On October 10 and 11, 2025, more than $19.3 billion in leveraged cryptocurrency positions were forcibly liquidated over 24 hours, according to CoinGlass data cited by CNBC.[s] Bitcoin fell as much as 14.5 percent, altcoins fell more sharply, and more than 1.6 million traders were liquidated.[s] The peak intensity came in a single 60-second interval, during which $3.21 billion vanished, with 93.5 percent of the peak-minute volume coming from forced selling and effectively no time for human intervention.[s]
This was not a traditional market correction. It was flash crash fragility made manifest: a cascade of algorithmic decisions reinforcing each other at speeds no human trader could interrupt. And while cryptocurrency markets grabbed headlines, the same dynamics are quietly embedding themselves into emerging markets around the world, creating systemic risks that regulators are only beginning to understand.
The Speed Arms Race
High-frequency trading firms compete on microseconds. Their servers sit inside exchange data centers, a practice called co-location, which lets them place orders faster than competitors whose machines are farther from the matching engine.[s] This speed advantage is lucrative: a 2025 BIS working paper notes that HFT firms account for more than 50 percent of equity market trading volume.[s]
The benefits are real. Narrower bid-ask spreads mean lower transaction costs for ordinary investors. But the same infrastructure that enables these gains also increases the tendency for extreme events, including flash crashes.[s] When algorithms react to price movements, their responses can compound. One system sells; another system sees the price drop and also sells; a third withdraws its buy orders entirely. In the October 2025 crypto crash, market-making algorithms withdrew liquidity faster than humans could assess what was happening. Order book depth evaporated by more than 98 percent.[s]
Why Emerging Markets Are Vulnerable
Flash crash fragility can be especially dangerous in emerging markets for a structural reason: their derivatives markets often dwarf their underlying cash markets. India provides a clear example. By April 2025, India’s derivatives markets accounted for 61 percent of worldwide equity options contracts.[s] On January 17, 2024, the Bank Nifty index saw $1.26 trillion in options turnover against only $3.6 billion in underlying stock trades: a 350-to-1 disparity.[s]
This imbalance creates fertile ground for sophisticated traders. When the derivatives market is hundreds of times larger than the cash market, a firm with enough capital can move the underlying prices to profit on options positions. That is precisely what India’s market regulator, SEBI, accused U.S. trading giant Jane Street of doing. Jane Street has denied wrongdoing.[s]
According to SEBI’s July 2025 interim order, Jane Street executed large purchases of Bank Nifty constituent stocks in the morning, sometimes representing 15 to 25 percent of total market volume. These purchases pushed the index up. Simultaneously, the firm held massive short positions in options that would profit if the index later fell. In the afternoon, Jane Street reversed course, aggressively selling the stocks it had bought, driving the index back down.[s] SEBI found that Jane Street’s options positions were 7.3 times larger than its stock and futures positions in delta-equivalent terms, an imbalance unusual for legitimate index arbitrage.[s]
The scale was staggering. On several trading days, Jane Street alone accounted for the entire positive price impact on the Bank Nifty index, while the rest of the market exerted net downward pressure.[s] SEBI says the firm made an estimated $4.3 billion from India in a little over two years.[s] Meanwhile, retail investors lost heavily. In fiscal year 2025, Indian retail derivatives traders lost 1.05 trillion rupees, approximately $11.6 billion, up from 750 billion rupees the previous year; BBC noted that these retail losses cannot be attributed to Jane Street directly.[s]
The Herding Problem
The Jane Street case exposed one firm’s alleged manipulation. But there is a more insidious problem: algorithmic systems can create flash crash fragility even when no single actor intends harm. When trading algorithms share similar architectures, similar training data, or similar market signals, they may reach similar conclusions at similar times. Regulators and legal analysts have warned that this kind of model concentration can foster herding behavior and highly correlated markets without explicit coordination.[s]
An April 2026 arXiv preprint argues that AI trading agents can appear diverse during normal market conditions but synchronize under stress. Representation homogeneity, the degree to which algorithms encode market data in similar ways, can compress the effective space of forecast disagreement during crises even when predictions seemed varied beforehand.[s] The authors also argue that low perceived volatility regimes can endogenously accumulate hidden leverage, which subsequently collapses when shocks trigger synchronized deleveraging.[s]
This creates a paradox. Research using Chinese market data suggests algorithmic trading can reduce volatility and dampen investor sentiment during normal conditions.[s] But when the market is in decline, this stabilizing effect weakens significantly.[s] The very mechanisms that smooth out small fluctuations can amplify large ones.
