Home / Libraries / ML4T Data / Docs
ML4T Data
ML4T Data Documentation
Unified market data acquisition from 19+ providers
Skip to content

Alternative Data Sources

This page maps sources of alternative data: news and text, social and web attention, app and web usage, job postings, card transactions, foot traffic, satellite and shipping data, ESG ratings, supply-chain relationships, and prediction markets, plus the marketplaces where such datasets are found. For each source it lists what the data measures, how far back it goes, what is known about its point-in-time behavior, the access tier, and whether ml4t-data wraps it.

Verified September 2026

Alternative data vendors are acquired, merged and repackaged often. Every entry below was checked in September 2026 against the vendor's own site, a press release, or published reporting. Few vendors publish their timestamp and revision policy; where it is not public, the table says "not documented" and the question belongs in your due diligence. Confirm the current terms before you build on any of them.


Choosing a Source

When was each record first observable?

A backtest is only valid if every record is used after the time it could first have been seen. Ask the vendor for the timestamp that marks first availability, not the event date, and whether history was backfilled when the dataset launched or when coverage expanded. A backfilled history looks continuous but was assembled later, often by a method that did not exist at the time.

Revisions and methodology changes

Panels change composition (card and device panels gain and lose users), models are re-estimated, and rating methodologies are revised. ESG ratings are the clearest case: providers disagree with each other and revise their own methods over time. Ask whether old values are restated when the method changes and whether earlier vintages are kept.

Coverage and representativeness

Check which names, sectors and regions a dataset actually covers, and whether coverage is stable across the sample. A signal that exists only for large US consumer companies is a narrower signal than its marketing suggests.

Provenance, permitted use and redistribution rights are hard constraints. Narrow panels can carry material non-public information, and granular location or device data carries privacy risk even without names. These questions are answered before any backtest, not after.

In the book

  • Chapter 4, "Understanding alternative data" (section 4.4): "The alternative data landscape", "The evaluation framework", "Build versus buy for alternative data", and "Data sources and implementation".
  • Chapter 4, "Using text data for NLP features" (section 4.5): extracting and storing text from filings point-in-time. Chapter 10 covers turning text into features.
  • Chapter 2, "A modern taxonomy of financial data" (section 2.1), subsection "Alternative data - High variety, high validation burden".

News, Text and Sentiment

Source What it measures History Point-in-time and revisions Access ml4t-data
GDELT Project Global news events and tone (Event database); themes, entities and sentiment from full text (Global Knowledge Graph) Events from 1979; Knowledge Graph from April 2013 Published in 15-minute batches; the batch time is a usable first-availability time. Older history was built retrospectively. Free and open No
RavenPack Entity-level news analytics: relevance, sentiment and event classification From 2000 (Dow Jones edition), 2007 (web edition) Records carry the news timestamp; revision policy not documented publicly Institutional; academic via WRDS No
LSEG MarketPsych Analytics Sentiment and emotion scores from news and social media across equities, macro, FX, crypto and commodities From 1998 Minute-level to daily series; revision policy not documented publicly Institutional No
Dow Jones Factiva DNA Licensed premium news archive with metadata, via API, snapshots and streams Archive depth varies by publication Publication timestamps per article Institutional No
Benzinga Newsfeed API Financial news headlines and articles, Wilshire 5000 and TSX coverage Not documented created and updated timestamps per article; updatedSince returns changes Paid No
SEC EDGAR full-text search Full text of SEC filings From 2001 Every filing carries its acceptance timestamp Free No

Social Media and Web Attention

Access to social-media data has narrowed since 2023, so check the current terms before planning a project around any of these.

Source What it measures Status Access ml4t-data
Stocktwits API Investor message stream with cashtags and self-reported sentiment Closed to new developer registrations Not open to new developers No
Reddit Data API Posts and comments, including investing subreddits Paid for commercial use since July 2023 Commercial license required No
X API Posts and engagement Pay-per-use credits since February 2026, replacing subscription tiers Paid No

Web, App and Workforce Data

Source What it measures History Point-in-time and revisions Access ml4t-data
Similarweb Website and app traffic and engagement estimates About 10 years Modeled estimates; revision policy not documented publicly Free tools, paid, enterprise No
Sensor Tower App downloads, revenue, rankings and usage estimates; includes the former data.ai Not documented Modeled estimates; revision policy not documented publicly Enterprise; free top charts No
Thinknum Web-collected company data: job listings, headcount, product prices, store locations Not documented Not documented Institutional No
YipitData Company KPI and revenue estimates from web, app and transaction data Not documented Not documented Institutional No
Revelio Labs Workforce composition and flows, job postings, employee sentiment, layoff notices Workforce composition from 2007, transitions from 2008, postings from 2021 Not documented Paid; academic via WRDS No
LinkUp Job postings collected directly from employer websites From 2007 New, updated and removed listings are captured nightly Institutional No

Card Transactions

Card panels measure the spending of a sample of consumers. Panel composition changes over time, so ask how the vendor weights the panel and whether history is reweighted when it changes.

