The Flash Crash of May 6, 2010

Key Takeaways

• The trigger was a single hedge executed badly, not a market-wide event. A mutual fund complex sold 75,000 E-Mini S&P 500 futures contracts, worth about $4.1 billion, using an algorithm that targeted 9 percent of trailing volume with no limit on price or time, and finished in roughly 20 minutes what a similarly sized program had taken the same firm more than 5 hours to execute months earlier.
• Liquidity did not just thin, it nearly disappeared. Buy-side depth in the E-Mini fell from a morning average near $5.5 billion to about $58 million in the final seconds before the low, and by 2:45:28 p.m. there were fewer than 1,050 contracts of buy-side interest resting in the entire book.
• High-frequency traders were not absent, they were just not net buyers. In a 14-second window they traded over 27,000 contracts, about 49 percent of all volume, while their own net position grew by only about 200 contracts, a pattern the joint SEC and CFTC investigation called a "hot-potato" effect.
• The stock-level damage was worse than the index-level damage. Over 20,000 trades across 326 securities, 5.5 million shares in total, executed 60 percent or more away from their price just minutes earlier. Accenture's bid fell from about $30 to one cent in seven seconds. Every one of those trades was cancelled the same evening.
• A five-second pause did more than an hour of analysis could have. The CME's Stop Logic Functionality halted E-Mini trading for five seconds at 2:45:28 p.m., and the rebound that followed brought the futures market back near its pre-drop level within about 23 minutes.
• The circuit breakers in place that day never activated. Adopted in 1988 after Black Monday and triggered only once since, they measured a full-day decline in the Dow Jones Industrial Average, a benchmark and a time frame that had nothing to do with a five-and-a-half-minute plunge in a different index.
• The regulatory response took years, and the enforcement case took five. The single-stock circuit breaker built within weeks of the crash was replaced in 2013 by a limit-up limit-down system built for this exact failure mode, and in 2015 the CFTC charged a London-based trader whose spoofing algorithm the agency says contributed to the conditions that produced the crash, a claim narrower than the popular account of what caused it.

What Happened During the Flash Crash of May 6, 2010?

The joint investigative report published by the staffs of the Securities and Exchange Commission and the Commodity Futures Trading Commission divides May 6 into five phases, and the shape of that division matters more than any single number in it. The first two phases, running from the market open through about 2:41 p.m., were an unusually bad but entirely ordinary bad day: broad stock index products lost roughly 3 percent in the morning session and another 1 to 2 percent by early afternoon on genuine bad news out of Europe. The third phase is the one the name refers to: between 2:41 p.m. and 2:45:28 p.m., about four and a half minutes, the broad market fell a further 5 to 6 percent, touching intraday lows of 9 to 10 percent below the prior day's close. The fourth phase, from 2:45 p.m. to about 3:00 p.m., is where the damage moved from the index level to individual securities: broad indices were already recovering while hundreds of stocks and ETFs traded in a disorderly fashion, in some cases as low as one cent or as high as $100,000. The fifth phase, from about 3:00 p.m. onward, was a return to ordinary trading, and by the close, major indices had recovered to a loss of about 3 percent for the day, roughly where the afternoon selloff had started before the plunge.

That structure separates three questions casual retellings conflate: why the broad market fell in the first two phases, why it fell so much faster in the third, and why individual stocks behaved so much worse than the index supposedly driving them. The third and fourth phases are where almost everything distinctive about this event happened.

Timeline of May 6, 2010

Times are Eastern. Figures are from the joint SEC and CFTC investigative report unless otherwise noted.

What Was Already Under Stress Before 2:32 P.M.?

The Flash Crash is often told as though it started from a calm market, which makes the speed of the plunge look inexplicable. It did not start from calm. The morning opened to worsening news about the European sovereign debt crisis: credit default swap premiums rose on Greek, Portuguese, Spanish, Italian and Irish government debt, and the Euro came under sustained pressure against both the dollar and the yen from around 1:00 p.m. FRED's own exchange-rate data confirms the direction and scale: the dollar-euro rate fell from 1.2890 on May 5 to 1.2689 on May 6, about 1.6 percent in a single session.

