PUBLIC DIGITAL PAYMENT RAILS AND LOCAL BUSINESS ADOPTION: EVIDENCE FROM BRAZIL'S PIX - replication package

# Brazil Pix Paper Replication Package This package reproduces the empirical tables and figures for the Brazil Pix paper. ## Repository Structure - `code/`: Python replication scripts. - `data/cleaned/`: frozen processed data for the cleaned-data replication path. - `docs/`: codebook, raw-file upload list, source notes and table/figure crosswalk. - `reference_outputs/`: reference CSV tables, QA outputs and one-format figure files. - `work/`: scratch directory created during replication runs. ## Environment Recommended setup: ```bash python -m venv .venv source .venv/bin/activate pip install -r requirements.txt ``` Conda alternative: ```bash conda env create -f environment.yml conda activate brazil-pix-replication ``` ## Path 1: Reproduce From Cleaned Data ```bash python code/run_cleaned.py --output outputs_from_cleaned && python code/check_outputs.py --outputs outputs_from_cleaned ``` This path uses the frozen processed data in `data/cleaned/` and reproduces the tables, figures, QA files and output checksums. Expected runtime on a modern laptop is roughly 30-60 minutes because the fixed-effect regressions and permutation checks are recomputed. ## Path 2: Reproduce From Uploaded Raw Public Data Upload every file listed in `docs/raw_files_to_upload.csv` to its `required_upload_path`, preserving filenames and folders. Then run: ```bash python code/run_raw.py --output outputs_from_raw && python code/check_outputs.py --outputs outputs_from_raw ``` The raw upload list contains 204 files totaling about 835 MB. The largest component is municipal COVID data. The raw path rebuilds the municipality-month panel, constructs exposures and diagnostics, then reproduces the same result outputs. ## Output Checks `code/check_outputs.py` verifies required output files and compares key CSV tables against `reference_outputs/` with a numerical tolerance. Each run also writes `sha256_outputs.csv` in the output directory. ## Notes The package includes public cleaned data needed for replication. Heavy public raw source files are not bundled; they are listed with sizes and checksums in `docs/raw_files_to_upload.csv`.