2ms WAN Connection: Are Exchanges "Slowing Down the Accelerator for Quant Trading"?

Fri, 31 Jul 2026 20:02:00 +0800

The “Technical Requirements for Wide Area Network Trading and Market Data Lines” recently discussed in the market, the most striking sentence is: the wide area network lines used for trading and market data services must have a bidirectional latency of no less than 2ms. According to reports, the original local area network trading and market data lines within the data center will also be phased out, and access will be uniformly transitioned to wide area network lines.

This sounds like slapping a speed limit on the fastest car. But it would be a stretch to conclude that “quant trading is banned” or that “retail traders now move at the same speed as institutions.” It’s more of a reordering at the access layer: replacing what were originally physical shortcuts potentially created by colocation with a manageable, dedicated wide-area channel that all participants must use. It will change a small class of strategies that are extremely sensitive to microseconds and milliseconds, but it won’t rewrite matching priority, nor will it eliminate the differences along the entire trading chain.

First, Separate the Evidence into Two Layers

As of July 31, 2026, the original exchange texts that can be publicly found and the market reports must be viewed separately.

The first layer is the explicit rules. On May 29, the Shenzhen Stock Exchange (SZSE) issued the Technical Specifications for Members and Other Relevant Entities Accessing SZSE Trading System Access Services (Ver1.06), which clearly states that this revision adds requirements for wide-area network (WAN) circuits and takes effect from the date of issuance. The revision history attached shows that Ver1.06 added the WAN circuit requirements on May 23. It is not just about speed: members are required to have at least two centralized operation centers that can back each other up, using circuits from different operators (including hosted access) to connect to the primary and backup data centers respectively; terrestrial lines must meet the minimum bandwidth, use Layer 3 routing ports, and employ high-speed one-way satellite as a backup for terrestrial market data. The type, latency, and bandwidth of the circuits must also meet the relevant technical requirements of the exchange. SZSE Announcement Original Technical Specifications

The second layer was directional notification reports at the end of July. Yicai reported that brokerage officials confirmed receiving the “Technical Requirements for Wide Area Network Trading Market Data Lines,” stating that bidirectional latency must not be lower than 2ms, applicable to existing and new lines; National Business Daily subsequently cited officials from major brokerages stating that both the Shanghai and Shenzhen stock exchanges had similar requirements. Yicai Report National Business Daily Report

As of the time of this article’s verification, the publicly available full text from the exchange specifying the “2 ms” figure, local network circuit cutover time, test methodology, and scope of application has not been found. Therefore, the most prudent statement is: The public technical specifications have confirmed the upgrade of the wide-area network access management; the “2 ms lower bound” and the specific cutover arrangements are media reports confirmed via brokers, and the detailed rules should still be subject to the exchange’s subsequent public or targeted documents. It is not appropriate to independently fill in “which exchange, which type of Level-2 market data, at which measurement point, and whether tested by mean or quantile.”

What exactly does this requirement change about which section of the road

Roughly breaking down a programmatic order, it generally goes through the following stages: “receiving market data — local computation — risk control/gateway — network — exchange access — matching — response.” The so-called local-area access and wide-area access mainly affect the segment “from the institutional access end to the exchange access end,” rather than the price-priority and time-priority rules within the matching engine.

Access Method	Intuitive Understanding	Relevance to This Change
Co-location in Data Center	The equipment and the exchange’s access facilities are in the same data center or very close together, resulting in a very short physical path	May produce extremely low transmission latency, and is reportedly the target of the adjustment
Dedicated WAN Line via Carrier	Access to the exchange network through controlled dedicated lines, facilitating management according to dual-path, carrier, and security requirements	Explicitly covered in the public specifications as an access form; reportedly to become a unified channel
Exchange Matching System	Matches orders entering the system according to established rules	Itself has not been reported as being changed due to “2ms”

Therefore, the statement that “the bidirectional latency must not be lower than 2ms,” if taken literally from the report, sets a lower bound on the round-trip latency of the line, rather than guaranteeing that every user’s end-to-end trade round-trip will be exactly 2ms. The actual end-to-end latency will also include additional factors such as the carrier’s routing path, quote decoding, strategy computation, risk checks, queuing, the trading gateway, and matching engine load. More importantly, latency is not a constant: a good-looking average does not mean the p99 or p99.9 tail latency during high-volatility market conditions also looks good.

The wide area network (WAN) is not the same as the public internet. The SSE Tech service page describes the market data WAN access as UDP market data access provided by carrier dedicated lines; its shared trading and market data access service for Shenzhen Communication users also emphasizes dual A/B route dual-active redundancy for both trading and market data. Market Data WAN Access Service WAN Shared Trading and Market Data Access Service

This also explains the dual context of the adjustment: on one hand, operational resilience and security—dual centers, different operators, bandwidth management, and active/standby failover are all fundamental capabilities of trading infrastructure; on the other hand, it is about fair access—if some participants can take shorter LAN paths within the same data center while others can only take longer paths, speed itself becomes a scarce resource. Unifying onto the wide-area channel with the lowest latency compresses these extreme advantages that arise directly from physical location.

It Is Not a Sequel to the “Quantitative Trading Ban”

Placing it back within the existing regulatory context makes its positioning clearer. The CSRC’s Regulations on the Administration of Programmatic Trading in Securities Markets (Trial) has been in effect since October 2024. It covers “report first, trade later,” real-time monitoring, information systems and co-location management, and provides differentiated arrangements for high-frequency trading; the Shanghai and Shenzhen stock exchanges subsequently refined reporting, conduct, information system, and high-frequency trading management through implementing rules. CSRC announcement Shanghai Stock Exchange implementing rules interpretation

Line requirements share the same direction as these rules, but the tools differ. The former focuses on who connects under what conditions, whether the connection is safe and controllable, and whether there are excessively disparate physical shortcuts; the latter also focuses on declaration, order cancellation, abnormal behavior, reporting, and risk control. One cannot simply equate line modification with a ban on quantitative trading, or with making a profit judgment about a certain strategy, simply because both objectives involve fairness and order.

