Code Directory

This directory contains all computational code for reproducing the results in "Welfare and Spending Effects of Consumption Stimulus Policies."

⚠️ Computational Time Warning

Full replication takes 4-5 DAYS on a high-end 2025 laptop. For quick validation (~1 hour), use the minimal reproduction: ../reproduce.sh --comp min

Directory Structure

`HA-Models/` - Heterogeneous Agent Models

Primary Entry Point: do_all.py

This is the main computational workflow controller. It orchestrates 5 sequential steps:

Step 1: Splurge Factor Estimation (~20 minutes)

File: Target_AggMPCX_LiquWealth/Estimation_BetaNablaSplurge.py
Purpose: Estimates the "splurge" factor (κ) using Norwegian lottery data
Paper Section: 3.1
Outputs:
- Figure 1 → Target_AggMPCX_LiquWealth/Figures/MPC_WealthQuartiles_Figure.pdf
- Table 1 → Target_AggMPCX_LiquWealth/images/MPC_WealthQuartiles_Table.tex
- Estimation results → Target_AggMPCX_LiquWealth/*.txt

Step 2: Discount Factor Distribution Estimation (~21 hours) ⚠️

Files:
- FromPandemicCode/EstimAggFiscalMAIN.py (main estimation)
- FromPandemicCode/CreateLPfig.py (creates Figure 2)
- FromPandemicCode/CreateIMPCfig.py (creates Figure 3a)
Purpose: Estimates distribution of discount factors (β) for three education groups
Paper Section: 3.3.3
Outputs:
- Figure 2 → FromPandemicCode/Figures/
- Model parameters → FromPandemicCode/Results/
Note: This is the most time-intensive step (~7 hours per education group × 3 groups)

Step 3: Robustness Check (~21 hours, OPTIONAL)

File: FromPandemicCode/EstimAggFiscalMAIN.py (with Splurge=0 parameter)
Purpose: Robustness analysis with no splurge factor
Paper Section: Online Appendix
Parameters: R=1.01, CRRA=2.0, ReplRate_w=0.7, ReplRate_wo=0.5, Splurge=0
Default: DISABLED in do_all.py to save time
Outputs: Table 8 → FromPandemicCode/Tables/Splurge0/welfare6_SplurgeComp.tex

Step 4: HANK Model (~1 hour)

Files:
- FromPandemicCode/HA-Fiscal-HANK-SAM.py (computes household Jacobians)
- FromPandemicCode/HA-Fiscal-HANK-SAM-to-python.py (runs HANK experiments)
Purpose: Robustness check with general equilibrium effects
Paper Section: 5
Outputs:
- Figure 5 → FromPandemicCode/Figures/
- HANK results → FromPandemicCode/Results/

Step 5: Policy Comparison (~65 hours) ⚠️

File: FromPandemicCode/AggFiscalMAIN.py
Purpose: Compares three fiscal policies (UI, checks, tax cuts)
Paper Section: 4
Outputs:
- Figure 4 → FromPandemicCode/Figures/
- Figure 6 → FromPandemicCode/Figures/ (uses Step 4 results)
- Table 6 (Multipliers) → FromPandemicCode/Tables/CRRA2/Multiplier.tex
- Table 7 (Welfare) → FromPandemicCode/Tables/CRRA2/welfare6.tex
Note: This is the longest-running step

`Empirical/` - Empirical Data Analysis

Contains Stata scripts for processing Survey of Consumer Finances (SCF) 2004 data.

Key File: make_liquid_wealth.do

Processes SCF 2004 data
Calculates liquid wealth statistics by wealth quartile
Generates empirical moments used in model calibration
Requires: Stata (tested with version MP/18.0)
Outputs: Statistics used in Table 2 (Panel B, lines 1-3), Table 4 (Panel B, line 1), Table 5 (Panels A & B, line 1)

Data Files:

rscfp2004.dta - SCF 2004 summary extract (Stata format)
rscfp2004.csv - SCF 2004 summary extract (CSV format)
ccbal_answer.dta - Processed credit card balance data (Stata format)
ccbal_answer.csv - Processed credit card balance data (CSV format)

Data Source: Federal Reserve Board - 2004 Survey of Consumer Finances

See main ../README.md section "Data Availability and Provenance Statements" for detailed data documentation.

