Quantitative investing never speaks for itself: it is an amplifier, not an ignition source.

Every time a market collapse occurs, the public always looks for a scapegoat — this is true not just in China, but also in the United States. Looking back at history, portfolio insurance was blamed in 1987, CDOs and credit rating agencies were the targets in 2008, high-frequency trading took the heat after the 2010 Flash Crash, and in 2026, quantitative trading became the scapegoat.

On June 26, the Shanghai Composite Index dropped 2.26%, the Shenzhen Component Index fell 3.44%, and the ChiNext Index slid 4.07%. Nearly 4,676 stocks across the entire market closed lower, and the phrase "quant funds smashing the market" immediately popped up in comment sections, just as it did after the market plunge on May 21. All these past scapegoats shared one common trait: they left highly visible footprints at the scene, making them look almost exactly like the culprit. But the hard truth is that leaving obvious footprints is never the same as committing the murder.

Who lit the initial spark, and who fanned the flames to make the fire burn far hotter? Drawing this exact distinction is the key to understanding this ongoing global tech stock correction. Framing quantitative trading as the initiator is not just a matter of wrongfully blaming an innocent party — it leads people to ask entirely the wrong questions.

01

What Exactly Is Being Sold Off

First, it must be clearly noted that this round of decline is by no means a one-man show exclusive to the A-share market.

Over in the U.S. stock market, the Nasdaq closed down 2.21% on June 22, with Micron, which had tumbled more than 10% that day, leading the decline. On July 2, the tech sector ETF fell 2.6% in a single trading day, with Applied Materials down 10% and SanDisk down 10.6%. On July 7, the Nasdaq 100 index once dropped as much as 2.4% during intraday trading.

The South Korean market came under synchronized pressure as well. On July 7, Samsung Electronics closed 6.92% lower, SK Hynix fell 6.06%, and the Kospi index dropped 4.91%.

On the A-share market side, the semiconductor, optical communication, PCB, and storage sectors — the very tracks that led the previous rally — are the same ones driving the current pullback. Leading popular stocks such as Hengtong Optic-Electric, GigaDevice, and TFC Optical Communication have all posted pullbacks of more than 30% from their respective highs reached on June 30.

In addition, a claim has been widely circulating in the market that "tech stocks have risen 80% in the first half of the year". But the actual gain of the Nasdaq 100 in the first half of 2026 stands at only 19.9%. For a broad market index, this figure is far from a crazy level. The real madness lies at the individual stock level: Micron once tripled its share price in the first half of the year, and leading stocks in the storage and optical communication industrial chains generally saw gains of over 80%, with many even doubling their value.

This stark contrast precisely points out the true nature of this market cycle: extreme crowding only occurred in several specific industrial chains, a dynamic that cannot be seen at the overall index level. The exact same batch of holdings is being sold off globally — these assets were all previously driven higher by the exact same market narrative, within the same single quarter, and by the exact same group of marginal capital flows. The three core narratives were the accelerated AI infrastructure construction, the upward cycle of memory chips, and the explosive surge in demand for advanced packaging. The superposition of these three logics compressed a massive amount of capital into an extremely narrow exit path.

The Samsung Electronics financial report released on July 7 made everyone clearly realize just how narrow that exit path was. Samsung Electronics' preliminary Q2 performance showed an operating profit of approximately 89.4 trillion won, a staggering year-on-year increase of roughly 19 times, which was about 6% higher than the sell-side consensus expectation of 87.3 trillion won compiled by Reuters. Despite this strong result, Samsung Electronics' share price still dropped 6.92%, dragging down the entire South Korean storage sector and Nasdaq futures along with it.

Bloomberg's headline clearly laid out the cause and effect: *Samsung's Record Profit Fails to Impress After AI Chip Rally*. The article put it more bluntly, noting that the results "topped analyst estimates but fell short of the inflated expectations on the buy side".

This may seem absurd, but John Maynard Keynes described this exact scenario all the way back in 1936. In his famous work *The General Theory of Employment, Interest and Money*, he used a well-known analogy: back then, British newspapers regularly ran beauty contest competitions, where readers were asked to select the six prettiest faces out of a hundred photographs. The winner was the contestant whose selection most closely matched the average choices of all participating readers. Keynes pointed out that smart contestants would never pick the faces they themselves thought were the prettiest; instead, they would select the faces they believed other contestants would think were the prettiest. Even smarter contestants, knowing that everyone else was following this same logic, would take their reasoning a step further and try to guess "what the average public opinion believes the average public opinion will be". Keynes called this the third level of reasoning, and added a note that as far as he knew, some people were even practicing the fourth, fifth, and higher levels of reasoning.

The core point of this analogy is this: once market reasoning enters the third or fourth level, share prices are no longer tied to any objective fundamentals. They are only anchored to a nested, unspoken subjective consensus that no one has ever explicitly articulated.