For investors, the more accurate conclusion is this: it may erode part of the speed advantage that relies on extremely short physical paths, but it does not change the simplistic narrative that “being faster, having more servers, and having more data inevitably makes trades easier to execute.” That narrative was never true in the first place. Whether an order gets filled also depends on price, quantity, the moment it enters the matching queue, cancel and risk-control statuses, and whether the market itself has counterparty liquidity.

“Speed fairness” is also regulated overseas, but usually without a unified millisecond lower limit

Outside the territory, the focus is not on tolerating colocation and low latency per se, but rather on breaking the issue down into three questions: “Can colocation be provided,” “Can similar users obtain the same conditions,” and “Are the conditions and fees transparent.” Compared with the unified wide-area circuits and latency floors reported in the press, the regulatory tools are not the same.

Market	Comparable Arrangement	Key Differences from This Reported Arrangement
EU	EU 2017/573 accompanying MiFID II requires colocation services to be fair and non-discriminatory; users of comparable services must be subject to identical conditions regarding cabinets, power supply, cooling, cable length, data and market connectivity, and connections and latency must be monitored; trading venues must also publicly disclose the different latency types they can provide and their space allocation procedures	Regulation focuses on “comparable services under comparable conditions, auditable and transparent,” rather than requiring all users’ lines to be uniformly slowed to a certain minimum value
US	Colocation services are permitted. SEC filings acknowledge that, within the same or adjacent data centers, advanced participants may achieve access latencies far below 1 ms; intentionally imposed transient delays must be handled through rule change procedures and are subject to unfair discrimination review	There is no universal minimum latency floor; low latency may exist as a purchasable service, but exchange services and rule changes are subject to regulatory scrutiny

The EU rules are quite representative in their level of detail: not only do they require the same conditions for users within a single class of co-location service, but also require the trading venue to monitor connections and latency measurements, and publish service lists, fees, access conditions, the different types of latency available, and the procedures for space allocation. EU 2017/573

The U.S. system does not equate to “co-located users get a different set of matching rules.” A NYSE American filing with the SEC regarding co-location services states that both co-located and non-co-located orders enter the trading and execution system through the same order gateway, and co-located users do not receive market data products that other participants cannot get at all; however, the filing also acknowledges that co-location typically results in lower order and market data transmission latency. NYSE American filing The relevant SEC staff guidance discusses intentional delays of less than 1ms within the specific context of Regulation NMS automated quotations, and specifically notes that it is not an official rule. SEC guidance

Therefore, there are “similar issue” rules overseas, yet it is hard to find an “exactly identical” rule. The EU emphasizes equal conditions and transparency, while the U.S. permits co-location but examines non-discrimination and rule changes; the practice presented in this domestic report is more like directly flattening a section of the shortest physical path within the access architecture. All three attempt to constrain unfair access advantages, but the boundaries are drawn in different places.

The Impact on Quantitative Trading Will Emerge in Layers

What is most affected is not “everyone who trades with code,” but strategies that treat extremely low latency itself as a core input. You can roughly divide it into three layers:

Ultra-short-cycle strategies that heavily rely on the market-quote-to-order close loop: For example, strategies that react instantly to tiny changes in the order book, or that depend on cancellation speed or cross-channel sequencing, are most in need of re-evaluation. If the original edge came from intra-co-location LAN paths, the ~2ms wire-floor increase and greater jitter can alter signal lifetime, queue position, and break-even points.
Market-making and short-cycle execution strategies: The impact depends on whether they originally relied on this access path. Risk lies not only in higher average latency, but also in tail latency, packet loss, primary/backup failover, and changes in the timing of quotes and acknowledgements. If models still treat the acknowledgement arrival order of the old environment as ground truth, they can easily become distorted in live trading.
Mid/low-frequency stock selection, ETF allocation, and slower intraday execution: The direct impact is usually small. Their main sources of risk and return are more likely to be signal quality, trading costs, impact costs, and execution discipline, rather than a few milliseconds.

For quantitative teams, the most practical action is not to rush to add 2ms to every backtest parameter, but rather, after the access details are clarified, to redo end-to-end measurements: separately record market data arrival, strategy decision, risk control approval, order submission, exchange response, and primary-backup switchover; and look at the median, p99/p99.9, out-of-order packets, and packet loss, rather than focusing on a single ping. Then replace the fixed latency assumptions in the backtest with the measured distribution, and incorporate abnormal market conditions and failover scenarios into drills.

Finally, a plain boundary must also be retained: a more equitable route does not mean market outcomes are automatically more equitable. System access, broker internal controls, market data licensing, order routing, capital, and R&D capabilities will still cause differences. Route upgrades can address one segment of differences that is most easily amplified by physical location; whether they achieve their fairness and stability goals still depends on scope of application, consistency of execution, monitoring methods, and subsequent public disclosure.

References

写作附记

Original Prompt

$blog-writer China’s mainland exchanges have an announcement: Technical Requirements for Wide Area Network Trading Market Data Lines. Explain the background and meaning of this announcement in detail, whether there are similar regulations outside of China, and the impact on quantitative trading.

Exchange on Uncle Xiang's Notebook