Reproduction Workflows

Quick Test (~1 hour)

Test that environment is working without full computation:

Option 1: Minimal reproduction script

# From project root
./reproduce.sh --comp min

Option 2: Enable only fast steps

cd Code/HA-Models
# Edit do_all.py: set run_step_1=True, run_step_4=True, others=False
python do_all.py

Full Reproduction (4-5 days on a high-end 2025 laptop) ⚠️

Complete replication of all paper results:

# From project root - recommended
./reproduce.sh --comp full

Or directly:

cd Code/HA-Models
# Edit do_all.py: ensure desired steps are enabled
python do_all.py

Recommended configuration for full replication:

# In do_all.py
run_step_1 = True   # Required - provides splurge estimate
run_step_2 = True   # Required - baseline model
run_step_3 = False  # Optional - robustness (adds 21 hours)
run_step_4 = True   # Required - HANK model
run_step_5 = True   # Required - policy comparison

Step-by-Step Execution

For development, debugging, or selective replication:

Navigate to computational directory:
```
cd Code/HA-Models
```

Edit do_all.py to enable/disable steps:

run_step_1 = True   # Modify these flags
run_step_2 = False  # to control execution
# ... etc

Run the configured workflow:
```
python do_all.py
```
Check outputs in respective directories (see Output Locations below)

Example: Reproduce only Figure 1:

cd Code/HA-Models
# Edit do_all.py: run_step_1=True, all others=False
python do_all.py

Example: Reproduce policy comparison (requires steps 1-2 first):

cd Code/HA-Models  
# Edit do_all.py: run_step_1=True, run_step_2=True, run_step_5=True
python do_all.py

Output Locations

All outputs are generated in subdirectories of Code/HA-Models/:

Figures

Figure 1: Target_AggMPCX_LiquWealth/Figures/MPC_WealthQuartiles_Figure.pdf
Figure 2: FromPandemicCode/Figures/LorenzPtsAll.pdf
Figure 3:
- Panel (a): FromPandemicCode/Figures/iMPC_CRRA2.pdf
- Panel (b): FromPandemicCode/Figures/UIexitcon.pdf
Figure 4: FromPandemicCode/Figures/ (6 panels)
- MPCall.pdf, CAll.pdf, YAll.pdf (with matching _WGGE versions)
Figure 5: FromPandemicCode/Figures/ (6 panels)
- HANK-SAM comparison figures
Figure 6: FromPandemicCode/Figures/MPCall_Combined.pdf

Tables

Table 1: Target_AggMPCX_LiquWealth/images/MPC_WealthQuartiles_Table.tex
Table 2-3: Not generated by code (parameter summaries)
Table 4-5: Partial generation, see FromPandemicCode/Results/AllResults_CRRA_2.0_R_1.01.txt
Table 6: FromPandemicCode/Tables/CRRA2/Multiplier.tex
Table 7: FromPandemicCode/Tables/CRRA2/welfare6.tex
Table 8: FromPandemicCode/Tables/Splurge0/welfare6_SplurgeComp.tex

Results Files

All numerical results: FromPandemicCode/Results/AllResults_CRRA_2.0_R_1.01.txt
- Contains values used in Tables 4-5
- Written by EstimAggFiscalMAIN.py (lines 1105, 1111)
Estimation results: Target_AggMPCX_LiquWealth/*.txt

Integration with Paper

The LaTeX compilation system automatically integrates outputs from this code directory:

Figures

Python scripts generate PDFs in Code/HA-Models/FromPandemicCode/Figures/
These are symlinked to ../Figures/ directory
LaTeX document references them via \FigsDir macro
No manual copying needed

Tables

Python scripts generate .tex files with table content
LaTeX document includes them directly via \input{}
Tables are automatically updated when code runs
No manual copying needed

Build Integration

# This workflow keeps everything synchronized:
./reproduce.sh --comp full   # Generate computational results
./reproduce.sh --docs       # Compile paper (uses generated results)

The build system ensures figures and tables are always up-to-date.