When a market sector is pushed to an extremely crowded position, public sell-side forecasts lose all their reference value — they are lagging, overly conservative, widely known, and thus already fully priced into the market. The real threshold becomes that implicit expectation, which has never been written into any research report, but has already been validated by the massive amounts of real money poured into accumulating positions throughout May and June. That is why Samsung Electronics' earnings report did not actually "fall short of expectations" — it simply failed to prove that Micron's tripled share price was justified. It did not meet the third-level consensus, and at that moment, the entire market was trading solely on that third-level consensus.

By late June, this structural tension had reached its absolute limit: all marginal buyers had already entered the market, everyone was positioned on the exact same side of the trade, and the only remaining participants were profit holders. The market completely lost its buffer against any news that was "not good enough". This is the literal meaning of crowding level: it is a descriptive statement about market structure, completely unrelated to market sentiment. The spark was right here, and this spark had nothing at all to do with the strategies running on any trading server.

02

The Denominator

So what exactly did quantitative trading do? This is the point where it finally enters the picture. Before answering this question, one critical fact must be clarified first: quantitative trading itself is one of the most thought-provoking realities in the financial industry, and no one truly knows the exact order of magnitude of the capital involved.

The most widely circulated figures claim that "around 2 trillion U.S. dollars globally is deployed in volatility targeting strategies, with another 300 billion U.S. dollars in risk parity strategies". These two numbers originate from the European Central Bank's *Financial Stability Review* published in May 2020. Interestingly, the ECB's citation itself traces back to a 2017 report from the Financial Times.

A capital pool large enough to move hundreds of billions of U.S. dollars in just a few days has its publicly available size estimates stuck at data that is 9 years old. The widely cited figure of 318 billion U.S. dollars for CTA (Commodity Trading Advisor) funds comes from Barclay Hedge, representing the stock of assets as of January 1, 2020. Moreover, trend-following CTAs generally use volatility scaling to manage their positions, which already overlaps with the previously mentioned volatility targeting strategies. As for the actual current size of risk parity strategies, Bloomberg cited estimates from Verus/eVestment in 2024 showing that the figure has shrunk from its 2021 peak of around 160 billion U.S. dollars to roughly 90 billion U.S. dollars by the end of 2023.

Therefore, the rigorous conclusion is that these different types of strategies intersect with each other, their statistical calibers overlap, and their supporting data is outdated — adding up their capital figures makes no practical sense. The only thing that can be confirmed is the approximate order of magnitude: roughly 2 trillion U.S. dollars of capital is entirely driven mechanically by mathematical models, the inverse of volatility, correlation signals, and momentum signals, with no subjective fundamental analysis involved. When compared against the hundreds of trillions of U.S. dollars in total global stock market capitalization, this sum is small enough to be rounded off.

However, it would be a huge mistake to dismiss it as irrelevant, because prices are determined by marginal transactions, not by the total stock of assets. The significance of this pool of capital lies in its ability to contribute a trading volume far exceeding its proportion over a few critical days, with highly consistent trading directions, highly concentrated timing, and absolutely no regard for whether asset prices are cheap or expensive. It sells more as prices fall, because the trigger for selling is not the price level itself — the real thing to watch out for is its perfect synchronization.

The most easily overlooked point is this: nominal scale does not equal actual exposure. The position size of a volatility targeting strategy is the result of a division calculation: target volatility divided by realized volatility. This is the standard formula for such strategies, with typical target volatility levels set between 10% and 15%, and leverage caps generally ranging from 1.5 to 2 times.

Calculated using the standard formula, when a fund sets its volatility target at 15%, and the market is so calm that the realized volatility is only 8%, this division yields a result close to 1.9 times — which is the maximum allowed leverage. This means the fund has to hold stock exposure nearly twice its net asset value to make the portfolio's volatility meet its target requirement.

This is the process by which models are forced to add leverage in a low-volatility environment, a dynamic that has nothing to do with whether fund managers are optimistic about the market outlook. Their performance metric is volatility, not return rate. If the realized volatility does not reach the target, they have no choice but to borrow money to make up for the gap. Once volatility jumps to 24%, the exact same division calculation will produce a result of 0.6 times, which means the fund must liquidate nearly two-thirds of its stock holdings in a very short period of time. It holds no opinion on the valuation of tech stocks, nor has it received any negative news. The denominator has moved, that is all.

The denominator is the key to understanding this entire mechanism, as it simultaneously explains two seemingly contradictory phenomena that actually stem from the exact same origin.

First, the lower the volatility, the higher the leverage. This relationship leads to a disturbing corollary: the maximum leverage of this trading machine will always appear at the moment when the market is the calmest, there is the least bad news, and all investors feel the most reassured. Danger accumulates in tranquility, building up to its maximum level silently through automatic, daily execution. The largest single sell order is destined to take place right after the most peaceful trading days.

