Exchange vs. Consolidated Market Data

Direct Answer

The US equity market operates as a fragmented system of 16+ individual exchanges plus FINRA ADF venues. Market data is distributed through two channels. Direct feeds (also called proprietary feeds) come directly from each exchange's matching engine with minimal processing delay, measured in microseconds to low milliseconds. Consolidated SIP feeds are produced by the Securities Information Processors (CTA for NYSE-listed securities, UTP for Nasdaq-listed securities) by aggregating, normalizing, and distributing data from all exchanges, adding 1-10 ms of additional latency versus direct feeds.

The SIP is the legally mandated baseline: broker-dealers must have access to SIP data to comply with best-execution rules, and the NBBO derived from the SIP is the regulatory reference price for execution quality assessment. Direct feeds provide faster, more granular data, including per-exchange order book depth and pre-trade indicators, but require separate connections, subscriptions, and software to handle each exchange's proprietary format. For any strategy operating on timeframes of one second or longer, the SIP is sufficient and significantly cheaper.

Key Takeaways

• CTA covers NYSE-listed, UTP covers Nasdaq-listed: Which SIP applies depends on where the security is listed, not where you are trading it. An NYSE-listed stock traded on Nasdaq's exchange still flows through CTA.
• SIP latency has dropped significantly: Following SEC-mandated infrastructure upgrades and competitive pressure, SIP latency from major venues to the SIP processor is now typically 1-5 ms, down from 10-30 ms a decade ago. Direct feeds add another few hundred microseconds of advantage at best co-location.
• Direct feeds require parsing multiple proprietary formats: Each exchange uses its own binary protocol (ITCH for Nasdaq, OPRA for options, FAST for CME). Consuming direct feeds from all 16 equity exchanges requires separate parser implementations and ongoing maintenance as formats evolve.
• The SIP NBBO is the regulatory reference price: Best-execution compliance, payment for order flow disclosures, and trade-through rule compliance all reference the SIP NBBO, not a direct-feed-derived NBBO. The SIP price is the standard.
• Direct feeds contain data not in the SIP: Individual exchange depth (beyond the NBBO), imbalance indicators for open/close auctions, order type distribution, and per-venue statistics are only available in direct feeds. The SIP delivers only consolidated trade and NBBO quote data.
• Most systematic strategies do not need direct feeds: End-of-day, daily/weekly rebalancing, and intraday strategies operating on 1-minute or longer bars have no meaningful use for the 1-5 ms latency advantage of direct feeds. The complexity and cost are not justified.
• Backtesting with SIP data is valid for most strategies: If your live system uses the SIP, backtest with SIP data. If you backtest with SIP data but then run live with direct feeds (or vice versa), you introduce a systematic measurement inconsistency between research and production.
• Cross-exchange arbitrage requires direct feeds: Strategies that exploit momentary price differences between exchanges need direct feed data from each exchange to detect and act on the discrepancy before the SIP publishes the update.

Core Concepts

The SIP Architecture: CTA and UTP

The Securities Information Processor system was established under Regulation NMS (National Market System) to ensure that all investors have access to consolidated best-bid-and-offer data and trade prints. The system is split by listing venue: the Consolidated Tape Association (CTA) Plan administers the tape for NYSE-listed securities (Tape A and Tape B); the Unlisted Trading Privileges (UTP) Plan administers the tape for Nasdaq-listed securities (Tape C). Both plans are joint-industry mechanisms governed by the exchanges themselves, overseen by the SEC.

Each exchange sends its trade prints and quote updates to the relevant SIP processor (CTA or UTP). The SIP processor consolidates these inputs, computes the NBBO, applies sequence numbers, and redistributes the consolidated stream to subscribers. Data vendors receive the SIP feed and redistribute it to their customers. End-to-end, a trade that occurs at NYSE Primary in New Jersey travels through the SIP processor (also in New Jersey) and arrives at a typical institutional co-location facility in 2-8 ms after the exchange timestamp.

The SIP infrastructure was significantly upgraded following the SEC's 2020 Market Data Infrastructure Rule (Rule 614). This rule required the SIP processors to offer competing consolidators, meaning third-party vendors can now aggregate and consolidate exchange feeds on behalf of subscribers, potentially achieving lower latency than the legacy SIP. The rule also expanded the data included in the SIP to include odd-lot quotes and depth of book information from exchanges at specified price levels.

