GVIA Perspective | Xia Chun, Chief Economic Advisor of the Global Value Investment Association: Reflections on Quantitative Trading

Starting with the new regulation that exchanges shut down local area network (LAN) market data lines in colocation rooms on July 31, 2026, this article systematically sorts out the origins and evolution of quantitative trading. It first clarifies three commonly conflated concepts — quantitative, programmatic and high-frequency — then compares the different origins of quantitative trading in the US and China, and introduces major strategy categories including multi-factor, equity market neutral, CTA, high-frequency market making and arbitrage.

The article points out that the high-frequency "arms race" is an endogenous outcome of the continuous auction system, and that the "speed tax" is real but borne mainly by institutions. Meanwhile, big data research from the Shanghai Stock Exchange shows that the primary cause of retail investors’ losses is their own behavioral bias of chasing gains and cutting losses: retail investors would lose money even without quantitative trading.

The article argues that the July 31 regulation is about "equalization" rather than a "ban". Quantitative trading is neither an angel nor a devil: it provides liquidity and smooths volatility in normal times, but may amplify fluctuations in extreme market conditions. Strategy convergence is a rational, voluntary equilibrium choice for the industry.

The apparent long-term weakness of A-shares stems from a combination of multiple factors, including earnings fundamentals, policy guidance and industry trends, and is by no means driven by quantitative trading. The way forward for retail investors is to overcome behavioral biases and adhere to rational investing based on fundamentals and valuations, rather than looking for scapegoats.

On the evening of July 31, 2026, a batch of network cables in the colocation rooms of national securities and futures exchanges was officially disconnected. What was shut down were the "LAN market data lines" — a type of market data access available only to institutions with servers hosted in exchange rooms. Previously, if a quantitative private fund colocated its servers in an exchange room, its market data traveled over dedicated fiber-optic lines within the same campus, tens to hundreds of microseconds faster than external institutions. Starting August 1, everyone must use "WAN dedicated lines": from brokerage rooms to carrier rooms to exchange rooms, all market data now starts from the same point on the physical network.

What is a microsecond? One microsecond is one millionth of a second. A human blink takes about 400,000 microseconds (equivalent to 400 milliseconds; one millisecond is one thousandth of a second). For ordinary investors, a few hundred microseconds of speed advantage means nothing. But for high-frequency quantitative institutions, those hundreds of microseconds are the "VIP channel" they pay large sums to buy. Under the matching rules of "price priority, time priority", an order arriving even one microsecond earlier may execute ahead of others.

The news triggered widespread market debate. Some called it "pulling the plug on quant", a heavy regulatory crackdown on quantitative trading. Others argued it was merely a routine technical adjustment whose impact was grossly exaggerated. On August 3, the first trading day after the switch, A-share turnover reached 2.01 trillion yuan, about 550 billion yuan lower than the July daily average. Yet no one can say how much of that was caused by the new rules and how much by market sentiment.

To truly understand the significance of this event, a bigger set of questions must be answered: What exactly is quantitative trading? Where did it come from? How does it make money? Is it good or bad for markets and retail investors?

During my PhD studies in the Department of Economics at the University of Minnesota and my teaching career in the School of Finance at the University of Hong Kong, my research and teaching focused mainly on market microstructure (the foundation of high-frequency quantitative trading), hedge fund trading strategies and behavioral finance investing. Starting in 2015, I published multiple articles on the Financial Times Chinese website explaining quantitative trading strategies. After leaving academia, I have worked in the financial industry for more than ten years, but have never served as a consultant to any quantitative institution. I have no conflicts of interest in writing this article. Of course, given the complexity and rapid evolution of quantitative investing, this article may contain errors in understanding or expression, and I warmly welcome criticism and correction from professionals.

First, Distinguish Three Terms: Quantitative, Programmatic and High-Frequency

In public debate, these three terms are often used interchangeably, but they do not refer to the same thing.

Quantitative trading is an investment methodology: it uses mathematical models and computers instead of human judgment to make investment decisions. Instead of placing orders based on intuition that a stock will rise, portfolio managers codify investment logic into rules and models, letting computers identify patterns in massive datasets and execute trades. Its counterpart is "discretionary trading".