That sentiment showed up first as volatility, not as a price collapse. Starting around 1:00 p.m., the number of Liquidity Replenishment Point pauses on the New York Stock Exchange, a mechanism that briefly slows trading in individual stocks moving too fast, rose well above average levels. By 2:30 p.m. the VIX was up 22.5 percent from its opening level, and by the end of the day it had risen 31.7 percent, the fourth-largest single-day VIX increase on record at the time per the joint report. FRED's own VIX series independently confirms the day's full magnitude: 24.91 on May 5 to 32.80 on May 6, almost exactly 31.7 percent. Ten-year Treasury yields fell from 3.58 percent to 3.41 percent as investors moved toward safety, and gold futures rose.

The part of this that matters most for what came next is liquidity, not price. By 2:30 p.m., buy-side depth in the E-Mini futures contract, essentially the total volume of resting buy orders across the full order book, had already fallen 55 percent from its morning average, from close to $6 billion to $2.65 billion. Buy-side depth in SPY, the S&P 500 ETF, had fallen 20 percent over the same window. Neither of those declines was itself a crisis. Both meant the market had less capacity to absorb a large sell order than it had two hours earlier, and at 2:32 p.m. a large sell order arrived.

What Was the Sell Algorithm, and Why Did a Hedge Move So Fast?

At 2:32 p.m., a mutual fund complex, identified in the joint report only as "a large Fundamental Seller," initiated a program to sell 75,000 June 2010 E-Mini S&P 500 futures contracts, valued at approximately $4.1 billion, to hedge an existing equity position. The trade itself was not unusual in size. It was the largest net change in position by any E-Mini trader since the start of the year, but the report notes that only two single-day sell programs of equal or larger size had been executed in the E-Mini in the twelve months before May 6, one of them by the same firm.

What made this program different was how it was executed. A large seller has three broad choices: hand the order to an intermediary who manages the execution, enter it manually over time, or route it through an automated algorithm that targets some combination of price, time and volume. The firm chose a fourth path within that third option: an algorithm programmed to feed sell orders into the market at a rate equal to 9 percent of the trading volume observed over the previous minute, with no constraint on price and no constraint on time. When the same firm had executed a similarly sized sell program earlier in the year, it used a mix of manual trading and algorithms that did account for price, time and volume, and it took more than five hours to work through the first 75,000 contracts. On May 6, the volume-only algorithm worked through the entire 75,000-contract program in about 20 minutes.

The mechanism behind that speed is a feedback loop the algorithm's own design created. As it sold, it added to trading volume, and because it targeted a percentage of trailing volume, higher volume made it sell faster still, even though the orders it had already sent were "arguably not yet fully absorbed," in the report's words, by the time the next batch arrived. In calm markets a volume-following algorithm without price limits is reasonable, because volume and liquidity usually move together. On May 6, with liquidity already thinned, volume kept rising even as the buyers capable of absorbing it kept withdrawing, and the algorithm read that rising volume as room to sell faster into a market with less and less room to buy.

The design flaw was structural, not exotic. Nothing about the Sell Algorithm was a bug. It executed exactly as programmed: sell at a rate proportional to volume, ignore price. The failure was that this rule, chosen for a $4.1 billion order on a day when liquidity had already fallen by half, treats a symptom of stress, rising volume, as a green light, when in a thinning market rising volume can mean the opposite. A volume-participation algorithm is not inherently dangerous. Running one without a price or time constraint into an already-stressed market is.

Was Waddell & Reed's Trade the Cause of the Crash?

The SEC and CFTC's own report never names the firm behind the Sell Algorithm. It was identified in contemporaneous financial press reporting, and confirmed by the firm's own public statement, as Waddell & Reed Financial, a Kansas-based asset manager, which described its trade as a bona fide hedge of an existing equity position and noted it was one of roughly 250 different accounts trading the E-Mini contract that afternoon. Both claims are consistent with the investigative report: the trade was a disclosed hedge rather than a directional bet. The report's own footnotes also track what happened to each of the two positions discussed above, after the fact: the short position built up by the firm's earlier, five-hour sell program was closed within a single day, while the short position the 20-minute Sell Algorithm created on May 6 itself took the firm more than six hours to offset, on a later date. The rapid execution did not produce an equally rapid exit.

