Congressional Disclosure and Retail Order Flow

Replication code for my MSc thesis. The main sample is the U.S. Senate over 2018 to 2024, built from TAQ trades, CRSP returns, and Quiver congressional trading records.

The code is organised as five short notebooks that run in order. Each notebook loads cleaned data, calls a helper function, and saves a table or figure. All the heavy logic lives in helpers.py, so the notebooks stay easy to read.

The notebook outputs are deliberately not cleaned. This shows that the numbers reported in the thesis are actual computations.

What the notebooks do

Notebook	What it does	Main output
00_build_panel.ipynb	Build the ticker-day panel (TAQ order flow + CRSP returns + controls).	data/licensed/work_panel.parquet
01_returns.ipynb	Disclosure-day return effect for Senate purchases, plus event-time and placebos.	return tables and figure
02_order_flow.ipynb	Retail and non-retail order-flow response, salience splits, top-senator channel.	order-flow tables
03_committee_relevance.ipynb	Match traded stocks to Senate committee jurisdiction (NAICS).	committee mapping
04_dongxu_replication.ipynb	Replicate the Dong and Xu (2025) House disclosure-day return effect.	outputs/tables/dongxu_replication.csv

Repository layout

msc_thesis/
  config.py            paths and sample dates
  helpers.py           all data loading and regression logic
  requirements.txt     Python packages
  notebooks/           00 to 04, run in order
  data/
    public/            small public files, tracked by git
    licensed/          restricted data, NOT in git (see below)
  outputs/
    tables/            saved CSV tables
    figures/           saved PNG figures
  wrds_cloud/          scripts that build the licensed data on WRDS

Setup

You need Python 3.11 or newer.

# from inside the msc_thesis folder
python -m venv .venv
.\.venv\Scripts\Activate.ps1        # Windows
# source .venv/bin/activate         # macOS / Linux
pip install -r requirements.txt

Then open the notebooks/ folder in VS Code or Jupyter, pick the .venv kernel, and run the notebooks in order from 00 to 04.

Data

The data is split into two folders:

• data/public/ holds small public files (FOMC dates, committee-to-NAICS maps, Senate committee membership). These are tracked by git.
• data/licensed/ holds restricted files from CRSP, TAQ, and Quiver Quantitative. These are not shared here. .gitignore blocks them.

To run the notebooks you must place the licensed files in data/licensed/ yourself, using the exact file names listed in data/licensed/README.md. Full provenance for every file is in DATA_AVAILABILITY.md.

If you have a WRDS account, the scripts in wrds_cloud/ rebuild the CRSP and TAQ inputs from scratch.

Run order

The notebooks depend on each other, so run them in order:

1. 00_build_panel.ipynb builds work_panel.parquet, which 01 and 02 read.
2. 01_returns.ipynb and 02_order_flow.ipynb use that panel.
3. 03_committee_relevance.ipynb rebuilds the committee match used in 02.
4. 04_dongxu_replication.ipynb is a standalone replication and can run on its own.