Programmatic trading is a method of trade execution: trading orders are automatically generated or submitted by computer programs. It is a tool, not a methodology. A fully discretionary fund manager can also use programs to execute large, manually determined order splits. Therefore, programmatic trading is not equivalent to quantitative trading, although the two overlap heavily.

High-frequency trading (HFT) is the most extreme category of programmatic trading: holding periods are measured in seconds or even microseconds, with tens of thousands of order submissions and cancellations per day, earning tiny spreads per trade and accumulating profits through massive volume. According to China’s regulatory criteria implemented in July 2025, a single account with more than 300 order submissions/cancellations per second, or more than 20,000 per day, qualifies as high-frequency trading.

It is worth noting that the widely circulated claim that "the US limits quantitative trading to no more than 15 orders per second" is a pervasive misconception. No US law or regulator has set such a hard limit for major exchanges. The figure applies only to alternative trading systems such as dark pools, as a reference metric for "disorderly trading", not a restrictive cap.

To summarize with an analogy: quantitative trading means "doing research with machines", programmatic trading means "placing orders with machines", and high-frequency trading means "placing orders rapidly and repeatedly with machines". Regulatory documents govern programmatic and high-frequency trading, while public debate about "exploiting retail investors" usually targets high-frequency trading. The three overlap but are not identical. In practice, almost all retail investors and even some financial experts have entirely blurred distinctions and definitions of quantitative trading.

Some of our financial institutions and research draw on US practices, and quantitative trading is no exception. To explain quantitative trading clearly, we must start with US institutions and experience.

Origins and Development of Quantitative Trading in the US

Quantitative trading in the US has two roots. The first grew out of academic ivory towers, giving rise to medium- and low-frequency quant. The second grew out of market structure, giving rise to high-frequency quant.

Start with the first root. In 1952, Harry Markowitz published modern portfolio theory, turning "diversification" into a mathematical problem for the first time. In the 1960s, the Capital Asset Pricing Model (CAPM) and Efficient Market Hypothesis (EMH) emerged one after another. Factor research by finance professors such as Eugene Fama showed the academic community that stock returns can be systematically explained and predicted — a large share of talent in quantitative trading comes from Fama’s academic lineage. In the 1970s, these theories gave birth to index funds and quantitative risk models. In 1982, mathematician James Simons founded Renaissance Technologies (predecessor: Monemetrics, founded in 1978). In 1988, computer scientist David Shaw founded D.E. Shaw, formally launching quantitative hedge funds onto the historical stage. This root produced medium- and low-frequency quant: using statistical models for stock selection, pricing and portfolio management.

It is particularly worth noting that in the first decade or so after founding his business, despite hiring countless top mathematicians to design futures trading strategies, Simons delivered highly unstable performance (detailed in the Chinese edition of The Man Who Solved the Market). Simons’ main talent lay in his superb interpersonal and business skills (he had a very unpleasant relationship with the mathematicians he hired) — every time he lost money, he always managed to find new investors who trusted him.

Renaissance Technologies’ later stellar performance relied mainly on three speech recognition experts from IBM, who in 1995 identified bugs in the code and improved the pairs trading strategy pioneered by D.E. Shaw. In the foreword to the Chinese edition of The Man Who Solved the Market, Liang Wenfeng, founder of Magic Quant, commented: "Simons and other pioneers, using techniques that do not seem complex today, quickly picked the lowest-hanging fruit in the market and accumulated their first pot of gold." Later media accounts attributing the fund’s outstanding performance to Simons’ mathematical ability are a major misconception. Of course, it is true that quantitative trading firms today hire PhDs and experts in mathematics, physics and engineering at high salaries — but the two should not be conflated.

Another major misconception concerns the actual performance of Renaissance’s flagship Medallion Fund, which is open only to insiders. In the foreword to the Chinese edition of The Man Who Solved the Market, Qiu Huiming, founder of Minghong Investment, writes: "Over the 30 years from 1988 to 2018, it delivered a 39.1% annualized compound return net of 5% management fee and 44% performance fee. The gross annualized return before fees would be around 66%. This return far exceeds those of better-known investment masters such as George Soros in macro investing and Warren Buffett in value investing."