Computational Requirements

Hardware

RAM: 8GB minimum, 32GB recommended
CPU: Multi-core processor beneficial (estimation can parallelize)
Storage: ~5GB for all intermediate results and outputs
Time: 4-5 days on a high-end 2025 laptop for complete replication

Benchmarking Reference

On Windows 11 laptop with 32GB RAM and AMD Ryzen 9 5900HS (3.30 GHz):

Step 1: 20 minutes
Step 2: 21 hours (~7 hours × 3 education groups)
Step 3: ~21 hours (similar to Step 2, optional)
Step 4: 1 hour
Step 5: 65 hours
Total: ~107 hours (4.5 days) with Step 3, ~86 hours (3.6 days) without

Your timing may vary based on hardware.

Software Dependencies

Python: 3.11.7 (other versions may work)
Core packages:
- econ-ark==0.14.1 (from conda-forge)
- numpy==1.26.4
- matplotlib==3.8.0
- scipy==1.11.4
- pandas==2.1.4
- numba==0.59.0
- sequence-jacobian (from pip)
Optional: Stata (for empirical analysis only)

Complete dependency list: See ../binder/environment.yml

Environment Setup

# Option 1: Using conda (recommended)
uv sync --all-groups  # or: conda env create -f ../environment.yml
conda activate hafiscal

# Option 2: Using pip
pip install econ-ark==0.14.1 numpy pandas matplotlib scipy numba
pip install sequence-jacobian

Troubleshooting

Installation Issues

Error: No module named 'HARK'

# Solution: Install econ-ark package
pip install econ-ark==0.14.1
# or
conda install -c conda-forge econ-ark=0.14.1

Error: No module named 'sequence_jacobian'

# Solution: Install via pip (not available on conda)
pip install sequence-jacobian

Error: ModuleNotFoundError: No module named 'numba'

# Solution:
pip install numba==0.59.0
# or
conda install -c conda-forge numba=0.59.0

Execution Issues

Computation taking too long / want to test first

# Solution 1: Use minimal reproduction
./reproduce.sh --comp min  # From project root

# Solution 2: Enable only fast steps
cd Code/HA-Models
# Edit do_all.py: run_step_1=True, run_step_4=True, others=False
python do_all.py

Results differ slightly from paper

Expected: Small numerical differences possible
Cause: Environment variations (OS, package versions, CPU)
Note: Optimization routines should be deterministic but may vary slightly
Acceptable: Differences in 3rd-4th decimal places
Investigate: Large differences (>1% in key results)

Script fails partway through

# Solution: Steps are mostly independent after Step 1-2
# You can restart from a later step by editing do_all.py

# Example: Step 2 completed, want to skip to Step 5
# In do_all.py:
run_step_1 = False  # Already done
run_step_2 = False  # Already done
run_step_3 = False  # Skip optional
run_step_4 = False  # Skip if not needed
run_step_5 = True   # Run this

Permission errors writing files

# Check write permissions in output directories
ls -ld FromPandemicCode/Figures/
ls -ld FromPandemicCode/Tables/
# Should show write permissions (drwxr-xr-x or similar)

Output Issues

Figures not appearing in paper

# Check if figures were generated:
ls -lh FromPandemicCode/Figures/*.pdf
ls -lh Target_AggMPCX_LiquWealth/Figures/*.pdf

# Check if symlinks exist:
ls -lh ../Figures/*.pdf

# Regenerate paper:
cd ..
./reproduce.sh --docs

Tables not updating in paper

# Tables are .tex files directly included
# Check if generated:
ls -lh FromPandemicCode/Tables/CRRA2/*.tex

# Regenerate paper:
cd ..
./reproduce.sh --docs

Development Notes

Modifying the Code

Adding a new step:

Create new Python script in appropriate subdirectory

Add step in do_all.py:

run_step_6 = True
if run_step_6:
    print('Step 6: Your new analysis\n')
    os.chdir('YourSubdir')
    os.system("python " + "your_script.py")
    os.chdir('../')
    print('Concluded Step 6.\n\n')

Update this README with step description
Update paper if adding new results

Modifying parameters:

Main parameters in FromPandemicCode/EstimParameters.py
Model parameters in FromPandemicCode/Parameters.py
Splurge factor (κ) in FromPandemicCode/Parameters.py (ADelasticity)

