What is alternative data?
From AltData.wiki, The Alternative Data Encyclopedia | Concept | 4 min read
Reviewed
Alternative data in investment research: what the term covers, how it differs from data format, and why provenance, timing, access and validation matter.
The short definition
In investment research, alternative data refers to information beyond conventional financial statements, filings, market data and similar established inputs. The SEC's Division of Examinations gives examples including satellite imagery, aggregated card transactions, social-media and search information, mobile geolocation and email-derived data. It explicitly notes that alternative data does not necessarily contain material non-public information.[1]
The boundary depends partly on source and use. CFA Institute's 2024 report distinguishes alternative data from unstructured data: the former describes the information's relationship to traditional investment inputs, while the latter describes its format. A conventional filing can therefore be the source for a nontraditional textual feature without turning every filing into a new alternative dataset.[2]
Where it came from
Two documented reference points illustrate the topic's place in investment practice. In April 2022, SEC examination staff reported deficiencies in advisers' diligence and controls around alternative-data service providers.[1] CFA Institute's 2024 report addresses alternative and unstructured data, their processing and ethical considerations.[2] Neither publication establishes a single origin date or a universal history of industry adoption.
Market structure today
A source, processor, distributor and analytical product are different roles. A collector may supply observations; a vendor may clean or model them; a marketplace may provide discovery or delivery; and a buyer may construct a signal. Determine which entity is responsible for the underlying collection and which contract governs the intended use. The SEC's alert highlights the importance of consistent provider diligence and reassessment when collection practices change.[1]
A marketplace listing is not an independent validation of the data or a replacement for its licence. Likewise, a processing platform is not the owner of every dataset it hosts. Google Earth Engine illustrates the distinction between data access and the compute and storage resources used to analyze it.[3]
The main signal families
The SEC's examples span consumer transactions, imagery, online activity and geolocation.[1] This encyclopedia groups sources into signal categories such as card transactions, jobs and workforce and satellite and geospatial data. These are navigation categories, not a claim that all products in one family share a method or predict the same outcome.
Alternative data can be structured, semi-structured or unstructured. A delivered table of sentiment scores is structured even if its source was free text. The conversion can remove context or introduce errors, so the transformed output needs evaluation as well as the original source.[2]
How funds actually use it
A researcher might compare an operational indicator with a company's disclosures, investigate an event or test a feature in a forecasting model. These are possible analytical workflows, not proof that the indicator adds useful information. For example, more advertised jobs could reflect expansion, replacement hiring or a change in recruiting policy.
Keep the measured quantity explicit. BLS defines job openings as a point-in-time stock and hires as additions to payroll over a month; an online posting count is not automatically either measure.[4] Comparable definitions, historical timing and a relevant benchmark are prerequisites to interpreting a signal.
What it costs
The label alternative data does not determine a price. Compare the product, history, geography, delivery, users and permitted uses in each offer. Free registration, free access to a sample and an open reuse licence are separate conditions; none establishes the other.
Processing can also be charged separately from access to source records. Earth Engine's published pricing distinguishes compute and storage usage from platform-plan fees and provides a separate noncommercial access path.[3] Such terms are examples of a platform's charging model, not a representative price range for the entire alternative-data market.
Risks, bias and decay
A research design must consider which observations are missing, how entities are matched and whether a historical file includes later corrections. Removing closed businesses from a present-day universe, for example, can change the apparent history of that universe. A collection or modeling change can also resemble a change in economic activity.
Revisions are not unique to alternative data: BLS documents monthly and annual revisions to JOLTS estimates.[4] Preserve releases and availability dates rather than assuming today's file reproduces what was known earlier. Evaluate out-of-sample results and monitor changes after deployment; no data label guarantees a durable predictive relationship.
Compliance in one paragraph
The SEC's April 2022 alert describes deficiencies in documented provider diligence, assessment of collection terms and legal obligations, and decisions about when diligence should be repeated.[1] It also distinguishes alternative data from MNPI: the former does not necessarily contain the latter. The alert represents examination-staff views, not a new rule or a legal approval of any dataset. Applicable privacy, collection, licensing and securities-law obligations require case-specific assessment.
Limitations and criticism
Novelty and quality are different properties. An unusual source may be irrelevant to a particular question, while a familiar public benchmark may be essential to evaluating it. A backtest must distinguish incremental information from exposure to existing factors, later revisions and unrealistic implementation assumptions.
Published coverage and performance figures should be read with their scope, date and methodology. A vendor case study does not establish a result for every buyer, and a survey of selected respondents is not a census of the industry. This article therefore does not present an unsupported market-size estimate or a universal return claim.
Is it worth it?
Suitability is specific to the question, data and implementation. Evidence should include a documented source, understandable transformations, realistic availability dates, permitted use and evaluation against an appropriate benchmark. These checks support a decision; they do not guarantee returns.
Use the encyclopedia for explanations of providers, marketplaces and signal families, and the dataset catalogue for individual products, documentation and access information.
Frequently asked questions
- What counts as alternative data?
- Nontraditional information used in investment analysis, with boundaries that depend on source and use. It is not synonymous with unstructured data, private information or a profitable signal.
- Who buys alternative data?
- Investment managers and other organizations can use these sources for research, monitoring and decision support. The same dataset can serve different purposes; this article does not rank buyer groups by spending.
- How big is the alternative data market?
- No single market size follows from the definition. An estimate needs a named source, period, methodology and boundary between alternative data, conventional data and analytical services. No market-size figure is asserted here.
- Is using alternative data legal?
- The label does not determine legality. Collection, privacy, licensing, contractual and securities-law obligations must be assessed for the source and intended use. The SEC notes that alternative data does not necessarily contain MNPI.