Uninsured Deposits and Funding-Cost Pass-Through - replication package

# Replication Package

This package reproduces the tables, figures, and memo for the JFS submitted paper. It contains a frozen cleaned bank-quarter panel and the Python code needed to reproduce the results from either cleaned data or uploaded public raw files.

The package intentionally excludes manuscripts, drafts, archived project material, cache files, and unrelated project files.

## Repository Structure

- `code/`: replication runners and analysis scripts.

- `data/cleaned/`: frozen cleaned analysis data for the short replication path.

- `docs/`: raw-file upload manifest, data-source notes, codebook, and output crosswalk.

- `reference_outputs/`: archived CSV tables, PDF figures, and memo used for output checks.

## Computational Environment

Use Python 3.11 or newer. Install dependencies with:

```bash

python3 -m venv .venv

source .venv/bin/activate

python3 -m pip install -r requirements.txt

```

An equivalent conda environment is specified in `environment.yml`.

## Path 1: Reproduce from Cleaned Data

One-line command:

```bash

python3 code/run_cleaned.py --output outputs_from_cleaned && python3 code/check_outputs.py --outputs outputs_from_cleaned

```

This path uses `data/cleaned/bank_quarter_panel_corrected_call_report.parquet` and reproduces all tables, figures, and the corrected measurement memo. In the package verification run, this path completed in about one minute; permutation checks are the slowest block.

## Path 2: Rebuild from Public Raw Data

Before running this path, upload the public raw files listed in `docs/raw_files_to_upload.csv` under the indicated relative paths. The required public raw-file set contains 123 files and is approximately 306 MB.

One-line command:

```bash

python3 code/run_raw.py --output outputs_from_raw && python3 code/check_outputs.py --outputs outputs_from_raw

```

The raw path rebuilds intermediate processed files from uploaded FDIC, FFIEC, and FRED files, then runs the same final corrected measurement pipeline.

## Outputs

Generated outputs are written under the selected output directory:

- `tables/csv/`, `tables/tex/`, `tables/xlsx/`

- `figures/pdf/`, `figures/png/`

- `memos/`

- `logs/`

- `data/`

- `sha256_outputs.csv`

`docs/table_figure_crosswalk.csv` maps each reference output to the generating script.

## Raw Data Sources

The data are public and come from:

- FDIC BankFind Suite Financials API.

- FDIC Summary of Deposits.

- FFIEC Central Data Repository Call Report single-period ZIP files.

- Federal Reserve Bank of St. Louis FRED.

See `docs/data_sources_and_terms.md` and `docs/raw_files_to_upload.csv`.

## Output Checks

`code/check_outputs.py` compares generated CSV tables to `reference_outputs/tables/csv/`, verifies that expected PDF figures were produced, and checks that the memo and checksum manifest exist.