This 66% gross annualized compound return has spread widely in Chinese-language circles, yet few have bothered to calculate: starting with the Medallion Fund’s initial $20 million, such performance would grow to $80 trillion in 30 years — more than the entire market capitalization of US stocks. Is that plausible? The error lies in the term "annualized compound return". In fact, the author, a Wall Street Journal reporter, clearly states in the book that it is the "average annual return" (i.e. the arithmetic mean of annual returns, see page 392 of the Chinese edition). Since the Medallion Fund caps its size every year (returning profits to investors), simply compounding annual returns geometrically produces a huge error.

In my article Myth and Reality: Did Simons Significantly Outperform Buffett?, I recalculated the Medallion Fund’s performance using the correct method. The conclusion is that for the same initial investment amount, the long-term net-of-fee return from Buffett’s Berkshire Hathaway likely exceeds that of Simons’ Medallion Fund (in fact, the two funds Renaissance offers publicly have delivered unremarkable performance).

This case shows that even within China’s professional financial circle, there are widespread misconceptions about quantitative investing — let alone retail investors.

Now turn to the second root of US quantitative trading, embedded in market structure. The US stock market is fragmented: for historical reasons, the same stock can trade on more than a dozen exchanges and over 30 dark pools simultaneously. In 2001, US stock quotes switched from fractions to decimals, shrinking the minimum tick size to $0.01. Bid-ask spreads narrowed sharply, human market makers were pushed out, and machines took over quoting. The implementation of Reg NMS in 2005 required every trade to execute at the "national best bid and offer". This investor-protection rule, in turn, created huge technical demand: someone had to monitor quotes across all markets at extreme speed and arbitrage away any discrepancies the moment they appeared. This root produced high-frequency trading: speed itself is the source of profit.

Origins and Development of Quantitative Trading in China

Quantitative trading started much later in China, driven by entirely different forces. In April 2010, the CSI 300 stock index futures were launched, giving China its first short-selling hedging tool. Without it, "market neutral" strategies — buying stocks while shorting futures to hedge out market exposure and capture only stock-selection alpha — would have been impossible. In the same year, the CTP trading system developed by Shanghai Futures Information Technology opened its full-featured API to the market, allowing any team to access futures exchanges for programmatic trading at low cost.

But the most fundamental driver was the "immaturity" of the Chinese market itself. A-shares have a high share of retail trading, with overconfidence and frequent turnover (every buy and sell incurs commissions and stamp duty, and also contributes order flow information to high-frequency strategies), chasing gains and cutting losses, lottery preference (favoring low-price, high-volatility, thematic stocks with a small chance of huge gains, fantasizing about getting rich overnight), and the disposition effect (selling winning stocks too early and holding losing stocks too long, taking profits but not cutting losses). These behavioral biases create massive pricing inefficiencies. In such a market, even the simplest multi-factor models — the inverse of low valuation, small market cap, high momentum and high turnover — can consistently capture excess returns. US quantitative trading emerged because "the market is too fragmented and needs machines to stitch it together"; Chinese quantitative trading emerged because "the market is too immature and machines find it easier to pick up profits".

Starting in 2010, China’s quantitative industry expanded at a staggering pace:

• In 2013, the Everbright Securities "fat finger" incident first introduced the power of programmatic trading to the public — a system glitch ignited a sharp market rally in an instant.

• In 2015, the firm Yishidun used illegal high-frequency tactics (concealing a group of controlled accounts and bypassing futures firms’ normal capital and position verification to connect unauthorized trading systems directly to exchange mainframes) to make nearly 400 million yuan illegally in stock index futures, proving that world-class speed players already existed in China’s market.

• In September 2021, A-share turnover exceeded 1 trillion yuan for dozens of consecutive trading days. Claims that "quantitative trading contributes half the volume" went viral, and the scale of quantitative private funds surpassed 1 trillion yuan.