Whether that trade "caused" the crash depends on which of two very different questions is being asked. If the question is which single order set the sequence in motion, the answer is close to yes: this was the largest, fastest-executed sell program in the E-Mini that day, and the joint report traces the initial price decline directly to its interaction with the market's already-thin liquidity. If the question is whether a similarly sized, similarly urgent sell order would reliably produce a crash of this magnitude on an ordinary day, the answer is almost certainly no. The same firm had sold a comparable position months earlier without incident, using a different execution method on a market with normal liquidity. The trigger and the vulnerability were both necessary. Neither was sufficient alone.

This distinction is not a technicality. A reader who concludes "one firm's algorithm caused the Flash Crash" learns to worry about large hedgers. A reader who understands that a large, price-insensitive order interacting with liquidity that had already halved learns a more durable lesson: the danger sits in the interaction, not in either ingredient alone. Waddell & Reed was never charged with any wrongdoing by the SEC or the CFTC in connection with May 6, and the joint report's own framing treats the firm's order as a catalyst that a fragile market amplified, not as misconduct.

How Did High-Frequency Traders Create a "Hot Potato" in the E-Mini?

The joint report classified more than 15,000 trading accounts active in the E-Mini on May 6 into six categories. Two of those categories, high-frequency traders and Intermediaries, were identified from each account's trading pattern over the three days before May 6 rather than from May 6 itself; the other four were based on the account's own behavior that day. Within the pool of market-maker-like accounts, the busiest 3 percent by trade count, just 16 accounts out of the more than 15,000 active that day, were designated high-frequency traders. Those 16 accounts traded over 1,455,000 contracts on May 6, close to a third of the entire day's volume, while their aggregate net position rarely exceeded three to four thousand contracts long or short at any point. That combination, enormous turnover paired with a near-flat net position, is the defining signature of the strategy: these firms make money on the bid-ask spread across a huge number of trades, not on holding a directional view.

In the early minutes of the Sell Algorithm's execution, that strategy worked as intended. High-frequency traders and other intermediaries were the natural buyers of the algorithm's initial orders, and by doing so they built up a temporary long position of about 3,300 contracts, providing exactly the liquidity a stressed market needed. Between 2:41 p.m. and 2:44 p.m., though, those same firms began aggressively selling about 2,000 of those contracts back into the market to bring their own inventory back down. At that moment they stopped providing liquidity and started competing for it, at the same time the Sell Algorithm was still selling and fundamental buyers were struggling to keep pace.

The most extreme phase came in the 14 seconds between 2:45:13 p.m. and 2:45:27 p.m. High-frequency traders traded over 27,000 E-Mini contracts in that window, about 49 percent of all trading volume, while their net position changed by only about 200 contracts. The joint report calls this a "hot-potato" effect: the same contracts changing hands rapidly among high-frequency firms, each trying to avoid holding inventory, without any of that trading actually connecting a genuine seller to a genuine buyer willing to hold the position. Volume looked enormous while liquidity, the capacity to absorb a new order without moving the price sharply, was nearly gone. By the end of that 14-second window, buy-side depth in the E-Mini had fallen to about $58 million, less than 1 percent of its level that morning, and the price dropped a further 1.7 percent in those same 15 seconds to touch its intraday low of 1056.

It is worth stating precisely, because it is routinely garbled in retellings: high-frequency traders did not cause the initial selling, and the joint report does not describe them as villains who withdrew and let the market fall. They were net sellers during the plunge because they were unwinding inventory they had taken on while providing liquidity minutes earlier. High trading volume, the report states directly, is not a reliable indicator of market liquidity, especially during periods of significant volatility, and May 6 is the clearest evidence of why that distinction matters.

What Actually Stopped the Free Fall?

At 2:45:28 p.m., with buy-side resting orders in the E-Mini down to fewer than 1,050 contracts, under 1 percent of the morning's depth, the CME's Stop Logic Functionality triggered a five-second trading pause. This is a CME-specific safeguard, distinct from the exchange-wide circuit breakers discussed below: it halts trading in a single futures contract when a cascade of stop-loss orders would otherwise push the price outside a predefined "no bust" range, which on May 6 was six index points, or about 0.6 percent of the E-Mini's price. The pause was not a broad market-wide halt. It applied to one contract, for five seconds.

