Market data licensing: what the contract lets you do with a price

Non-display use means the data never reaches a human screen: backtesting, research, an internal risk model, an algorithmic signal. Display use means someone sees it, in an app, a web page, a chart or an email. Intrinio's licensing guide describes the practical consequence: display licenses commonly carry a per-user fee, and the split between data shown to authenticated users and data shown publicly to any visitor matters again, with public display the more expensive of the two.

The counterintuitive part is that non-display is often the cheaper right even though it drives the trading decision. The reason is administrative: a non-display license covers a use, and a display license covers a headcount, which has to be counted and reported every month. A firm whose user numbers move finds the display side harder to administer than the trading side, and the trading side is where the money is made.

Derived data is anything computed from licensed data: a signal, a score, an index level, a model output. The test a contract applies is whether a recipient of your output could reconstruct the underlying values from it. A daily average of one instrument fails that test, because the input is recoverable. A portfolio risk number built from hundreds of inputs passes it.

Exchanges protect derived data in several ways, and CME Group's own description of its data licenses is the reference a European firm usually ends up reading, because the same structures appear in the European venue contracts. The protections are a separate derived data license, a restriction of derived output to non-display use, an additional fee, and a limit on who may receive the derived figure. The last one is the trap: a firm licensed for internal derived use that sends the number to a client has redistributed, which needs its own permission.

Redistribution is passing data outside your legal entity, and a group with several entities is already redistributing between them unless the contract says otherwise. That catches firms with a shared data platform across subsidiaries more often than it catches anything else.

Per-user reporting follows from the display license. The firm declares how many people had access in a period, and the vendor reserves the right to check. An audit in this market reads access logs, entitlement systems and application screenshots, and the common findings are a developer account that kept production entitlements, a dashboard embedded in an internal wiki that a wider group could open, and a test environment pointing at live data. Each of these is an entitlement question, not a technical one, which is why market data administration sits with a dedicated team in larger houses.

MiFID II made the European trading venues price their data on a reasonable commercial basis, which means fees based on the real cost of producing and distributing the data, offered as a standalone product and not bundled into other services. The implementation of that principle has moved. The regulatory technical standards on reasonable commercial basis entered into force on November 23, 2025, with a transition until August 22, 2026 for providers authorized before that date to align existing contracts, and ESMA withdrew its earlier guidelines on the market data obligations in favor of the standards.

Two consequences reach a data consumer. Published fee schedules and disclosures are now the norm and no longer a negotiation point, so a firm can compare terms before it signs. And the free delayed data that MiFID II requires, available after the delay period, is a legitimate option for a use case that does not need the live number, such as an end-of-day report or a public website. Finance Loop's background on the directive is at MiFID II.

A crypto data vendor licenses on the same four axes, with one difference that matters: there is no exchange monopoly behind the data, because the trades happened on venues that mostly publish their own public APIs. That gives a buyer leverage a buyer of listed equity data does not have, and it also means the vendor's product is the normalization, the history and the coverage, and not exclusive access.

The second difference is the free tier. A free crypto data API usually permits non-commercial use with attribution and forbids redistribution and commercial display, which rules out exactly the internal dashboard a firm wanted it for. Reading those terms before the pilot saves the rebuild. The vendors and what each supplies are at crypto price feed providers, and the question of who may publish a benchmark built from the data at crypto price discovery.

Model training is a use the older contracts do not name, and silence in a contract is not permission. Where a license is silent, the vendor's position is usually that training is a derived data use, that the trained weights are a derived product, and that model output reaching a client is redistribution of derived data. Firms that got this wrong found out at the audit, not at the launch.

Three things settle it in practice: a written amendment that names model training as a permitted purpose, a clause on whether the weights survive the end of the contract, and a position on whether model output may go to a client. Finance Loop covers the data side of this at AI and financial market data and the governance side at the EU AI Act in finance.

The display fee per user is not one number, because the exchanges charge a professional subscriber several times what they charge a non-professional one. The test is not seniority: a subscriber counts as non-professional when the person uses the data for private purposes only, is not registered with a securities regulator, is not employed in a role that uses market data, and does not act for a business. Everyone else is professional, including the sole trader who trades his own book as a business.

The classification is the firm's to make and the firm's to defend, and misclassifying a group of users is the single most expensive audit finding in this field, because the correction applies per user and per month for the whole period. A retail app with a professional user it treated as retail owes the difference for every month that user had access. Firms that get this right ask the classification question at registration, store the answer with the account, and re-ask it when the account changes.

A vendor's contract rarely clears a firm with the exchange that produced the data. Intrinio's licensing guide states the practical division: obtaining the exchange's approval is the licensee's responsibility, and the vendor supplies the data once the approval exists. A firm that signs a vendor agreement and launches a feature without the underlying exchange permission has a vendor who is paid and an exchange that has not agreed to anything.

That division explains why a data project has two timelines. The vendor contract is a commercial negotiation measured in weeks. The exchange approvals are per venue, per license category and per use, each with its own form and its own review, and a product that displays data from several venues needs all of them before launch. Planning the approvals after the build is the common failure, because a refused approval removes a venue from a product that already assumed it.

Market data licensing is the contractual permission to use price and trade data for a named purpose, separate from the technical delivery of that data. One subscription can sit under several licenses at once, because the vendor prices the purpose and not the pipe. The four purposes nearly every contract distinguishes are display, non-display, derived data and redistribution.

Non-display use is any use where no person reads the data: an algorithm that places an order, a risk model that computes an exposure, a backtest, a reconciliation job, an internal research notebook. The moment the number appears on a screen that a person looks at, it has become display use, and an internal dashboard is display use even though it never leaves the firm.

Yes, and the terms of service are that license. A free tier typically permits non-commercial use, requires attribution, caps the request rate and forbids both redistribution and commercial display. A firm that builds a client-facing feature on a free tier is in breach of the terms it clicked through, independent of whether anyone invoiced it.

The vendor or the exchange audits, usually through a specialist firm it appoints, and the right to audit is in the contract the subscriber signed. The audit examines entitlement records, user counts and the applications that consume the feed, and an underdeclaration produces a retroactive invoice for the periods concerned. This is a commercial process and not a supervisory one, so it runs separately from anything BaFin or ESMA does.

Start from the product and not from the contract. Write down every place a number from the feed appears or is used, including internal dashboards, emails, PDF reports and API responses to clients. Classify each of those as display, non-display, derived or redistribution. For the display ones, name the user category and the professional classification, and estimate the count at ten times the current size, because the fee curve is the thing that surprises firms later. Then check the real latency need: a use case that tolerates delayed data can take the free delayed feed and drop out of the licensing question. Finally, give one named person the ownership of the entitlement records, because an audit asks a person and not a team.

Delayed data from an EU trading venue is free of charge after the delay period under MiFID II, and the venue's terms of use still govern it. In practice that means attribution, a prohibition on presenting the data as real time, and in many cases a restriction on redistribution even though the data costs nothing. A free feed with terms is still a license, and a public website showing delayed prices without the venue's attribution is in breach of them.