Market data feeds and the practical cost of a tick

Data cables connecting trading servers in a broker's market data infrastructure.
Photo by Brett Sayles on Pexels

Market data is not a rounding error in a trading stack. The exchange invoice may look small beside a matching engine or a colocation bill, but it is only one line in the cost of producing a usable tick.

For a broker or prop firm, that cost includes the rights to consume or display data, the connections that carry it, the compute that normalises it, and the storage and recovery systems that keep it available. Counting only exchange messages misses both the commercial obligations and the engineering work.

Start by defining the unit. A packet can contain several updates, while one exchange message can represent a trade, a quote change or a book-depth delta. A normalised event may combine or discard input messages. Choose a consistent unit, such as accepted feed messages or published quote updates, before comparing systems.

Build the cost from the rights outward

Market data fees depend on venue, product, use and distribution. A screen shown to a professional user can be priced differently from data used by an algorithm, a risk process or an internal dashboard.

Derived data and redistribution can trigger separate reporting or fee rules. The agreement, not the feed handler, defines what the firm may do with each field.

The figures below are typical planning ranges in USD, not vendor quotes. Actual tariffs vary by exchange, user classification, access method and contract. Treat them as prompts for a budget model, then verify current schedules and definitions with each venue.

Illustrative typical market ranges for monthly planning

Cost lineTypical rangeBasisMain cost driver
Professional display entitlementUS$50-US$150Per user, per venueUser classification and venue tariff
Non-display usageUS$500-US$10,000Per application or venueUse case, data scope and reporting terms
Feed access or portUS$500-US$5,000Per venue connectionDirect access, port type and service tier
Colocation cross-connectUS$100-US$500Per connection, per monthFacility and connection type
Cloud data egressUS$0.02-US$0.10Per GBProvider, region and traffic volume

Display entitlements are often the easiest item to count and the easiest to understate. Reconcile exchange reports with identity records and application logs. Count users who can view real-time data, including support and operations staff where the licence requires it. For non-display use, map each feed to the consuming application and record whether it creates derived data or supplies an external service.

Choose the feed architecture for the workload

A consolidated feed reduces the number of venue connections and simplifies normalisation. It can suit a broker showing top-of-book prices or a risk service that needs broad coverage.

Direct feeds provide venue-specific depth and control over the path, which matters for latency-sensitive strategies and execution analytics. They also bring more channels, sequence handling, recovery logic and licence administration.

Typical active venue feeds can require processing from tens of thousands to several hundred thousand messages per second, with short bursts several times higher. The number depends on the instrument set, depth, session and volatility.

Benchmark bursts, not just the quiet part of the session. A feed that works at an average rate can still build a queue or miss recovery deadlines when the market moves quickly.

Before adding another direct feed, test these trade-offs:

  • Latency: measure exchange-to-process and process-to-publish time at the median and high percentiles. Kernel bypass and pinned threads can reduce jitter, but they add operational complexity and restrict where the handler can run.
  • Completeness: use sequence numbers, gap detection and snapshot recovery. A faster line is not useful if the book remains incomplete after packet loss.
  • Redundancy: dual multicast paths improve resilience, but duplicate handling, monitoring and sometimes connection charges add cost.
  • Scope: subscribe to only the instruments and depth the strategy needs, where venue rules permit it. Filter after the network interface and the feed still consumes bandwidth and processing capacity.

Transport choices shift costs rather than remove them. Colocation can shorten the path and avoid repeated cloud transfer, while a cloud deployment can simplify scaling and regional distribution. Compare the full monthly cost of ports, cross-connects, compute, egress and on-call support. A low per-GB rate does not make replicated raw feeds cheap if every service copies them.

Measure cost per usable event

Use a monthly numerator that includes market data fees, access charges, network, feed-handler compute, storage, monitoring and an allocated share of operations. Divide it by the number of accepted events actually delivered to consuming services.

Keep a second figure for raw input messages. The gap between them exposes duplicate traffic, rejected updates and normalisation choices.

For example, assume a system processes 100,000 accepted messages per second during a 6.5-hour session for 21 trading days. That is about 49.1 billion events in the month. If the allocated monthly cost is US$12,000, the result is roughly US$0.24 per million events, or US$0.00000024 per event. This is an illustrative calculation, not a typical firm result. Change the session length, event definition or allocated cost and the answer changes.

Include storage in the calculation. At 100,000 messages per second and an average raw record size of 100 bytes, an uncompressed capture produces about 864 GB in a full day. Indexes, metadata and replicas increase that total. Retaining every feed indefinitely is rarely a neutral choice; set retention by use, replay needs and regulatory obligations, then test that compression and replay preserve the fields required by downstream systems.

Track these measures by venue and feed handler:

  • Accepted, dropped, duplicate and rejected messages, alongside gap and recovery counts.
  • Peak message rate, queue depth, CPU use and memory pressure during volatile intervals.
  • End-to-end latency percentiles, stale-book duration and time to rebuild after a gap.
  • Monthly cost per million accepted events, with licensing and infrastructure shown separately.

Those splits help locate the expensive layer. Rising cost per accepted event can come from a higher exchange tariff, a growing user count, excess subscriptions or a handler that discards too much input after paying to receive it. Do not optimise only for a lower event count: dropping an update may cut compute while leaving fees unchanged and degrading a price, risk check or audit trail.

Make licensing and operations part of release control

Data rights belong in the same change process as feed configuration. A new symbol universe, client display, derived field or analytics service can change the licence position even when the network design stays untouched. Maintain a register linking venue, product, data type, consuming application, display status, user population and retention policy. Review it when teams deploy a new consumer or change the purpose of an existing one.

Operationally, keep the raw capture and the normalised output distinguishable. That makes it possible to replay a session, check a book reconstruction and trace whether a bad price came from the venue stream, recovery logic or a downstream transformation. Set alerts for sequence gaps, stale timestamps and sustained queue growth. Test the alert path and the recovery process outside market hours, then repeat after changes to the handler or network.

For a wider view of the systems around feeds, Trading Technology's infrastructure coverage follows the platforms, software and providers brokers and prop firms use. In procurement, ask vendors to separate recurring access, usage, connectivity and support charges. In engineering, publish the same cost categories beside latency and completeness metrics so a cheaper design cannot hide a weaker service.

The exchange still charges for the feed whether the strategy reads every tick or not.

Frequently Asked Questions

How do you calculate the cost per tick?

Add the monthly data rights, connectivity, processing, storage and operations costs, then divide by accepted events delivered to consumers. Keep the event definition consistent and report the cost per million events for easier comparison.

What is the difference between a market data message and a tick?

A message is an item received from a feed, while a tick is a market update as defined by the system counting it. One packet can carry several messages, and a normalised tick may combine or omit input messages.

Are direct market data feeds always faster than consolidated feeds?

Direct feeds can provide a shorter, venue-specific path, but performance depends on network location, handler design and recovery behaviour. Measure latency and completeness under peak load before choosing a feed based on its label.

What are non-display market data fees?

They are charges or licence terms for using market data in applications rather than presenting it to a human user. Algorithms, risk systems and internal analytics may fall under non-display rules, depending on the venue agreement.

How can a broker reduce market data feed costs?

Review subscriptions, user entitlements, application use and retention against current business needs and licence terms. Filtering and consolidation can reduce processing or transport costs, but they do not automatically reduce exchange fees.