Alternative Data in 2026: What's Working, What's Not

In 2018, satellite imagery of parking lots was alpha. By 2026, every major quant fund has the same parking lot data from the same three vendors, and the signal has been largely arbitraged away. This is the pattern across all successful alternative data signals: discovery, adoption, commoditisation. The question for any alternative data investment is where you are in that cycle, and whether your processing capability gives you an edge even in a commoditised signal.

Earnings call transcript analysis remains one of the most robust alternative data signals. Not because the transcripts are novel — they're public — but because the signal is in the linguistic patterns that humans overlook: hedge word frequency, certainty language shifts quarter-over-quarter, management tone versus analyst question tone. LLM-based analysis of earnings calls has extended the signal by extracting entity-level sentiment for subsidiary businesses not separately reported, and by detecting inconsistencies between prepared remarks and Q&A responses. Funds using LLM-augmented earnings analysis are seeing 15–25% improvement in signal precision versus traditional NLP.

AIS shipping data, port congestion metrics, and trucking GPS data provide real-time supply chain visibility that leads official economic data by 3–6 weeks. This signal is most valuable in commodities, industrials, and consumer staples — sectors where supply chain dynamics directly drive earnings. The processing challenge is entity resolution: matching vessel names to cargo manifests to company supply chains requires a knowledge graph that most funds are still building. The funds with proprietary entity resolution have a durable edge.

Credit card transaction data from aggregators like Earnest Research and Second Measure was a genuine edge in 2019–2022. Consumer spending signals led official retail sales data by 2–3 weeks. Today, the major panel providers all have comparable coverage, prices have risen 3–5× as demand increased, and the factor has been incorporated into sell-side models and quant strategies widely enough that the excess return is minimal. The remaining edge is in panel-level segmentation — isolating spending by high-income cohorts, or by geographic markets not well-covered by aggregators — rather than the headline signal.

The volume of regulatory filings — 10-Ks, 10-Qs, 8-Ks, proxy statements, comment letters, no-action letters — exceeds what any research team can read. LLMs are now being used to extract structured data from these filings at scale: capex guidance, litigation risk disclosures, related-party transaction changes, executive compensation structure shifts. The signal is real but the processing is non-trivial — financial filings use complex cross-references, XBRL taxonomies, and accounting conventions that require domain-specific parsing before LLM extraction.

The bottleneck in alternative data alpha is rarely data access — most signals are available for purchase. The bottleneck is data infrastructure: ingestion pipelines that handle irregular delivery formats and schedules, entity resolution that maps vendor IDs to your internal security master, point-in-time correctness to avoid look-ahead bias in backtesting, and feature engineering that translates raw signals into tradeable factors. Funds that invest in this infrastructure can extract alpha from commoditised signals that competitors are misprocessing. Funds that don't have it will consistently underperform even their purchased signals.

Found this useful?

Share the key finding on X.