HARBINGER

Pre-announcement M&A signal detection using public alternative data.

Hedge funds pay $5–50M/year for datasets that move before M&A announcements. This is the infrastructure to build one.

---

The Thesis

Mergers leave a trail of computable public signals weeks before announcement:

Signal	Source	Why It Matters
SEC form clustering	EDGAR	S-4s, SC TO-Ts, 13-D amendments cluster before deals
Corporate jet co-routing	OpenSky Network / FAA	Executives meet secretly; planes don't lie
Job posting freeze	Indeed / LinkedIn	Targets freeze hiring 3–6 weeks before close
Patent cross-licensing	USPTO	IP agreements often precede full acquisitions
Law firm lateral hires	State bar / LinkedIn	M&A lawyers move before deals, not during
Lobbying disclosure spikes	Senate LBDB	Regulatory prep happens before announcement

None of this is insider information. All of it is public. Nobody has aggregated it systematically.

---

Architecture

collectors/          → Pull raw signals from public APIs
  sec_edgar.py       → EDGAR full-text search, filing parse
  flight_tracker.py  → OpenSky Network corporate jet tracking
  job_postings.py    → Job freeze detection
  patent_signals.py  → USPTO assignment API

signals/             → Feature engineering on raw data
  features.py        → Per-company signal vectors
  scoring.py         → Composite anomaly scores

models/              → ML prediction layer
  predictor.py       → XGBoost model: P(acquisition | signals)
  trainer.py         → Training on historical deal data

api/                 → Serve predictions
  routes/predictions.py  → GET /predict/{ticker}
  routes/signals.py      → GET /signals/{ticker}

scripts/
  backtest.py        → Validate signal lead time on historical deals
  ingest_historical.py → Seed DB with known deals for training

---

Quickstart

cp .env.example .env
docker-compose up -d
pip install -r requirements.txt

# Seed historical deals and train
python scripts/ingest_historical.py
python scripts/train.py

# Run the API
uvicorn src.api.main:app --reload

Endpoints:

• GET /predict/{ticker} — Current M&A probability score
• GET /signals/{ticker} — Raw signal breakdown
• GET /watchlist — Top 20 highest-scoring companies right now
• GET /alerts — New high-score crossings in last 24h

---

Revenue Model

1. License the dataset — Quant funds pay $1–10M/year for edge. One confirmed signal sells the product.
2. Signal API — SaaS, $50K–500K/year per fund. Scales to zero marginal cost.
3. Run your own book — Use the model yourself. No ceiling.

---

Legal Status

All data sources are public. The legal theory is identical to how firms like Quiver Quantitative, Thinknum, and Eagle Alpha operate. MNPI (material non-public information) law applies to information from insiders — not to pattern recognition on public filings.

We are not trading on tips. We are trading on math.