Five seconds does not sound like enough time to matter, and the joint report's own order-book data shows why it was. During the pause, sell-side pressure eased and buy-side interest had a moment to arrive without being immediately overwhelmed by the Sell Algorithm's continuing execution. When trading resumed at 2:45:33 p.m., prices stabilized almost immediately and began to climb. The Sell Algorithm itself kept running until about 2:51 p.m., selling its remaining contracts into a now-rising market, and between 2:45 p.m. and 3:08 p.m. fundamental buyers bought more than 110,000 E-Mini contracts net, over twelve times the net buying seen in the same window on an ordinary day. By 3:08 p.m. the E-Mini had returned close to its pre-drop level.

This runs against the usual instinct that halting a market is itself dangerous, because it prevents price discovery. On May 6, a five-second interruption in one contract appears to have been the single most effective intervention of the entire afternoon, precisely because it was short enough not to trap capital and long enough to let algorithms reset and buyers reassess, the same design principle behind the 2012 circuit-breaker rewrite discussed later in this article.

Why Did Stocks Like Accenture Trade at One Cent While the Index Was Recovering?

The most visually dramatic part of May 6 happened after the index-level crisis was already ending. Between 2:40 p.m. and 3:00 p.m., roughly 2 billion shares traded across the whole market, more than $56 billion of volume, and over 98 percent of those shares traded within 10 percent of their 2:40 p.m. reference price. That statistic is easy to miss because the remaining fraction produced the images that defined the day: individual stocks and ETFs trading as low as one cent or as high as $100,000, prices that bore no relationship to any reasonable estimate of value.

The mechanism centers on a market-making obligation called a stub quote. Exchange rules require a registered market maker to maintain a continuous two-sided quote in the securities it covers, even when it has decided the market is too uncertain to trade normally. Rather than withdraw entirely, which the rules did not clearly permit, some market makers left standing quotes far from any realistic price, a bid of a penny or an offer of $100,000, purely to satisfy the obligation while signaling no genuine interest in trading. When a stop-loss or market order arrived for a stock whose real liquidity had evaporated, with nothing better resting in the book, it executed against that stub quote instead.

Individual stocks in the order-book reconstruction

From the joint report's detailed order-book analysis of selected securities.

The joint report's aggregate analysis of the 100 largest ETFs found they were disproportionately affected relative to individual large-cap stocks: buy-side liquidity for that ETF basket fell to about 10 percent of its midday value at the worst point, against about 20 percent for the top 100 non-ETF stocks. The report's explanation is that ETF liquidity is concentrated much closer to the current price than stock liquidity, provided mostly by professional market makers rather than resting retail orders further from the market, so when those professionals paused, there was comparatively little liquidity left standing behind them.

How Many Trades Were Cancelled, and Under What Rule?

Over 20,000 individual trades, representing 5.5 million shares across 326 separate securities, executed at prices 60 percent or more away from their value at 2:40 p.m. that afternoon. Almost two-thirds of the shares in those trades executed below $1.00, and about 5 percent executed above $100. After the market closed on May 6, the exchanges and FINRA jointly agreed to cancel, or "break," every one of those trades under their existing "clearly erroneous" trade rules, a mechanism that already existed before May 6 but had never been applied at this scale or with this much uncertainty about where the line should sit.

That uncertainty was itself a problem the joint report flagged directly. Market participants raised concern that the 60 percent threshold used that day, and the process used to apply it, were not transparent in advance, which affects how willing a market maker is to keep quoting during the next stress event if it cannot predict which trades will later be unwound. In response, the SEC worked with the exchanges and FINRA to adopt clearer, price-scaled standards for breaking erroneous trades going forward, rather than a single discretionary judgment applied after the fact.

Almost 90 percent of the broken trades were entered as limit orders, higher than expected if most of the flow were retail market orders, since brokers often convert market orders into other order types before routing them. Retail stop-loss orders that triggered during the decline were still a meaningful part of the flow that hit stub quotes, a detail that matters for the broader lesson about stop-loss orders discussed later in this article.

Why Didn't the Circuit Breakers That Existed in 2010 Stop This?

Market-wide circuit breakers already existed on May 6, 2010, adopted in October 1988 in direct response to Black Monday, and they had been triggered exactly once in the intervening twenty-two years, in 1997. They did not activate at any point during the Flash Crash, and the reason is a straightforward mismatch between what those breakers measured and what actually happened.

