Plataformas · Software · Provedores · Mercados
Week one: a broker chooses two venues, prices top-of-book access and expects a modest monthly bill. Week six: the feed is live, but the invoice also includes non-display usage, a second site, normalisation support and charges tied to how the data is consumed.
The gap is not unusual. The cost of a tick is rarely just the price of receiving one exchange message.
Consider a mid-sized broker or prop firm adding direct feeds for a new cash-equity market. The figures below are planning ranges, not audited results for a particular company. Actual charges depend on venue tariffs, professional status, distribution rights, volume and contract terms.
Week one: define what the feed must do
Separate the use cases
Start by listing every consumer. A trader's display, a pricing engine, a smart order router and a historical capture service may all use the same market data, but licensing rules can treat them differently. A display licence does not automatically cover non-display use, redistribution or derived data.
Then specify the required view. Top-of-book is cheaper and lighter to process than full depth, but it may not be enough for an execution algorithm that needs queue position or order-book imbalance. Trades-only data can support some analytics, but cannot reconstruct the live bid and offer.
- List each venue, instrument group and required depth.
- Map each feed to its consumers, including test, recovery and research systems.
- Record whether data is displayed, used for decisions, stored or redistributed.
- Set a peak message-rate estimate, not just an average daily volume.
A common sizing mistake is to count users and ignore applications. A service running on one server can still create a non-display entitlement, while a shared display may carry per-user charges. Confirm the exact definitions with each exchange or authorised vendor before committing.
Choose a delivery path and failure mode
A vendor feed reduces the number of exchange connections to manage and can simplify normalisation across venues. Direct feeds offer more control over the path and often lower latency, but the firm takes on more work around connectivity, protocol changes, sequence recovery and monitoring. Neither option removes the need to understand the licence.
For a small deployment, a primary feed and a tested recovery path may be enough. A latency-sensitive firm may run redundant lines into separate network paths or sites. Redundancy can double some connectivity costs and can also create duplicate data that must be detected and reconciled. Decide whether failover is hot, warm or manual, and test it under packet loss rather than relying on a diagram.
Weeks two to four: build the cost model
Count recurring costs, not just the exchange tariff
For planning, separate charges into data rights, delivery, processing and operations. The table shows typical market ranges for a modest multi-venue deployment.
These are broad monthly planning figures, not quotations. Exchange tariffs vary sharply, and some are billed per professional user, application, device or use case rather than as a flat feed fee.
| Cost area | Typical monthly range | What drives it |
|---|---|---|
| Exchange data rights | US$1,000 to US$25,000+ | Venues, professional users, non-display use and depth |
| Vendor feed or normalisation support | US$1,000 to US$8,000 | Number of feeds, service level and software scope |
| Connectivity and cross-connects | US$500 to US$4,000 | Sites, circuits, ports and resilient paths |
| Compute, capture and monitoring | US$500 to US$3,000 | Peak throughput, retention and deployment model |
These rows should not be added as if every firm pays the same tariff. A broker distributing data to customers may face a different cost profile from a prop firm using it internally.
Engineering and operations time also belongs in the model. It is usually tracked as staff capacity rather than an invoice line, but feed upgrades, incident response and entitlement audits consume real hours.
For an internal unit-cost measure, define a tick before dividing. Here, a tick means one incoming market-data message, such as a quote update or trade, not a minimum price increment.
Suppose the deployment costs US$18,000 a month across these categories and records 900 million normalised updates. That is about US$20 per million updates. It is useful for comparing venues or environments, but it is not an exchange price and it does not replace the licence basis.
- Record raw messages at the feed handler and normalised updates after deduplication.
- Report both average and peak rates, since bursts set network and CPU requirements.
- Allocate fixed charges separately from volume-sensitive costs.
- Keep a second view by application or desk, so unusually expensive consumers are visible.
For a broader view of platform and provider decisions, Trading Technology's infrastructure coverage follows the systems brokers and prop firms build around their feeds.
Week five: test the path under load
Feed capacity is not the same as feed availability. A connection can remain up while messages are delayed, dropped or processed out of sequence.
Instrument the full path: exchange or vendor timestamp, receive timestamp, normaliser output and consumer receipt. Use a monotonic clock for local intervals, and document clock synchronisation for cross-system comparisons.
Replay peak periods against production-sized hardware. Measure packet loss, sequence gaps, recovery time, queue depth and the time from socket receipt to consumer delivery.
A normaliser that handles 150,000 messages per second on average may still fall behind during a burst of 500,000. Those are illustrative rates; use the venue's published peak guidance and your own capture data to size the system.
Run these checks before launch:
- Drop a primary path and verify the recovery feed closes gaps without duplicating trades.
- Restart a handler during a busy period and check recovery from the last valid sequence number.
- Compare top-of-book and depth snapshots across primary and backup sources.
- Alert on stale data, growing queues and sequence gaps, not only a disconnected socket.
Capture enough raw data to investigate a dispute or reproduce a processing fault, subject to retention and licensing limits. If raw packet capture is not permitted or practical, agree what application-level records can be retained. Agree retention with compliance and the relevant data provider before production.
Week six: reconcile the invoice and the tick count
At go-live, compare contracted entitlements with actual consumers. A new strategy, a shadow deployment or a disaster-recovery site can change the declared use. Reconcile exchange reports, vendor usage reports and internal inventory monthly. Keep an owner for each entitlement and an expiry or review date for temporary access.
Then compare the planned cost model with the live one. Check whether the busiest venue consumes most compute, whether a backup path has added recurring charges, and whether storage is retaining more data than required.
If the monthly total is higher than expected, identify the line item before reducing depth or switching vendors. Cutting the wrong feed can save a small fee and create a much larger execution or incident cost.
Track at least these operational measures alongside spend:
- Cost per million normalised updates, with fixed and variable costs shown separately.
- Peak messages per second, handler CPU and maximum queue depth.
- Gap count, recovery duration and stale-data incidents by venue.
- Licensed consumers compared with active consumers and deployed applications.
The mistake most readers will still make is treating the exchange subscription as the cost of a tick, while leaving entitlements, duplicate paths and the systems that process each update outside the calculation.
Perguntas frequentes
What does a tick mean in market data?
In a market-data feed, a tick commonly means a message such as a quote update or trade. It does not mean a minimum price increment. Define the unit in monitoring and cost reports, because feeds and handlers may count messages differently.
How much do market data feeds cost a broker?
There is no single tariff. A modest multi-venue deployment can have monthly costs from a few thousand US dollars to tens of thousands or more, depending on exchange rights, professional users, non-display use, depth and connectivity. Treat published planning ranges as estimates and confirm charges with the exchange or authorised vendor.
Should a prop firm use direct feeds or a data vendor?
A vendor can simplify connectivity and normalisation across venues. Direct feeds give more control over delivery and processing, but require more engineering for protocol changes, recovery and monitoring. Compare total operating cost and licensing terms, not latency or subscription price alone.
