT_BASELINE represents TCJA law and OUTPUT_REFORM represents the 2017-law counterfactual; the ratio is oriented accordingly.
Before the paired run, we selected Table 7/8 of the May 2018 manuscript, preferred age-and-income-source specification, steady-state GDP change +0.07%, with absolute tolerance 0.005 percentage points (half its printed 0.01-point unit). Actual change: +0.069545535143%, difference 0.000454464857 percentage points, which passes that rounding comparison. The six baseline aggregate outputs also match the author’s archived steady state within 3e-15; the declared regression tolerance was rtol=1e-6, atol=1e-8.
This run also exposes an economic-feasibility problem: government spending is −0.01294664762 in the TCJA steady state and −0.01051179371 in the counterfactual, in model units. The author solver warns that the debt closure requires infeasible negative spending. Thus the numerical table match is useful reproduction evidence for a bounded implementation stage; it does not establish a feasible policy prediction. Native resource residuals are about 5.3e-14/1.6e-14 and maximum Euler residuals below 4e-13, but those residuals do not cure the negative-spending condition.
Historical dependency recovery required Python 3.6 and Dask 0.17.5; newer Dask removes the solver’s get= API. Failed Python 3.7/Dask attempts are retained in the research record. The source and public fitted inputs are CC0; this original harness is MIT and guide CC BY 4.0. Do not add restricted IRS PUF data to the lab. Human scientific review remains pending.
Download the recorded files#
- results.json — execution evidence.
- comparison.csv — native result.
- pair-results.json — execution evidence.
- SS_vars.pkl — native result.
- SS_vars.pkl — native result.
- SS_vars.pkl — native result.
- guide.md — portable driver or environment/source record.
- get-source.sh — portable driver or environment/source record.
- run_fitted_ss.py — portable driver or environment/source record.
- run_fitted_pair.py — portable driver or environment/source record.
- environment.yml — portable driver or environment/source record.
- environment-explicit.txt — portable driver or environment/source record.
- requirements-legacy.txt — portable driver or environment/source record.
- requirements-tested.txt — portable driver or environment/source record.
- target.json — portable driver or environment/source record.
- pair-target.json — portable driver or environment/source record.
Complete run record · All candidate attempts. Human scientific review is pending.