The 1988-era breakers were built around the Dow Jones Industrial Average, and they measured a decline from the prior day's closing price, triggering a halt only at 10, 20, or 30 percent, thresholds designed to interrupt a slow, grinding, full-session collapse like October 19, 1987, described in Black Monday. The Flash Crash's defining feature was speed within a single afternoon, not depth over a full session: the market's worst intraday level was down 9 to 10 percent, a fraction of even the lowest trigger threshold, and it happened and substantially reversed within about nine minutes. A circuit breaker calibrated to catch a slow-motion 1987-style crash has no mechanism for noticing a fast, narrow, self-reversing plunge, because by the time enough of the day's decline had accumulated to approach even the 10 percent threshold, the plunge itself was already over.

This gap is what the joint report's "Lessons Learned" section identifies as the central structural problem, distinct from any single firm's behavior: the market-wide breakers, the single-stock quoting obligations, and the erroneous-trade rules had all been designed around a market structure that had already changed by 2010, and none of the three was built to interrupt a liquidity crisis that develops and resolves within minutes. The policy response, covered next, is a rebuild of all three around that faster failure mode.

What Did Regulators Build After May 6, and How Is It Different?

The response happened in two stages: an emergency patch within weeks, and a permanent rebuild about two years later.

The interim fix: single-stock circuit breakers

Within weeks of the crash, SEC staff worked with the exchanges and FINRA to build a pilot pausing trading in an individual security for five minutes if its price moved 10 percent within the preceding five minutes. The SEC approved applying this to S&P 500 stocks on June 10, 2010, and expanded it to the Russell 1000 Index and certain ETFs that September, running as a pilot through year-end. This was explicitly a stopgap: fast enough to interrupt another Accenture-style collapse, but built stock by stock rather than as an integrated system.

The permanent rebuild: Limit Up-Limit Down

On June 1, 2012, the SEC approved two proposals from the exchanges and FINRA that replaced the interim system entirely, implemented by February 4, 2013. The first, the Limit Up-Limit Down Plan, prevents trades in a listed stock from occurring outside a price band set as a percentage above and below the security's average price over the preceding five minutes, rather than pausing trading only after a move had already happened. For S&P 500 and Russell 1000 stocks and certain exchange-traded products, that band is 5 percent; for other listed securities it is 10 percent, doubled near the open and close when volatility is naturally higher, with wider bands for stocks priced at $3 or less. If a stock cannot trade inside its band for more than 15 seconds, a five-minute pause follows, the same design principle validated by the CME's five-second E-Mini pause on May 6 itself: short, automatic, and triggered by the band rather than by human judgment in the moment.

The second proposal rebuilt the market-wide circuit breakers that had sat unused on May 6. The decline thresholds dropped from 10, 20, and 30 percent to 7, 13, and 20 percent of the prior day's closing price, halt durations shortened from 30, 60, or 120 minutes to a flat 15 minutes for the two lower thresholds, the reference index changed from the Dow Jones Industrial Average to the broader S&P 500 Index, and trigger levels are now recalculated daily rather than quarterly. SEC Chairman Mary L. Schapiro described the goal as an "automated and appropriately calibrated way to pause or limit trading if prices move too far too fast," a description that reads, in hindsight, as a direct response to what May 6 showed the old rules could not do.

Alongside those two headline changes, the SEC also adopted rules effectively prohibiting stub quotes, and a separate rule requiring broker-dealers to have risk controls in place before giving customers direct market access, closing off two mechanisms documented in this article: the quoting practice that let Accenture trade at a penny, and the possibility of an unchecked algorithm reaching the market with no broker-level safeguard in front of it.

Did Navinder Sarao's Spoofing Cause the Flash Crash?

Five years after the joint SEC and CFTC report was published, the story gained a second chapter. On April 21, 2015, the CFTC and the Department of Justice unsealed civil and criminal actions against Navinder Singh Sarao, a London-based futures trader, and his company, Nav Sarao Futures Limited. Scotland Yard arrested Sarao in London the same day at the request of U.S. authorities.