Source What it measures History Access ml4t-data
Bloomberg Second Measure US consumer card spending by merchant and company 8+ years; 2 to 7 day reporting lag Bloomberg Terminal and enterprise No
Consumer Edge Transact Card transactions by merchant, company and demographic; now includes Earnest Analytics About 9 years Institutional No
Facteus Debit and credit card transactions, synthesized for privacy Not documented; about 1-day lag Institutional feeds (API, AWS, Snowflake) No

Foot Traffic, Satellite and Shipping

Source What it measures History Access ml4t-data
Placer.ai Visits to stores and venues from a mobile-device panel Not documented Free tools, paid platform, enterprise API No
Advan Research Patterns+ Foot traffic to points of interest, US and Canada; took over SafeGraph's Patterns product Weekly updates; history from 2019 Institutional No
Planet Labs Satellite imagery: near-daily medium resolution and tasked high resolution Archive from 2009 (RapidEye, 2009 to 2020) Commercial; education and research program for university users No
RS Metrics Asset-level signals from satellite imagery: metals production at smelters, retail parking lots Not documented Institutional; data for research No
Kpler Commodity flows from ship tracking (AIS), satellite, customs and port data; now owns MarineTraffic and Spire Maritime Not documented Institutional No

ESG

ESG ratings disagree across providers and change when methodologies change. Treat a rating history as a sequence of vintages, and check which methodology produced each value.

Source What it measures History and methodology Access ml4t-data
MSCI ESG Ratings Industry-relative ESG ratings (AAA to CCC), about 17,000 issuers Time series since 2007 Institutional; free public lookup for a subset No
Morningstar Sustainalytics ESG Risk Ratings Unmanaged ESG risk score, 16,000+ companies A score change log exists; vintage policy not documented publicly Institutional No
LSEG ESG Scores ESG scores from 220+ indicators, plus controversies New suite launched March 2026 on 220 standardised indicators; how it relates to earlier LSEG scores is not documented publicly Institutional No
S&P Global ESG Scores Scores from the annual Corporate Sustainability Assessment Annual assessments; history not documented publicly Institutional No
RepRisk Reputational and conduct risk from news and stakeholder sources Daily series with a consistent methodology since January 2007 Institutional No

Supply Chain and Trade

Source What it measures History Access ml4t-data
FactSet Supply Chain Relationships (formerly Revere) Customer, supplier, partner and competitor links between companies North America from 2003; other regions from 2011 to 2016 Institutional; academic via WRDS No
Bloomberg Supply Chain (SPLC) Supplier and customer relationships with revenue and cost exposure, 100,000+ companies From 2006 Bloomberg Terminal and enterprise feed No
S&P Global Panjiva Shipment-level customs records, 2 billion+ records Not documented Institutional No
ImportGenius US and international bill-of-lading records US imports from 2006, exports from January 2014 on the Pro and Enterprise tiers Self-serve subscriptions; enterprise plans No

Prediction Markets

Prediction-market prices are market data: each trade and quote has an exchange timestamp and is never revised. The caveats are thin liquidity outside headline contracts and markets that are listed and resolved over time, so a set of markets picked today is a survivor sample.

Source What it measures History Access ml4t-data
Kalshi CFTC-regulated event contracts: economics, politics, weather, sports From the July 2021 launch Public market-data API; trading needs a verified account Yes: KalshiProvider
Polymarket Event contracts on a crypto platform; CFTC approval for intermediated US access, November 2025 From 2020 Public market data Yes: PolymarketProvider

Marketplaces and Discovery

Marketplaces simplify procurement; they do not validate point-in-time behavior or methodology.

Source What it offers Access
Nasdaq Data Link (formerly Quandl) Catalog of financial, economic and alternative datasets, some free Free account for free datasets; paid for premium
Snowflake Marketplace Datasets shared into a Snowflake account and queried in place Free, trial and paid listings
AWS Data Exchange Datasets delivered into AWS, billed through AWS Per-product subscriptions
Eagle Alpha Alternative data discovery and advisory, 2,500+ dataset profiles Institutional
Neudata Alternative data discovery and research, 7,000+ datasets in its catalog Institutional
Datarade Directory of data providers and data marketplaces Free to browse; each provider sets its own terms

Of the free datasets that made Quandl popular, the WIKI prices table is frozen at April 2018 (ml4t-data wraps it as Wiki Prices), and the Zillow real-estate database no longer has a catalog page and was last refreshed in July 2025.


Alternative Data in ml4t-data

ml4t-data wraps two prediction-market sources, Kalshi and Polymarket. The FXMacroData provider also exposes news, risk-sentiment and CFTC positioning endpoints alongside its release-timestamped macro data. Everything else on this page needs its own client.


Name Changes and Closures

Then Now
data.ai (formerly App Annie) Part of Sensor Tower since 2024
Earnest Analytics Part of Consumer Edge
SafeGraph Patterns (foot traffic) Sold to Advan Research; SafeGraph now sells places, geometry and spend data
Orbital Insight Acquired by Privateer, May 2024
MarineTraffic Acquired by Kpler, March 2023
Spire Maritime Acquired by Kpler, April 2025
Refinitiv ESG scores LSEG ESG data; a new LSEG ESG Scores suite launched March 2026
Reddit and X APIs Paid access (Reddit since July 2023; X pay-per-use since February 2026)
Stocktwits API Closed to new developers
Polymarket (closed to US users from 2022) Amended CFTC order of designation for intermediated US access, November 2025
PredictIt CFTC approval to operate a regulated exchange, September 2025
Quandl Nasdaq Data Link (September 2021)

See Also