AI Finance Knowledge Base on Uncle Xiang's Notebook

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.

Can Low Interest Rates Erase Debt: From RMB, Japan to Financial Repression

Mon, 27 Jul 2026 20:36:55 +0800

Is China using low interest rates to “resolve its debt”? This assessment is half straightforward: with low interest rates, debtors pay less in interest each year, and the rollover pressure on the government and local government financing vehicles eases. The other half, however, is easily overstated: low interest rates do not mean the principal of the debt has disappeared, nor do they automatically mean that household deposits are filling the government’s holes.

To understand this clearly, we first need to separate four things that are often conflated: the policy rate, the real rate, the exchange rate, and capital flows.

First, let me correct a premise: the RMB has not appreciated continuously over the past five years

If “appreciation” refers to the nominal exchange rate against the U.S. dollar, the past five years have not been a continuous line of appreciation, but rather have shown distinct periodic fluctuations. The RMB exchange rate against a basket of currencies is not the same indicator as the exchange rate against the U.S. dollar. Going one step further, the implications of exchange rates for trade competitiveness and purchasing power also depend on the inflation differential between China and other countries.

The IMF, in its 2025 Article IV consultation with China, emphasized a phenomenon that differs from the notion that “the RMB has been continuously appreciating”: because China’s inflation is lower than that of its trading partners, the real exchange rate has depreciated, which has helped boost export growth. The State Administration of Foreign Exchange’s 2025 Balance of Payments report showed a current account surplus of US$735 billion, accounting for 3.7% of GDP, and stated that the RMB is basically stable at a reasonable and equilibrium level. Together, they remind us that low interest rates, trade surpluses, and periodic currency strength can occur simultaneously, but this does not warrant the conclusion that “one policy deliberately engineered another outcome.”

Capital account management should not be simplistically translated as “blocking capital outflows.” China’s capital account is not fully freely convertible; cross-border funds are subject to authenticity, compliance, and macroprudential management requirements, which affect the speed and channels of capital outflows. However, current account surpluses also naturally generate external asset allocation. The State Administration of Foreign Exchange’s (SAFE) description for 2025 is precisely this: capital inflows formed by current account surpluses are converted into outward investments under the financial account. The existence of management does not mean that every external allocation is prohibited; it is more akin to a set of valves rather than a one-way plug.

How Low Interest Rates Help “Resolve Debt”

Start with the simplest layer: the interest bill on debt. In May 2025, the 1-year LPR fell to 3.0%, and the over-5-year LPR fell to 3.5%. For floating-rate loans, newly issued debt, and debt being refinanced at maturity, lower rates will gradually reduce financing costs. This is also the main role of local implicit debt swaps: replacing short-term, high-interest, cash-flow-pressured debt with longer-tenor, lower-cost, and clearer-credit debt. The IMF’s characterization of this round of implicit debt swaps—five trillion yuan over five years starting in 2024—is “easing refinancing pressure.”

But “alleviating the pressure” is not the same as “eliminating the debt.” When looking at the debt ratio, an approximate relationship is quite useful:

$$ \Delta d_t \approx \frac{r-g}{1+g}d_{t-1}-pb_t $$

Here, $d$ is the debt-to-GDP ratio, $r$ is the average nominal interest rate, $g$ is the nominal GDP growth rate, and $pb$ is the primary fiscal surplus (revenue minus non-interest expenditure, with surplus being positive). A low $r$ suppresses the first term; higher nominal growth $g$ and a sustained primary surplus $pb$ also help bring the debt ratio down. Conversely, if nominal growth is weak and the government runs persistent large deficits, low interest rates alone can only drag the problem out longer and smooth it over—they cannot make the arithmetic disappear.

This is also why real estate adjustments, local government finances, and low inflation make the problem harder: low inflation suppresses the nominal GDP growth rate, so the denominator no longer grows quickly; although bond yields are low, $r-g$ may not be favorable enough.

Are Depositors Being “Silently Devalued”

This requires distinguishing between low nominal interest rates and negative real interest rates:

$$ \text{Real Interest Rate} \approx \text{Nominal Deposit Rate} - \text{Inflation Rate} $$

If the deposit interest rate is 1% and inflation is 3%, the real purchasing power return for savers is approximately -2%; debtors repay with lower real costs, and the real value of government debt is also eroded by inflation. This is close to a mechanism of “using savers’ purchasing power to reduce historical debt.”

In academic literature, this entire set of institutional arrangements is often referred to as “financial repression.” The typical characteristics listed by Reinhart and Sbrancia include: constrained domestic funds absorbing government debt, explicit or implicit interest rate ceilings, restrictions on cross-border capital flows, and closer ties between the government and the banking system. Low nominal interest rates save on interest expenses; only when a negative real interest rate forms over the long term does the “liquidation” effect they describe occur, in which the real value of debt is eroded.

However, applying this label to China in recent years requires particular caution. First, low inflation or even falling prices raise real interest rates rather than produce persistently negative real rates; a decline in nominal deposit interest does not necessarily translate into a proportional decline in the purchasing power of deposits. Second, households hold more than deposits; housing, wealth-management products, equity assets, and employment income are also affected by macroeconomic adjustments, so the distributional consequences cannot be captured by deposit interest rates alone. Third, observing capital flow controls, a high share of state-owned banks, and low interest rates only shows that they resemble certain elements in the definition of “financial repression”; this does not prove that the sole intent of any given policy is to tax households.