According to the CFTC's complaint, Sarao had modified an off-the-shelf trading platform in or around June 2009 to automatically place four to six exceptionally large sell orders into the visible E-Mini order book simultaneously, spaced one price level apart and continuously repositioned to stay three or four price levels from the best asking price, a technique the complaint calls the Layering Algorithm. Because these orders were designed to be cancelled before execution rather than filled, they created the visual impression of heavy selling interest without Sarao intending to sell at those levels, a manipulative practice known as spoofing. The complaint alleges he used this algorithm on more than 400 trading days between April 2010 and April 2015, producing total E-Mini profits exceeding $40 million over that period.

On May 6, 2010 specifically, the CFTC's 2015 complaint alleges Sarao ran the Layering Algorithm continuously for more than two hours immediately before the crash, applying what the agency described at the time as close to $200 million of persistent downward pressure on the E-Mini price. The 2016 consent order, in which Sarao admitted the underlying conduct rather than merely facing an allegation, gives a larger and more precise accounting: it puts his cumulative use of the Layering Algorithm on May 6 at more than four hours and twenty-five minutes, representing roughly $170 million to more than $200 million of downward pressure and 20 to 29 percent of the entire sell-side of the E-Mini order book, before the program was cancelled at 2:40 p.m. Eastern, just as the sharpest phase of the crash was beginning. The CFTC's own language on causation is precise and worth quoting exactly, because it is narrower than how the case is usually summarized: his activity "contributed to an extreme E-mini S&P order book imbalance that contributed to market conditions that led to the Flash Crash." That is a claim about one contributing input into an already-stressed order book, not that his trading caused the crash, and it does not appear anywhere in the original 2010 joint report, published five years before his activity was identified.

Sarao pleaded guilty to wire fraud and spoofing in November 2016. A separate CFTC consent order that same month ordered him to pay $12,871,587.26 in disgorgement and a $25,743,174.52 civil penalty, a combined $38,614,761.78. He was sentenced in a Chicago federal court on January 29, 2020, to time served, roughly four months already spent in UK custody after his 2015 arrest, plus one year of home confinement, with no additional prison time; prosecutors had recommended against further incarceration, citing his cooperation and his diagnosis with autism.

Two true stories, one crash. The 2010 joint report explains May 6 as a mechanical failure: a price-insensitive algorithm, a thinning order book, and a hot-potato liquidity spiral among high-frequency traders, none of which required anyone to have broken any rule. The 2015 enforcement case adds a second, independent story: a trader spent over two hours before the crash placing orders he never intended to execute, worsening the exact order-book imbalance the mechanical account describes. Both are documented and both are true. What is not supported by either document is the popular compression of the two into "one man caused the Flash Crash," a claim the agency that built the case explicitly did not make.

Which Warning Signs Were Visible Before 2:32 P.M., and Which Only in Hindsight?

May 6 offers an unusually clean test of this question because the joint investigation reconstructed the order book second by second, which means it is possible to say with precision what an observer could and could not have known at each point in the afternoon.

Signals classified by whether they were observable in real time before the 2:41 p.m. plunge began.

The honest reading of that table is that the risk-off mood of the morning was genuinely visible and gave a reasonable observer cause to expect a bad, volatile afternoon. Nothing available to anyone outside the exchanges' own order-book data could have specified that the mechanism would be a single algorithm's execution style colliding with a liquidity spiral among market makers, or that the outcome would be a nine-minute round trip rather than a sustained decline. Compare the Silicon Valley Bank failure, where the balance-sheet fragility was fully public and only the timing was not: here, even the fragility itself was invisible until after the fact.

Common Myths About the Flash Crash

"Navinder Sarao caused the Flash Crash." The CFTC's own complaint describes Sarao's activity as contributing to "market conditions that led to" the crash, not as its cause. The 2010 joint report, built entirely on order-book reconstruction with no knowledge of Sarao's existence, attributes the crash to the Sell Algorithm's interaction with thinning liquidity and the high-frequency hot-potato effect. Sarao's spoofing worsened an already-stressed order book for over two hours beforehand. It did not, on the government's own account, trigger the plunge.

"High-frequency traders pulled out and let the market fall." They did the opposite of pulling out. In the most extreme 14-second window, high-frequency accounts traded 49 percent of all E-Mini volume. What they stopped doing was providing net new liquidity, because they were unwinding a temporary long position built up minutes earlier. Rapid trading with a flat net position looks, from the outside, like withdrawal. The order-book data shows it was closer to a large volume of activity that never connected a genuine buyer to a genuine seller.