Testing Changes

# Quick test with Step 1 only
cd Code/HA-Models
# Edit do_all.py: run_step_1=True, others=False
python do_all.py

# Verify outputs
ls -lh Target_AggMPCX_LiquWealth/Figures/
ls -lh Target_AggMPCX_LiquWealth/*.txt

Code Organization

Key files:

do_all.py - Main orchestrator
reproduce_min.py - Minimal reproduction script
EstimParameters.py - Estimated parameters
Parameters.py - Model parameters
Output_Results.py - Results formatting and table generation
Welfare.py - Welfare calculations

Directory structure:

HA-Models/
├── do_all.py                      # Main entry point
├── reproduce_min.py               # Minimal reproduction
├── Target_AggMPCX_LiquWealth/    # Step 1: Splurge estimation
│   ├── Estimation_BetaNablaSplurge.py
│   ├── Figures/                  # Figure 1 output
│   └── images/                   # Table 1 output
└── FromPandemicCode/             # Steps 2-5: Main analysis
    ├── EstimAggFiscalMAIN.py     # Step 2: Estimation
    ├── CreateLPfig.py            # Figure 2
    ├── CreateIMPCfig.py          # Figure 3a
    ├── EvalConsDropUponUILeave.py # Figure 3b
    ├── AggFiscalMAIN.py          # Step 5: Policy comparison
    ├── HA-Fiscal-HANK-SAM.py     # Step 4: HANK household
    ├── HA-Fiscal-HANK-SAM-to-python.py  # Step 4: HANK experiments
    ├── Figures/                  # Figures 2-6 output
    ├── Tables/                   # Tables 6-8 output
    └── Results/                  # Numerical results

Additional Resources

Main README: ../README.md - Setup and overview
AI Documentation: ../README_IF_YOU_ARE_AN_AI/ - AI-optimized guides
- Especially: 030_COMPUTATIONAL_WORKFLOWS.md - Detailed workflow guide
Reproduction Scripts: ../reproduce/ - Script documentation
Paper: ../HAFiscal.pdf - Complete paper with results
Slides: ../HAFiscal-Slides.pdf - Presentation version

Citation

If you use this code, please cite:

Carroll, Christopher D., Edmund Crawley, Weifeng Dai, Ivan Frankovic, and Hakon Tretvoll (2025). "Welfare and Spending Effects of Consumption Stimulus Policies."

Support

For issues or questions:

Check troubleshooting sections above
Review main README troubleshooting: ../README.md
Check AI troubleshooting guide: ../README_IF_YOU_ARE_AN_AI/080_TROUBLESHOOTING_FOR_AI_SYSTEMS.md
Verify environment: ../reproduce/reproduce_environment_comp_uv.sh (or activate: source ../.venv/bin/activate)
Open an issue on the project repository (if applicable)

Quick Reference Command Summary:

# From project root:
./reproduce.sh --comp min       # Test (~1 hour)
./reproduce.sh --comp full      # Full replication (4-5 days on a high-end 2025 laptop)
./reproduce.sh --comp max       # Maximum replication with robustness (~5 days on a high-end 2025 laptop)
./reproduce.sh --docs           # Compile paper

# From Code/HA-Models/:
python do_all.py               # Run configured steps
python reproduce_min.py        # Minimal test

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

Code Directory

⚠️ Computational Time Warning

Directory Structure

HA-Models/ - Heterogeneous Agent Models

Step 1: Splurge Factor Estimation (~20 minutes)

Step 2: Discount Factor Distribution Estimation (~21 hours) ⚠️

Step 3: Robustness Check (~21 hours, OPTIONAL)

Step 4: HANK Model (~1 hour)

Step 5: Policy Comparison (~65 hours) ⚠️

Empirical/ - Empirical Data Analysis

Reproduction Workflows

Quick Test (~1 hour)

Full Reproduction (4-5 days on a high-end 2025 laptop) ⚠️

Step-by-Step Execution

Output Locations

Figures

Tables

Results Files

Integration with Paper

Figures

Tables

Build Integration

Computational Requirements

Hardware

Benchmarking Reference

Software Dependencies

Environment Setup

Troubleshooting

Installation Issues

Execution Issues

Output Issues

Development Notes

Modifying the Code

Testing Changes

Code Organization

Additional Resources

Citation

Support

`HA-Models/` - Heterogeneous Agent Models

`Empirical/` - Empirical Data Analysis