There are also two related but distinct words:

Term	Core Meaning	Conclusion That Cannot Be Directly Drawn
Financial Repression	Using regulation, interest rates, capital allocation, etc., to keep domestic financing costs lower than what a free market might provide	Any interest rate cut constitutes financial repression
Debt Monetization	The central bank directly or indirectly and continuously provides monetary financing for fiscal deficits	A central bank rate cut or bank purchases of government bonds necessarily constitute monetization
Fiscal Dominance	Fiscal sustainability in turn constrains monetary policy, making it difficult for the central bank to act solely according to inflation targets	High debt alone means monetary policy independence has already been lost

Mixing up these three terms often turns reasonable questions into already-verified conclusions.

Japan Followed a Similar Path, but Not the Same Script

Japan is indeed a good reference for understanding persistently low interest rates. The Bank of Japan (BOJ) began its zero interest rate policy in 1999 and moved into quantitative easing in 2001; in 2016, it introduced negative interest rates and yield curve control (YCC), targeting the long-term rate near zero. In March 2024, the BOJ judged that negative interest rates and YCC had fulfilled their roles and ended this framework, returning the short-term rate to its primary policy instrument.

Japan’s starting point was low growth, deflationary pressure, and the repair of its banking system following the collapse of an asset bubble. Persistently low interest rates objectively lowered the cost for the Japanese government to roll over its massive government debt, and high domestic savings combined with domestic financial institutions holding Japanese government bonds also made its financing structure more stable. However, this is not sufficient evidence that “the Japanese government intentionally let savers lose value to reduce its debt.” Japan’s long-term inflation was very low, with deflation in many years, so negative real interest rates were not stable. A more accurate way to put it is that Japan demonstrated how a highly indebted economy can extend its adjustment period by relying on domestic-currency financing, low interest rates, and a strong base of domestic investors; it also demonstrated the costs—bank profitability, distorted asset allocation, fiscal dependence on low interest rates, and the difficulties of exiting monetary easing.

The commonality between China and Japan is that both face high savings rates, banking systems of significant importance, debt rollover pressures, and low interest rates. The differences are equally critical: China still has higher potential growth space, a greater degree of capital account management, a different exchange rate mechanism, and a greater weight of local governments and the real estate sector in the debt chain. Therefore, “China will become Japan” is a hypothesis that requires continuous testing, not a prediction that can be established by the single variable of low interest rates.

What sustains this arrangement, and where does it break down

If an economy uses low interest rates to help roll over debt over a long period, it needs at least five pivot points:

Sufficiently large domestic savings and investors willing to hold local-currency-denominated assets, avoiding excessive reliance on foreign debt.
The banking system, insurance, pension funds, and bond markets are able to absorb longer-term government and quasi-government debt while remaining well-capitalized and with identifiable risks.
Nominal growth ultimately exceeds the average cost of financing, or the fiscal position can achieve a sustainable improvement in the primary balance; this is more fundamental than simply “suppressing interest rates.”
A credible balance is maintained among inflation, the exchange rate, and capital flows. If interest rates are too low while exchange-rate expectations become destabilized, evasive outflows may be induced; the tighter the controls, the more transparent and predictable rules are needed to maintain confidence.
Beyond debt rollover, there must be genuine fiscal and balance-sheet adjustments: clarifying who bears legacy debt, shutting down financing vehicles that lack cash flow, and restoring local tax sources and social security—rather than repeatedly renaming old debt.

The warning signs are equally intuitive: nominal growth consistently falls short of financing costs, banks see their capital erode as they absorb the debt, depositors shift into avoidance channels, exchange-rate expectations become one-sided, or fiscal authorities rely solely on rollover rather than addressing unsustainable debt. At that point, low interest rates are no longer a buffer but instead expose the fragility of an economy that “cannot afford interest-rate normalization.”

So, the shortest answer to the original question is: low interest rates are indeed one of the tools available to defuse debt pressures, especially by reducing rollover costs; however, they are not an independent debt solution. A situation resembling “financial repression”–style debt dilution only emerges when sustained negative real interest rates, constrained funding absorption capacity, and a degree of inflation all coexist. China’s low inflation in recent years means instead that the continuous erosion of depositors’ purchasing power through inflation cannot be treated as an established fact. What really determines how far this path can go is not whether interest rates can be cut a little further, but whether the economy can return to nominal growth, whether fiscal policy can transform implicit debt into manageable explicit obligations, and whether households are still willing to hold domestic-currency financial assets.

References

People’s Bank of China: Chronology of China’s Monetary Policy in the First Three Quarters of 2025, accessed: 2026-07-27.
State Administration of Foreign Exchange: China’s International Balance of Payments Report 2025, accessed: 2026-07-27.
IMF: People’s Republic of China 2025 Article IV Consultation, accessed: 2026-07-27.
Reinhart & Sbrancia: The Liquidation of Government Debt, accessed: 2026-07-27.
Bank of Japan: Unconventional Monetary Policy since the Latter Half of the 1990s, accessed: 2026-07-27.
Bank of Japan: Adjustment of the Monetary Policy Framework in 2024, accessed: 2026-07-27.

写作附记

Writing Instructions

This article discusses macro-financial mechanisms and publicly available materials, and does not constitute investment, currency exchange, or asset allocation advice. Regarding policy intentions, only verifiable measures and their possible mechanisms are stated, and relevance is not equated with the sole motive.

Original Prompt

In the past five years, China has entered a low-interest-rate mode, but the RMB has been appreciating in tandem, while various measures have been taken to intercept capital outflows. Is China using lower interest rates to resolve government debt issues? Is it equivalent to allowing the slow depreciation of household deposits to resolve historical debt? Has Japan ever taken this path? What is this called in economics? Explain the corresponding basic economic knowledge and what is needed to sustain this model’s operation?