"The market lost trillions of dollars in value." This conflates the intraday low with the closing price, and treats a paper decline that reversed within the same session as a permanent loss. By the close, the major indices were down about 3 percent for the day, unremarkable by the standards of 2008 or 2020. The stocks that traded at a penny or $100,000 did so for seconds, and every one of those trades was subsequently cancelled, so no investor's account was ever settled at those prices.

"Circuit breakers failed on May 6." The circuit breakers that existed did not fail; they simply did not apply. They were built to interrupt a slow, full-day decline in the Dow Jones Industrial Average, and May 6's defining feature was a fast, narrow, largely self-correcting plunge in a different index over a few minutes. A tool not designed to catch a given failure mode failing to catch it is a design gap, which regulators closed with Limit Up-Limit Down, not a malfunction.

"This could not happen again because of the new rules." Limit Up-Limit Down and the rebuilt market-wide circuit breakers address the specific mechanisms of May 6: they slow a single security before it can print at a stub-quote price, and they use a faster, broader-based market-wide trigger. Neither rule eliminates the underlying condition that made May 6 possible, which is that automated execution and thinning liquidity can still interact badly under stress. The system is better instrumented to interrupt that interaction quickly. It has not removed the interaction itself.

What a Reader Can Actually Carry Forward

The temptation with a case this technical is to file it away as a story about market plumbing that has nothing to do with an individual investor's decisions. That misses the part of May 6 that touched real accounts directly.

What generalizes

• A resting stop-loss order has no protection against the price it executes at. A stop-loss order becomes a market order once triggered, and a market order fills against whatever is available, which on May 6 included one-cent stub quotes. Retail stop-loss orders were a meaningful share of the flow that hit those quotes. A reader who wants a guaranteed exit price, not just a guaranteed trigger, needs a stop-limit order instead, covered in stock order types, understanding that a stop-limit can fail to fill at all in a fast market.
• Displayed liquidity can be far larger than actionable liquidity. The E-Mini's buy-side depth looked adequate at 2:00 p.m. and had fallen 99 percent within 45 minutes, without a single news headline announcing the change. Depth shown on a screen describes a moment, not a guarantee.
• Extreme, brief price prints are not automatically real losses. Every trade more than 60 percent away from its reference price on May 6 was cancelled hours later. An investor who panicked and sold into that same window at a real, executable price would have locked in a loss the market itself judged illegitimate for the anomalous trades and never actually experienced for a normal one.
• Automated execution needs constraints, not just a target. The Sell Algorithm was not reckless by design; it simply lacked a price or time boundary that would have slowed it when the market stopped cooperating. The same logic applies at retail scale: a recurring trade or a standing order benefits from an explicit boundary for the scenario where the market behaves nothing like it did when the rule was written.

What does not generalize

• The nine-minute round trip. It depended on the specific combination of an already-thin order book, a single very large price-insensitive order, and a five-second circuit-breaker pause that happened to arrive at close to the worst moment. Most liquidity events do not resolve this fast or this cleanly.
• Trading at a literal penny or $100,000. Those prices were a direct consequence of stub-quoting rules that no longer exist in that form. The rules that made Accenture's one-cent trades possible were changed specifically because of this event.
• The absence of any real economic damage. May 6 left no bank insolvent, no company impaired and no depositor at risk; it was a pure market-structure event. A liquidity crisis with real balance-sheet consequences, like the 2008 financial crisis, does not resolve in an afternoon because the underlying losses are not artifacts of a broken quoting mechanism.

The one question worth asking now

Not "could a flash crash happen again," a question about market microstructure regulators are better positioned to answer than any individual investor. The more useful question is narrower: if every order in your account right now, a stop-loss, a limit order, a standing rebalance, executed in the next 30 seconds against whatever price the market happened to offer, would that price be one you could live with? For most long-term positions a few minutes of dislocation do not matter, because the order was never going to trigger in that window. For anyone running a tight stop-loss on a volatile position, the answer is worth checking before the next stress event, not during it.