• Starting in 2023, regulators began taking systematic action: the programmatic trading reporting system, penalties for abnormal trading in 2024, the official implementation of the Implementation Rules for the Administration of Programmatic Trading in July 2025, and finally the shutdown of the LAN data lines on July 31, 2026.

It should be noted that the July 2025 new rules are regarded by the industry as "standardizing" rather than "stifling" quantitative trading. Since the final rules had a transition period of over a year from consultation to implementation, most compliant leading quantitative private funds completed system adjustments and strategy optimization in advance, resulting in limited short-term market impact. The more far-reaching effect is that it has pushed the industry’s core strategies to shift from high-frequency strategies reliant on trading speed to medium- and low-frequency strategies focused on model R&D and fundamental analysis.

Quantitative Strategy Camps: How Do They Actually Make Money?

Quantitative strategies fall into two broad camps by trading frequency, with entirely different profit logic.

The first is the medium- and low-frequency camp, with main strategies including:

• Multi-factor stock selection / index enhancement: This is currently the largest quantitative strategy by scale in China. Models score thousands of stocks on dozens or hundreds of factors (valuation, growth, quality, sentiment, price-volume characteristics, etc.), buy a portfolio of high-scoring stocks, and aim to outperform the index. Index enhancement products pursue the goal of "rising more than the index when it goes up, falling less when it goes down". Holding periods range from days to weeks.

• Market neutral: On top of the above stock selection portfolio, stock index futures are used to hedge out market exposure, leaving pure stock-selection excess returns. Around 2021, this was the signature product of Chinese quantitative private funds.

• CTA (Commodity Trading Advisor): Trend-following or arbitrage strategies in futures markets, able to go long or short, with holding periods ranging from hours to weeks.

These strategies earn money from research and data: whoever has more effective factors, more accurate models and more refined portfolio optimization wins. They have very low reliance on speed. The line change after July 31 has almost no impact on annualized returns.

The second is the high-frequency camp, with main strategies including:

• Market making: Placing orders on both the bid and ask sides simultaneously, buying low and selling high to capture the bid-ask spread while providing liquidity to the market. This is the dominant high-frequency strategy in futures markets, and futures exchanges have dedicated schemes and fee waivers for market makers. In the stock market, due to T+1 settlement (shares bought the same day cannot be sold the same day), US-style stock market making is structurally impossible.

• Arbitrage: Cash-futures arbitrage between futures and spot, creation-redemption arbitrage between ETF primary and secondary markets, and cross-product spread arbitrage. Arbitrage windows often exist for only milliseconds to seconds — whoever arrives first captures the profit. For example, the price correlation between S&P 500 ETFs and E-mini futures contracts approaches 1 at minute, hourly and daily intervals, but drops to near zero at 250-millisecond intervals. This creates profit opportunities for high-frequency arbitrage strategies.

• Order book prediction (high-frequency alpha): Using order book data (order volume at each bid/ask level, cancellation speed, large order flow) to predict price direction over the next few seconds to minutes, and buying or selling in advance. Such strategies are known in China’s stock market as "high-frequency price-volume factors", essentially extracting information from other people’s order flow and trading ahead of them.

• T+0 trading: In the A-share market, strict T+0 applies only to ETFs, convertible bonds, stock index futures and other instruments. However, if an investor holds a base position of stocks, they can repeatedly buy low and sell high on the same batch of stocks intraday to capture spreads, circumventing the T+1 restriction. Retail investors with base positions can also execute such strategies. The popular claim that "institutions get T+0 while retail investors get T+1" is also inaccurate.

High-frequency strategies earn money from speed, infrastructure and market microstructure. They are extremely sensitive to latency, which is exactly why the July 31 data line change struck a nerve. It should be clarified that high-frequency trading accounts for a small share of the overall quantitative industry: official data shows programmatic trading accounts for about 29% of A-share turnover. Industry estimates put active trading by pure quantitative funds at 15%–25%, while high-frequency trading in the strict sense accounts for only single-digit percentages, concentrated mainly in futures markets.