China’s Response
China has taken aggressive action against flash crash fragility. In late 2025, reports said the Shanghai and Shenzhen exchanges were discussing rule amendments that could require securities firms to remove client-dedicated HFT equipment from exchange data centers within three months.[s] In January 2026, 5paisa reported that the Shanghai Futures Exchange had ordered brokers to clear HFT servers by the end of the month.[s]
The market impact was immediate. Metals prices on both the Shanghai and London exchanges dropped sharply after the crackdown was announced.[s] The move suggested how deeply HFT activity had become embedded in price formation: the same report said high-frequency transactions had helped propel earlier price surges, and prices then corrected downward.
What Comes Next
The Jane Street case sets a precedent for regulatory thinking globally, especially for emerging markets with shallow underlying volumes but deep derivatives activity.[s] Countries facing an emerging market debt crisis also confront a second structural vulnerability: their financial infrastructure may be designed for speed rather than stability.
Commentators have proposed several reforms. Sandbox testing or pre-clearance protocols could require exchanges to vet the risk implications of new algorithms before deployment. Enhanced disclosure rules could require consolidated reporting across all entities under common ownership. Settlement methodology itself could be revised: expanding the time window for price calculation or introducing participation caps during expiry hours could insulate closing prices from concentrated trades.
Advanced AI systems could create more brittle and highly correlated markets during periods of stress, exacerbating swings by acting in unison and undermining liquidity when it is needed most.[s] Algorithmic trading strategies often include safety mechanisms that trigger de-risking or complete shutdowns during high volatility. Their simultaneous activation across multiple participants could create destabilizing feedback loops.[s]
Flash crash fragility is not a bug in modern markets. It is a feature of infrastructure designed to maximize speed at the expense of resilience. The question is whether regulators will act before the next cascade, or after.
Co-Location Infrastructure and Latency Arbitrage
High-frequency trading depends on physical proximity to exchange matching engines. Co-location services allow firms to place servers inside exchange data centers, reducing round-trip latency to microseconds.[s] A 2025 BIS working paper says HFTs represent more than 50 percent of equity market trading volume.[s]
A September 2025 BIS working paper used NASDAQ co-location and latency improvement upgrades as natural experiments to examine HFT’s economic effects. The findings were counterintuitive: on average, HFT leads to higher cost of capital.[s] The mechanism is systematic risk amplification. HFT’s correlated trading strategies make low-beta stocks more responsive to market-wide information, increasing their systematic risk exposure and therefore their required returns.[s] While HFT narrows bid-ask spreads, it also increases the tendency for flash crashes.[s]
The October 2025 cryptocurrency cascade demonstrated flash crash fragility at scale. CoinGlass recorded more than $19.3 billion in forced liquidations over 24 hours, the largest liquidation event ever tracked by the firm as of CNBC’s October 2025 report.[s] During peak intensity, $3.21 billion liquidated in a 60-second interval, with 93.5 percent of that minute’s volume coming from forced selling.[s] Market-making algorithms withdrew quotes while liquidation engines fired simultaneously, draining more than 98 percent of order book depth.[s]
The Derivatives-Cash Disparity in Emerging Markets
Flash crash fragility is structurally amplified where derivatives notional turnover vastly exceeds underlying cash market liquidity. India exemplifies this pattern. By April 2025, Indian derivatives markets accounted for 61 percent of global equity options volume.[s] On January 17, 2024, Bank Nifty recorded $1.26 trillion in options notional against $3.6 billion in underlying stock trades, a 350:1 disparity.[s]
SEBI’s July 2025 interim order against Jane Street documented a two-patch trading strategy. Jane Street has denied wrongdoing.[s] During Patch I (09:15-11:46), the firm executed aggressive buys of Bank Nifty constituent stocks, representing 15-25 percent of market-wide volume in select scrips. Simultaneously, it built short options exposure through sold calls and bought puts. During Patch II (11:49 to close), it reversed the cash market positions, exerting downward index pressure coinciding with expiry settlement.[s]
