A quién sirve — Funds and quant desks seeking edge beyond price and filings. Alternative data can front-run KPIs — if quality holds and cost is sustainable.
Alternative data means non-standard sources for investment and trading — satellite imagery, card transactions, foot traffic, web scraping, app data, employment, shipping, weather — integrated into fundamental or quant models via feature engineering.
In plain terms — «Off-balance-sheet» information measuring real activity before earnings — powerful but fragile.
Common categories
| Type | Use case | Risk |
|---|---|---|
| Consumer transactions | Retail revenue nowcast | Sample bias, privacy |
| Satellite | Oil inventory, mall parking | Cloud cover, cost |
| Web / app | Downloads, engagement | Bots, seasonality |
| Text | Lag, spam |
Due diligence: lineage, scraping TOS, survivorship, timestamp alignment (data leakage).
Integration
- Normalize and aggregate to model frequency
- Ablation: alpha vanishes OOS? possible overfitting
- Legal/compliance: GDPR, MNPI
- Cost vs marginal edge — many datasets commoditize after diffusion
Error típico — Buying expensive data without OOS test — in-sample correlation that fails live.
Ejemplo — Foot traffic + card spend on retail chain → nowcast same-store sales T-2 weeks vs consensus — feature in weekly model with walk-forward.
Card
- Check: sample bias, timestamp, legal.
- Test: OOS + decay monitor.
- Hub: AI & markets.