Academic Consensus Once Favored High-Frequency Trading

Whether quantitative trading (especially HFT) is good or bad for markets has been debated in academia for two decades, producing a body of solid research. Ordinary readers do not need to read the papers, but should know the most important findings.

Starting in 2003, I delved into theoretical models of market microstructure and closely followed empirical research. It is fair to say that until the Flash Crash in US markets in May 2010, the basic academic consensus was that high-frequency quantitative trading, by buying low and selling high, increases market liquidity, narrows bid-ask spreads, reduces transaction costs, lowers volatility, makes prices react faster to information, and improves market efficiency. For both institutions and retail investors, HFT was seen as beneficial. One intuitive observation: exchanges used to be crowded with human market makers; as information technology was introduced and trading became electronic, exchanges grew empty, and investors’ transaction costs fell significantly.

When the 2008 financial crisis broke out, programmatic and high-frequency quantitative trading drew little explicit criticism. Five hedge fund managers who made large profits during the crisis — including Soros, Simons and Paulson — were summoned to testify before Congress, and received positive media coverage.

On August 7–9, 2007, US stocks suffered three consecutive days of "quant quake / black swan" events. Short-term reversal factor strategies that had worked for years suddenly failed en masse, and several major quantitative funds posted huge losses simultaneously. Goldman Sachs CFO David Viniar said, "We were seeing 25-standard-deviation moves, several days in a row." Yet the S&P 500 performed normally over the same period. By August 10, strategies returned to normal and losses were recovered by month-end. This event did not change attitudes toward quantitative strategies. The main academic reflection was on factor overcrowding and similarity. In other words, there were simply too many of Fama’s academic descendants.

But the May 2010 Flash Crash — in which the Dow Jones Industrial Average plunged nearly 1,000 points, about 9%, in minutes intraday before recovering most of the loss in just over 20 minutes — triggered a re-examination of existing theories on high-frequency trading. It took five years for US authorities to arrest the perpetrator, whose use of high-frequency spoofing techniques exacerbated the crash that day. He made $40 million in profits over five years, yet in January 2020 was sentenced to only one year of home confinement. The real consequence of this event was that US regulators revised multiple trading rules. It can be said that half of today’s stable framework for US equity markets is a legacy of that flash crash.

Theoretical Breakthrough: Trading Structures Endogenously Generate Speed Races and Worsen Liquidity

In 2015, University of Chicago economist Eric Budish and co-authors published a highly influential paper in the Quarterly Journal of Economics (QJE) (the working paper was completed in 2013), presenting two counterintuitive conclusions. First, the total profit from high-frequency arbitrage strategies is fixed, and the optimal strategy is a speed race rather than price competition. Second, the problem does not lie with high-frequency traders — it lies with the market rules themselves.

For example, the total annual profit from arbitrage between the S&P 500 ETF and E-mini futures contracts mentioned earlier hovers around $75 million, and does not decrease as competition intensifies. This runs completely counter to basic economic common sense, which holds that price competition drives economic profits to zero. The total prize pool of this fixed arbitrage opportunity is determined by trading volume and volatility, regardless of how many firms compete.

The speed race has not reduced the total profits of high-frequency traders, and investors have not actually benefited significantly from the HFT arms race. The improved market liquidity, narrower bid-ask spreads and lower overall costs that academia once cited as benefits of HFT came mainly from the continuous introduction of information technology by exchanges. It is hard to find direct evidence that high-frequency traders contributed to improved liquidity. Moreover, the benefits of IT adoption were concentrated in the late 1990s to mid-2000s, before high-frequency trading became widespread.

New entrants do not divide the pie more thinly and pass savings on to the market. Instead, they burn profits on more expensive servers and shorter data cables — a never-ending, socially unproductive "arms race". For example, in June 2010, just one month after the flash crash, Chicago telecom firm Spread Networks announced it had spent $300 million building a fiber-optic link connecting New York and Chicago financial markets, cutting round-trip message latency from 16 milliseconds to 13 milliseconds. The project drew widespread public criticism.