Jane Street’s option positions were 7.3 times larger than stock and futures positions in delta-equivalent terms.[s] On several days, SEBI attributed the entire positive price impact on Bank Nifty to Jane Street alone, while aggregate market flow was net negative.[s] SEBI says the firm made $4.3 billion from Indian markets in approximately two years.[s] Retail investors lost $11.6 billion in FY25 derivatives trading, up 40 percent from FY24, though BBC noted that those losses cannot be attributed to Jane Street directly.[s]
Representation Homogeneity and Synchronized Deleveraging
Even absent intentional manipulation, algorithmic systems can create flash crash fragility through emergent coordination. An April 2026 structural multi-agent model examined how AI trading agents with similar data representations generate systemic instability.[s]
The key theoretical contribution distinguishes representation homogeneity (similarity in how agents encode market states) from forecast overlap (similarity in predicted returns). These are related but not equivalent. Under normal conditions, agents with similar representations may generate diverse forecasts. Under stress, representation homogeneity compresses the effective space of forecast disagreement, causing predictions to converge precisely when diversity matters most.[s]
The preprint argues that low perceived volatility regimes can endogenously accumulate hidden leverage through position stickiness, which collapses when shocks trigger synchronized deleveraging.[s] It also hypothesizes that increasing representation similarity amplifies synchronization in beliefs and positions, generating volatility clustering, liquidity stress, and elevated tail risk.[s]
Regulators have warned about AI herding: correlated behavior without explicit coordination.[s] Traditional surveillance may struggle to distinguish legitimate convergent analysis from prohibited coordination. Shared machine-learning models and common data feeds can create herding dynamics: when one node de-risks, correlated agents may follow, draining order books.[s]
Conditional Stabilization and Asymmetric Effects
Research using Chinese market level-2 data found algorithmic trading generally reduces volatility by dampening investor sentiment, with the sentiment channel accounting for approximately 25 percent of the total effect.[s] However, this stabilizing effect is conditional. When the market is in decline, algorithmic trading’s volatility-reducing role weakens, and the mediating effect on investor sentiment also decreases.[s]
This asymmetry is critical to understanding flash crash fragility. The same mechanisms that smooth normal fluctuations can amplify tail events. Algorithmic safety mechanisms that trigger de-risking during high volatility, when activated simultaneously across participants, create destabilizing feedback loops rather than stabilizing ones.[s]
Regulatory Responses: China’s Co-Location Ban
China’s approach has been aggressive. In late 2025, reports said the CSRC, Shanghai exchange, and Shenzhen exchange were discussing rule changes that could limit co-location services and require securities firms to remove client-dedicated HFT equipment within three months.[s] In January 2026, 5paisa reported that the Shanghai Futures Exchange had ordered brokers to clear HFT servers by the end of the month.[s]
The crackdown produced immediate price effects. Metals on both Shanghai and London exchanges declined sharply after the announcement.[s] The same report said high-frequency transactions had helped propel earlier price surges across global metals markets, and prices then corrected downward.[s]
The Jane Street case sets regulatory precedent for emerging markets with shallow cash volumes and deep derivatives activity.[s] Proposed reforms include sandbox pre-clearance for algorithms impacting expiry prices, enhanced FPI disclosure across commonly-owned entities, VWAP settlement window expansion or participation caps, and version-controlled algorithmic audit trails.
Concentration in data providers and AI-as-a-Service platforms raises monoculture concerns: market participants drawing from identical data and employing similar models reach similar conclusions.[s] Advanced AI could create more brittle, highly correlated markets during stress, exacerbating volatility and undermining liquidity precisely when it is needed.[s]
Flash crash fragility is an infrastructural feature, not a transient bug. Whether regulators move to address representation homogeneity, latency asymmetry, and derivatives-cash imbalances before the next cascade remains an open question.