Second, in normal times, these strategies act as a sedative for market volatility. Low volatility gives rise to high leverage, and high leverage means continuous mechanical buying flow — this is the driving force that no one mentioned during the market rally from May to June. But once the critical threshold is crossed, the exact same division calculation reverses direction, and the strategies turn into a liquidity black hole, draining liquidity away at the exact moment the market needs it most. Many people see this as two separate behaviors, but in reality, it is nothing more than a single division operation, with the number in the denominator changing.

How should we properly characterize this mechanism? Jón Daníelsson and Hyun Song Shin provided an explanation in their paper *Endogenous Risk*, using the Millennium Bridge in London as their analogy. The bridge was opened to the public on June 10, 2000, but it began shaking violently that very day due to an extremely slight lateral sway. It was forced to close two days later, and did not reopen until February 2002, after nearly 20 months of repairs.

The engineering community's explanation for what exactly happened on the bridge has been revised over time. The early popular version claimed that pedestrians, trying to stay steady on their feet, synchronized their steps with each other, and the synchronized lateral force caused resonance. A research paper published in *Nature Communications* in 2021 proposed a different mechanism: the instinctive corrective movements each pedestrian made to maintain their personal balance, on average, added net energy to the bridge, creating negative damping. Even if pedestrians did not synchronize their steps, this effect alone was enough to make the swaying self-amplify. The observed "step synchronization" was more of a result of the swaying, not its initial cause.

This revised explanation makes the analogy even more appropriate. On today's tech stock trading desks, that 2 trillion U.S. dollar pool of capital is doing exactly the same thing: as prices drop, volatility jumps, the denominator increases, and positions are mechanically reduced. Position reduction pushes prices lower, which makes prices fall further, volatility rises again, and triggers the next batch of model-driven selling. Every institution's risk control adjustment is completely reasonable on its own, each is just trying to keep its footing in the market — but every single adjustment adds net selling pressure to the entire system. They do not need to collude with each other, or even truly synchronize their actions, because the negative damping effect is more than enough on its own.

What Daníelsson and Hyun Song Shin wanted to convey is this: mainstream risk management treats risk as exogenous "weather", something that comes from outside the financial system, which we can only forecast and defend against. But the truly fatal risk in financial markets is endogenous — it grows from within the system, generated by the mutual reactions of all participants. And mechanical sell orders never go to uncrowded places; they only head for the most crowded spots, because that is where their positions are concentrated, and where the liquidity is deep enough for large amounts of capital to enter and exit smoothly.

Goldman Sachs' position tracking data provides quantitative evidence for this mechanism. A report published on February 8, 2026, estimated that in a scenario of deteriorating liquidity, CTAs might sell 33 billion U.S. dollars worth of global stocks over the following week, with a maximum selling amount of 80 billion U.S. dollars over the next month. The trigger condition was the S&P 500 falling below 6707. A report on June 2 showed that CTAs held roughly 93 billion U.S. dollars in long exposure to global stocks at that time, with about 34 billion U.S. dollars of that exposure in S&P 500 futures. Scenario simulations showed that CTAs would buy around 18 billion U.S. dollars if the market moved sideways, buy over 37 billion U.S. dollars if the market continued to rise, and trigger more than 100 billion U.S. dollars in selling if the market kept falling.

This asymmetry itself is direct evidence of "one single division operation working in two directions". As Goldman Sachs pointed out in the exact same report, trend signals in North America were still far above the levels that would trigger large-scale mechanical selling. In other words, CTAs had not yet truly pulled the trigger in this decline, and the main driving force came from the volatility targeting strategies.

Therefore, the judgment that "quantitative trading made this decline steeper, faster, and more like a sudden plunge" is valid, with supporting mechanisms, arithmetic, and position data to back it up. But it must be noted that quantitative trading systems do not have any judgment of "what price a tech stock should be worth", because that variable does not even exist in the machine at all.

03

The Real Question We Should Be Asking

After the sharp A-share market decline on May 21, multiple public and private quant fund managers publicly denied the claim that "quant funds smashed the market", arguing that there was no evidence that a large number of strategies in the market had set specific price levels as unified selling points. That rumor initially arose after individual stocks like Jiangfeng Electronics and Advanced Micro-Fabrication Equipment Inc. China pulled back at round price levels such as 222.22 yuan and 533.33 yuan.

This rebuttal is factually valid: different institutions have vastly different factor configurations, trading windows, turnover rates, and risk control thresholds. There is no collusion structure between them, and they would never hold a meeting to agree to sell together when the index hits the 3400-point level.

But there is a misleading substitution here that must be pointed out: "there is no unified selling trigger" and "quantitative trading does not amplify market volatility" are two completely different propositions. Fund managers are refuting the former, while the market is complaining about the latter — the fact that the former is not true in no way means the latter does not exist.

The pedestrians on the Millennium Bridge did not hold a meeting to agree on when to synchronize their steps, but the bridge still shook violently enough that it had to be closed. The most critical feature of endogenous risk is precisely that it can be generated without any collusion at all.

So why is the accusation that "quantitative trading is the initiator" untenable? If we removed all volatility targeting, risk parity, and CTA strategies from the market, with the same