Yet even as society debated the pros and cons of quantitative trading, financial firms began replacing fiber optics with microwave communications (which travel through air only 1% slower than the speed of light in a vacuum). HFT speeds have moved from the millisecond and microsecond eras into the nanosecond (one-billionth of a second) era.

Arbitrage opportunities in global capital markets span all asset classes and markets. Budish et al. estimate that annual arbitrage profits in the US stock market alone amount to billions to $20 billion. This explains the strong appeal of HFT to participants. For other investors, however, HFT brings no obvious benefits. From a social perspective, the HFT speed race is a waste of resources.

Similar criticism has begun to emerge in China. Some of the country’s brightest minds are flowing from laboratories and frontline engineering into trading server rooms. Top graduates in mathematics, physics and computer science — who could be working on semiconductors, aerospace or basic research — are instead crowding into quantitative private funds, applying their talent to millisecond-level order book games. The scarcest intellectual resources are shifting from productive sectors to redistributive ones.

Budish et al.’s greatest insight is that the above characteristics of HFT markets, and the behavior of high-frequency traders, are actually caused by existing trading structures. Today’s major exchanges worldwide (including China) use "continuous double auction": time flows continuously, and orders are matched one by one in the order they arrive. They prove that as long as these two conditions hold, "speed arbitrage" is an endogenous product of the mechanism: when new information arrives, old quotes are technically outdated but have not yet been canceled. Whoever snatches those stale quotes first makes a guaranteed profit. This requires no insider information — it is purely a race for speed.

The biggest difference from the era of manual trading is that once trading becomes electronic, computers allow infinite subdivision of trading time. As a result, high-frequency traders engage in a speed race: milliseconds are too long, only microseconds offer an edge. The essence of arbitrage trading is buying low and selling high based on public information. If there were only one high-frequency trader, speed would not matter much, and its optimal strategy would be to act as a liquidity provider, earning profits from buying low and selling high.

But with multiple competing HFT firms, given that public information changes constantly and trading can proceed continuously with infinitely subdivided time under the current system, two options must be considered: either act as a liquidity provider, or act as a "sniper": when other HFTs’ quotes fail to keep up with public information changes and they attempt to cancel, use the speed race to "snipe" those quotes (buy low, sell high), hoping to execute before others’ cancellation instructions go through. The "spoofing" technique used by the perpetrator of the US flash crash, mentioned earlier, involves submitting prices with no intention of executing and then quickly canceling them, in order to disrupt other high-frequency traders’ strategies.

Because everyone can be sniped in a speed race, their optimal pricing strategy produces wider bid-ask spreads than would exist without a speed race, which in turn worsens market liquidity.

Budish et al.’s proposed remedy is to replace "continuous double auction" with "frequent batch auctions": first, slice time into discrete intervals (e.g. one trade every 100 milliseconds); second, aggregate all orders within the same interval and match them at a single uniform price, making a one-microsecond speed advantage worthless. In theory, two such minor changes — barely noticeable to most people — would completely eliminate "sniping" strategies. High-frequency traders would no longer compete on speed, but on price, genuinely boosting market liquidity.

Unfortunately, although this trading structure designed by Budish et al. received unanimous academic support, it was opposed by high-frequency traders and has been shelved to this day. Readers interested in the reasons can read my article Does High-Frequency Quantitative Trading Improve Market Liquidity?.

Key Empirical Research Findings

In recent years, empirical research on high-frequency quantitative trading has grown significantly. I select some important findings below. Attentive readers will notice conflicting conclusions across studies, largely due to different data sources, as well as academic journals’ tendency to publish divergent results. Research in China’s relevant fields is still emerging, and we look forward to more high-quality studies on China’s quantitative practices.

• The bill for the arms race has been quantified. In 2022, Budish and researchers from the UK Financial Conduct Authority published a quantitative study in the QJE. Using message-by-message data from UK exchanges (which captures even failed orders and cancellations, equivalent to seeing "all the losers in the race"), they found that FTSE 100 stocks experienced an average of 537 speed races per stock per day, with over 20% of trading volume involved. The "implicit tax" imposed by speed arbitrage on all investors is