For practical purposes, most data vendors provide SIP data as their standard offering. When a vendor quotes you "real-time Level 1 US equity data," they almost certainly mean SIP-derived consolidated data unless they explicitly state otherwise.

Direct Exchange Feeds: Structure and Content

Each US equity exchange publishes its own proprietary market data feed directly from its matching engine. These feeds deliver data faster than the SIP by eliminating the aggregation step. They also deliver data that is not included in the SIP: per-exchange order book depth, auction imbalance indicators, open/close reference prices, and proprietary statistics.

Each exchange uses its own binary message format. Nasdaq uses ITCH 5.0, a simple sequential message stream with messages for order add, cancel, replace, and execution. NYSE uses an Integrated Feed (XDP). CBOE uses its own binary format across its four equity exchanges (BZX, BYX, EDGX, EDGA). To consume all US equity exchanges via direct feeds, a firm needs parsers for approximately eight distinct protocols, plus the bandwidth and infrastructure to receive and process them simultaneously.

Direct feed connections are typically established via co-location at the exchange data center. Each exchange offers co-location services where trading firms rent rack space physically adjacent to the exchange's matching engine servers, minimizing network distance. At Nasdaq's data center in Carteret, NJ, a co-located server receives Nasdaq direct feed data approximately 150-300 microseconds after the matching engine processes an order. Via the SIP, the same event arrives 1,000-5,000 microseconds later.

The practical content advantage of direct feeds is in auction data. The NYSE's designated market maker and the opening/closing auction processes generate imbalance indicators, showing how many shares are on the buy side versus the sell side at various reference prices, that are published only on the direct feed, not on the SIP. Firms that trade at the close (which accounts for a substantial fraction of daily volume in index-linked products) use these direct-feed imbalance indicators to calibrate closing order submission.

NBBO Construction from Direct Feeds

Firms that subscribe to direct feeds from all exchanges can construct their own NBBO, often called a "proprietary NBBO" or "direct-feed NBBO", that reflects the true best bid and offer 1-5 ms before the SIP NBBO updates. This proprietary NBBO is the basis for HFT strategies that trade against stale SIP-dependent orders.

The construction process works as follows: maintain a local order book for each exchange, updated in real time from each exchange's direct feed. Compute the best bid across all exchange books and the best offer across all exchange books. This is your NBBO at a given moment. When any exchange updates its best bid or offer, recompute and update your NBBO. This requires low-latency processing of messages from all exchanges simultaneously.

The complexity and cost of maintaining a proprietary NBBO are substantial. Beyond the infrastructure, there is a timing issue: messages from different exchanges arrive at slightly different times even when processed in parallel. When Exchange A updates its quote at timestamp T, and Exchange B's quote update for the same symbol arrives at T+100 microseconds, your proprietary NBBO at time T reflects a mix of simultaneous and slightly stale inputs. Handling this correctly requires careful sequence number management and event ordering logic.

Regulatorily, the SIP NBBO remains the reference price for compliance purposes regardless of what your proprietary NBBO shows. A trade-through violation is assessed against the SIP NBBO, not the faster direct-feed NBBO. This distinction matters for strategies that route orders based on a proprietary NBBO that may be ahead of the SIP, the faster information can be used for routing but the SIP price is used for compliance assessment.

When Direct Feeds Are and Are Not Necessary

The practical question is not whether direct feeds are faster, they are, but whether the latency difference produces economically meaningful improvements for your strategy. For any strategy with a decision latency of 100 ms or more (e.g., a signal fires and an order is sent within 100 ms), the 1-5 ms SIP latency is a small fraction of the total pipeline. The alpha from faster data cannot be extracted within that timeframe, and the cost of direct feeds is not justified.

Strategies that require direct feeds are typically in specific categories: pure latency arbitrage (detecting price differences between venues in microseconds), co-located execution algorithms that must react to auction imbalance data, and market-making strategies that need per-exchange depth to compute optimal quote prices. These strategies are rare in the systematic trading community outside specialized HFT firms.

For the vast majority of quantitative strategies, including most mid-frequency (1-second to 1-day) systematic strategies, SIP data provides sufficient information. The additional cost and complexity of direct feeds (which typically run tens of thousands of dollars per month in exchange fees alone, plus infrastructure) is not recouped through improved strategy performance. The correct decision process is: identify the latency of your full order pipeline from signal to execution; if the SIP's 1-5 ms lag is more than 5% of that pipeline, consider direct feeds; otherwise, use the SIP.