References

Every figure on this page was verified against the following sources, each retrieved on 26 August 2026:

• U.S. Commodity Futures Trading Commission and U.S. Securities and Exchange Commission: Findings Regarding the Market Events of May 6, 2010: the day's timeline and five-phase structure, the 75,000-contract Sell Algorithm and its parameters, the E-Mini and SPY liquidity figures, the high-frequency trader classification and hot-potato volume statistics, the CME Stop Logic pause, the order-book data for Accenture, Procter & Gamble, 3M and IBM, the 20,000-plus broken trades across 326 securities, and the single-stock circuit breaker pilot adopted after May 6.
• SEC: SEC Approves Proposals to Address Extraordinary Volatility in Individual Stocks and Broader Stock Market, Press Release 2012-107, June 1, 2012: the Limit Up-Limit Down Plan's price bands and pause mechanism, the February 4, 2013 implementation date, the rebuilt market-wide circuit breaker thresholds and durations, the switch to the S&P 500 as reference index, the October 1988 adoption and 1997 trigger of the original breakers, and the SEC Chairman's quoted statement.
• CFTC: CFTC Charges U.K. Resident Navinder Singh Sarao and His Company Nav Sarao Futures Limited PLC with Price Manipulation and Spoofing, Release 7156-15, April 21, 2015: the Layering Algorithm's mechanics and June 2009 origin, its use on 400-plus trading days between April 2010 and April 2015, the over $40 million in total E-Mini profits, the two-hour, approximately $200 million pressure applied on May 6, 2010, and the CFTC's exact language on what that activity contributed to.
• CFTC: Federal Court in Chicago Orders U.K. Resident Navinder Singh Sarao to Pay More than $38 Million in Monetary Sanctions for Price Manipulation and Spoofing, Release 7486-16, November 17, 2016: the $12,871,587.26 disgorgement and $25,743,174.52 civil penalty figures, and the admitted consent order's larger cumulative figures for May 6 (over four hours and twenty-five minutes of Layering Algorithm use, $170 million to over $200 million in pressure, and cancellation at 2:40 p.m. Eastern).
• Federal Reserve Bank of St. Louis, for the independently sourced cross-check series: CBOE Volatility Index, Series VIXCLS (24.91 on May 5, 2010 and 32.80 on May 6), 10-Year Treasury Constant Maturity Rate, Series DGS10 (3.58 percent on May 5, 3.41 percent on May 6), and U.S. Dollars to Euro Spot Exchange Rate, Series DEXUSEU (1.2890 on May 5, 1.2689 on May 6).

Figures deliberately not stated. This page gives no Dow Jones point figure or dollar figure for market capitalization erased during the intraday low, because the joint SEC and CFTC report describes the day's declines only in percentage terms and no primary series available for this page carries an intraday, point-level Dow Jones series for May 6, 2010. It gives no figure for Sarao's net worth, and no outcome for the DOJ's separate criminal case beyond the sentencing terms reported in contemporaneous financial press coverage, since the Department of Justice's own case pages for this matter could not be retrieved during this review. Waddell & Reed's identification as the firm behind the Sell Algorithm rests on the firm's own public statement as reported by financial press at the time, not on the joint report itself, which does not name any firm.

Method note: percentage changes for the VIX, the 10-year Treasury yield and the dollar-euro exchange rate were computed by Swoopr Investment from the daily closing values in the FRED series named above. The E-Mini's intraday low of 1056 is stated in index points, as reported in the joint SEC and CFTC investigative report, and is not adjusted for the E-Mini futures contract's separate notional multiplier.

This is educational content about a historical episode. It is not investment advice, it is not a forecast, and nothing here should be read as a claim about the current design of any exchange's trading systems or circuit breakers.

Related Reading

• Black Monday, 1987: the crash that produced the original, Dow-based market-wide circuit breakers that sat unused on May 6, 2010.
• Silicon Valley Bank and the 2023 Regional Banking Stress: a different kind of speed problem, where the fragility was fully disclosed in advance and only the timing was not.
• The GameStop Squeeze: another episode where market-structure mechanics, not company fundamentals, drove extreme short-term price moves.
• Sessions, Auctions, Halts & Volatility Controls: how circuit breakers, Limit Up-Limit Down bands and trading halts work today.
• Algorithmic Trading: how execution algorithms like the Sell Algorithm described on this page are designed and constrained.

Frequently Asked Questions