Event-Driven Signal Hypothesis Engine

A Python framework for testing whether discrete market events (e.g. congressional trades) change the distribution of future returns.

---

Objective

Answer a single question:

Does a specific event provide a measurable edge in future price movement?

This project does not build a trading system. It builds a hypothesis testing engine.

---

Current Status

• Congressional trade ingestion: working
• Event parsing and enrichment: working
• Backtesting engine: working
• Hypothesis framework: partially implemented

Current Research Direction

This project is evolving from a simple insider-event backtest into a dataset-discovery and event-study pipeline. Current focus areas:

• locating reliable disclosure datasets
• improving scraper coverage
• normalising event data
• comparing insider signals against market benchmarks

Known Issues

• filing_date mostly missing → event timing not reliable
• Current tests use transaction date (not ideal)
• Sample size small
• No statistical significance testing yet

---

System Overview

Data Sources (Free Only)

• Congressional trades (Senate disclosures)
• Price data via yfinance

---

Pipeline

raw data → parser → event table → backtest → report

---

Event Schema

ticker
event_date
filing_date
transaction_type
shares_before
shares_after
event_type
direction (always None at this stage)
source

---

Event Types

purchase
partial_sale
full_exit_sale

Derived:

cluster_flag
cluster_size

---

Key Design Principles

• Parser produces facts only
• Hypothesis layer assigns interpretation
• No narrative logic in code
• All signals must be testable
• Weekly iteration discipline

---

Example Hypothesis

hypothesis: purchase_bullish
event_type: purchase
expected_direction: bullish
horizons: [5, 20, 60]
min_sample_size: 10

---

Output

Each test produces:

mean return
median return
hit rate
sample size
baseline comparison
decision

Saved to:

reports/

---

How to Run (current)

Install dependencies:

pip install -r requirements.txt

Run pipeline (temporary entry point):

python main.py

(Note: entry script will be replaced with scripts/run_hypothesis.py)

---

Limitations

• Event timing not yet correct (disclosure vs transaction)
• Congressional data is delayed and incomplete
• Small dataset → results exploratory only
• No cross-validation yet

---

Roadmap

See:

ROADMAP.md

---

Non-Goals

• No trading bot
• No portfolio optimisation
• No news sentiment analysis
• No paid data (until edge is proven)

---

License

MIT