Exchange Fairness and Access
Prerequisites
- matching-engine-system-design — sequencer, price-time priority
- trading-fundamentals — market makers, spreads, the timing problem
- order-books — CLOB mechanics
“Fairness” in financial markets has at least three distinct meanings that must not be conflated. A matching engine can guarantee deterministic priority (layer 1), but cannot guarantee equal access (layer 2) or cross-venue fairness (layer 3) — those require infrastructure policy and regulation respectively.
Layer 1: Engine-level deterministic priority
The matching engine’s sequencer assigns a monotonically increasing sequence number to every incoming message, creating a total order. Orders are processed in strict sequence: whoever arrives at the sequencer first wins. Given the same inputs, the engine produces the same outputs — determinism guarantees auditability and reproducibility.
This is the only layer the matching engine controls directly. It is fair in the sense that no participant receives preferential treatment from the engine itself — the rules are identical for everyone.
But “first to arrive at the sequencer” is a function of physical proximity and network investment, not of trading intent.
Layer 2: Access fairness — co-location and speed bumps
Co-location
Exchanges sell co-location services: participants pay to place their servers in the same data center (often the same rack row) as the exchange’s matching engine. The structural latency advantage:
| Location | Round-trip to sequencer |
|---|---|
| Co-located (same rack) | ~1 microsecond |
| Same data center, different cage | ~5-10 microseconds |
| Same city, different building | ~100-500 microseconds |
| Cross-country (NY to Chicago) | ~4-8 milliseconds |
| Retail trader (home internet) | ~5-50 milliseconds |
The sequencer is fair — whoever arrives first wins. But “first” is determined by physics and capital expenditure. A co-located participant has a 10,000x–50,000x latency advantage over a retail trader. Price-time priority, under co-location, becomes an arms race for physical proximity and network technology.
IEX’s speed bump
IEX (Investors Exchange), founded in 2012 by Brad Katsuyama (subject of Michael Lewis’s Flash Boys), introduced a 350-microsecond intentional delay on all inbound and outbound communications. The delay is implemented as 38 miles of coiled fiber optic cable — a physical device called the “magic shoe box” — through which all messages must pass.
What the speed bump does: Even if an HFT firm is co-located with IEX, their orders are delayed by 350 microseconds — the same delay experienced by remote participants. This equalizes the latency advantage of co-location for the specific attack it targets.
What it targets — latency arbitrage: An HFT firm detects a price move on Exchange A (e.g., NASDAQ). It immediately sends an order to Exchange B (e.g., NYSE) to trade against a market maker’s stale quote — a quote that hasn’t been updated yet because the market maker’s cancel message is still in transit. The 350μs delay gives market makers enough time to cancel their quotes on IEX after a price move is detected elsewhere.
SEC approval (2016): The SEC approved IEX as a national exchange, ruling that a sub-millisecond intentional delay is “de minimis” (negligibly small) and does not violate Regulation NMS’s “immediate and automatic” access requirement. This was controversial — opponents argued that any intentional delay violates the principle of immediate access; proponents argued that co-location itself creates unequal access.
Other speed bump implementations
After IEX’s approval, several venues adopted variants:
- NYSE American (formerly AMEX): introduced a delay for certain order types
- TSX Alpha (Toronto): 1-3ms randomized delay on aggressive orders
- Cboe EDGA: experimented with asymmetric delays (delaying only certain aggressive order types while allowing passive orders through immediately)
Layer 3: Regulatory fairness — Reg NMS and the NBBO
Regulation NMS (2005)
Regulation NMS (National Market System) is the SEC’s framework for ensuring fair competition among US equity exchanges. Its most consequential provision:
Rule 611 — the Order Protection Rule: Exchanges must route orders to the venue displaying the NBBO (National Best Bid and Offer — the best available buy and sell prices across all 16 US equity exchanges) rather than executing at an inferior price.
Example: if NASDAQ shows a bid of $100.02 and NYSE shows a bid of $100.03, a sell order arriving at NASDAQ must be routed to NYSE (where the seller gets the better price) rather than executed at NASDAQ’s inferior bid.
Consequences of Rule 611
Market fragmentation: The US equity market is split across ~16 exchanges (NYSE, NASDAQ, Cboe BZX/BYX/EDGX/EDGA, IEX, MEMX, LTSE, etc.) plus dozens of dark pools. Orders flow to whichever venue displays the best price at any given instant.
Latency arbitrage opportunity: The NBBO is a distributed concept — it takes time for a price change on one venue to propagate to all others. During that propagation window (microseconds to milliseconds), the NBBO on other venues is stale. HFT firms exploit this staleness:
- Detect a price move on Exchange A
- Race to Exchange B to trade against quotes that haven’t yet been updated
- Profit from the difference between the new true price and the stale quote
This is the same attack IEX’s speed bump targets. Rule 611 inadvertently created it by requiring fragmentation across venues with non-zero inter-venue latency.
SIP vs. direct feeds
The SIP (Securities Information Processor) is the official NBBO calculator — it aggregates quotes from all exchanges and publishes a consolidated view. But the SIP is slower than direct feeds (exchange-specific data feeds that co-located participants receive directly). An HFT firm with direct feeds from all exchanges computes its own NBBO faster than the official SIP, creating an information advantage.
Summary: what each layer provides
| Layer | Provided by | Guarantees |
|---|---|---|
| Deterministic priority | Matching engine (sequencer) | Same inputs → same outputs; no preferential treatment |
| Access fairness | Exchange policy (co-location rules, speed bumps) | Reduced latency disparity between participants |
| Cross-venue fairness | Regulation (Reg NMS, Rule 611) | Best price available regardless of venue |
A well-designed matching engine provides layer 1. Layers 2 and 3 require decisions outside the engine’s scope — infrastructure policy and government regulation respectively.
Questions to sit with
- IEX’s speed bump targets latency arbitrage by delaying everyone equally. But does equal delay actually create equal access, or does it just shift the arms race from “fastest network” to “fastest signal detection”? What advantage remains for co-located firms on IEX?
- Reg NMS assumes that “best price” (lowest ask, highest bid) is the right metric for fairness. But a market maker quoting the best price for 100 shares is not equivalent to one quoting it for 10,000 shares. Should size be part of the fairness equation?
See also
- matching-engine-system-design — how the sequencer creates deterministic ordering
- trading-venues — venue landscape including dark pools and ATS
- flash-crash-2010 — how market structure enabled the crash
- Price-time vs pro-rata — how the matching algorithm itself changes incentives