News as alternative data for quant funds and systematic trading

Systematic desks don't want a news reader — they want a clean, point-in-time signal they can backtest without lookahead bias. Deterministic scoring and event clustering turn a news feed into exactly that.

What makes news usable as alternative data

For a quant workflow, headlines are useless but structured, point-in-time events are gold. Every article here carries a UTC fetched_at timestamp plus urgency (0–10), topic_tags, country_tags, political_lean and an event cluster_id — event-level features, not raw text you have to NLP yourself.

Deterministic = backtestable

The urgency engine is deterministic rules, not a drifting model, so a score computed in a 2025 replay is the same one you'd get live today — no silent re-labeling, no lookahead leaking in through a retrained model. Thresholds mean the same thing across your whole sample. That reproducibility is the difference between a usable factor and noise; the scheme is documented in the methodology.

Construct signals

curl -H "X-API-Key: YOUR_KEY" \
  "https://api.newsagentdata.com/v1/feed?audience=trading&topic=rates&min_score=6&days=90"

Aggregate urgency by topic/country/day for an attention factor; use cluster_size as event intensity; count high-urgency clusters as event density. The audience=trading filter pre-narrows to market-relevant items.

Backtest on the archive, deliver live

Historical rows share the same schema and timestamps, so you can pull a window and join to prices by fetched_at without lookahead (see the historical data guide). For live, a webhook or SSE stream pushes scored events in ~60s, de-duplicated so one central-bank decision is one signal, not forty (see event clustering).

Honest note

This is a signal input, not a strategy — validate it in your own pipeline before trading. Russian and English are the deepest-enriched; political_lean "neutral" means unclassified. Complements the real-time-alerts approach in the news API for trading guide.