Worked Scenario

A quant team is building a pairs trading strategy on two correlated ETFs. They need to decide between SIP data and direct feeds. Here is their evaluation:

1. Signal latency: Their pairs signal fires when the spread between the two ETFs deviates more than 2 standard deviations from a 20-day mean. The signal is computed every 5 seconds on closing prices of 5-second bars. Decision latency from signal detection to order submission is approximately 50 ms (Python signal computation plus order submission API call).
2. Relevant SIP latency: The 5-second bar uses the close price at the end of each 5-second window. The SIP delivers this price within 1-3 ms of the exchange timestamp. At 5-second bar resolution. This is 0.02-0.06% of the bar window, negligible.
3. Does faster data help? Their edge is the mean-reversion over 5-second windows, not speed of reaction. Even if they had direct feeds and reduced effective data latency to 100 microseconds, their signal would still compute on 5-second bars. The strategy is not competing for the first order after a price move. It is holding positions for seconds to minutes.
4. Decision: SIP data is sufficient. The team subscribes to a standard real-time SIP data vendor at approximately $200/month versus the $15,000+/month cost of direct feed co-location and exchange fees. They reinvest the difference in research capacity.
5. Exception check: The team notes that if they later want to execute at the close via MOC orders with better queue position, they would need direct access to NYSE's closing auction imbalance feed. This is a separate consideration from their signal data needs.

Measurement Framework

Common Failure Modes

Backtesting with Direct-Feed Data but Running Live on SIP

A team uses tick data reconstructed from direct exchange feeds for backtesting, achieving 100-microsecond precision, but their live system connects to a standard SIP data vendor at 3 ms latency. The backtest assumes signal computation at exchange timestamp; the live system computes the same signal 3 ms later. For a strategy whose edge depends on reacting within 10 ms, this 30% latency difference can cause the live system to miss fills or execute at worse prices than the backtest predicted.

The fix is to either use SIP-latency-corrected data in backtesting (add 3 ms to all data timestamps before computing signals) or to adopt direct feeds in production to match the backtest assumptions. Mixing data sources between research and production without correcting for latency differences is a systematic bias.

Assuming SIP Data Covers All Venues Equally

The SIP aggregates data from registered national exchanges, but it does not include trades from ATS (Alternative Trading Systems), dark pools, or internalized flow until those trades are reported. Internalized trades by broker-dealers are reported to FINRA ADF and appear in the tape, but with a delay of up to a few seconds. Dark pool executions appear in the tape after the fact but were never visible as quotes. A strategy that tracks whether all volume at a price was buy-initiated or sell-initiated cannot make this inference reliably from SIP data alone.

Conflating SIP Latency with Network Latency

The commonly cited "1-10 ms SIP latency" measures the delay from exchange matching engine to SIP processor. The total latency from market event to your application also includes: SIP-to-vendor distribution time, vendor-to-client delivery (typically TCP/IP over internet or co-location fiber), and your application's deserialization and processing time. For a client outside a data center, total end-to-end latency can be 20-50 ms even though the SIP processing step is only 1-3 ms.

Ignoring Exchange-Specific Quote Data for Routing

The SIP NBBO shows the best price across all exchanges, but it does not tell you which exchange is offering the most depth at that price. If Nasdaq is posting the best offer with 100 shares and CBOE EDGX is posting the same offer with 5,000 shares, a large order routed blindly to the cheapest venue based on NBBO may receive only a partial fill. Smart order routing (SOR) algorithms use per-exchange depth from direct feeds to route efficiently. Strategies that execute significant size without per-exchange depth data will incur unnecessary market impact.

Frequently Asked Questions

References

• SEC Release No. 34-90610: Market Data Infrastructure Rule (2020)
• CTA Plan: Consolidated Tape Association Plan documentation and fee schedules
• UTP Plan: Unlisted Trading Privileges Plan specifications
• SEC: Regulation NMS (Release No. 34-51808, 2005): Regulation NMS (original rule, including Rule 611 trade-through prohibition)
• Angel, J. J., Harris, L. E., & Spatt, C. S. (2011). "Equity Trading in the 21st Century." Quarterly Journal of Finance, 1(1), 1-53. (Market fragmentation and SIP data quality)

Educational Disclaimer

This guide describes US equity market data structure as of 2026. Exchange fees, SIP infrastructure, and regulatory requirements change. Verify current specifications and costs with your data vendor and the relevant exchange or SIP plan administrator.