Packet Path/

The 2010 Flash Crash: Five Minutes That Erased — Then Restored — the Market

On the afternoon of May 6, 2010, U.S. stock and futures markets — already down more than 4% on a turbulent day — fell another 5 to 6 percent in a matter of minutes, then largely recovered just as fast. No single firm failed. Instead, a large sell program run by an algorithm that watched only trading volume, and ignored price, collided with thinning liquidity and high-frequency traders passing positions back and forth like a hot potato. Buy-side depth in the main index future collapsed from nearly $6 billion to about $58 million. This is the documented story, from the joint staff report of the U.S. Commodity Futures Trading Commission (CFTC) and the Securities and Exchange Commission (SEC).

Case Studies · Intermediate · 13 min · November 11, 2026

Editorial illustration on a dark teal background: a stock-market index line holding steady, then plunging sharply into a deep V and climbing straight back up, with a small five-second pause marker at the bottom of the V.

The short answer

On May 6, 2010, U.S. markets were already having a bad day. News about the European debt crisis had pushed the major indices down more than 4 percent from the previous day's close. At 2:32 p.m. Eastern Time, a large fundamental trader — described in the official report only as a mutual fund complex — started an automated program to sell 75,000 E-Mini S&P 500 futures contracts, worth about $4.1 billion, as a hedge (an offsetting position meant to protect against losses elsewhere). The selling program used an algorithm set to target 9 percent of the previous minute's trading volume, without regard to price or time. It finished in about 20 minutes. High-frequency traders first absorbed the selling, then rushed to shed it, passing positions back and forth so fast that buy-side depth in the E-Mini collapsed to about $58 million — less than 1 percent of its morning level. The E-Mini fell more than 5 percent in about four and a half minutes, the selling spilled into stocks, and some individual shares traded at one cent, and others at $100,000, on placeholder quotes. At 2:45:28 p.m. an automatic five-second pause in the futures market broke the fall. By about 3:00 p.m. most securities were back near their earlier values, and the indices closed about 3 percent down. Afterwards, more than 20,000 trades in over 300 securities were cancelled as clearly erroneous. One honesty note before the detail: this was a staff report — the report itself states that the Commissions expressed no view on its analysis, findings or conclusions.

The company: what America's stock and futures markets did

There is no single company at the centre of this case. The "company" is the U.S. market itself — and that is exactly why the case matters for engineers. Two linked instruments carried the shock.

The first is the E-Mini S&P 500 futures contract (the "E-Mini"). A futures contract is an agreement to buy or sell something at a set price on a future date; the E-Mini tracks the S&P 500 stock index and trades electronically on the Chicago Mercantile Exchange (CME). It is one of the most actively traded index products in the world, which makes it a place where big investors quickly adjust their overall market exposure.

The second is SPY, an exchange-traded fund (ETF — a fund that trades on an exchange like a single stock) that also tracks the S&P 500, traded in the stock market. Because the E-Mini and SPY track the same index, traders called cross-market arbitrageurs constantly buy whichever is momentarily cheaper and sell the other. That link normally keeps the two prices in step. On May 6 it also carried the pressure from futures straight into stocks — a coupling every network engineer will recognise: two systems that are supposed to correct each other can also drag each other down.

Liquidity is the word the whole case turns on, so here it is plainly: liquidity, in this report, means market depth — the resting buy and sell orders sitting ready to trade at prices at or near the current price. Deep markets can absorb a large sell order without the price moving much. Thin markets cannot: the same order eats through the available buyers and the price drops to find the next one.

The problem: a stressed day, thinning depth, and an algorithm that ignored price

Three conditions lined up before anything "broke."

First, the day was already turbulent. May 6 opened to unsettling news about the European debt crisis, and concern about a possible Greek government debt default pushed up the price of protection against that default. At about 1:00 p.m. the euro began a sharp decline against the U.S. dollar and the Japanese yen, and volatility pauses in individual stocks on the New York Stock Exchange began rising well above average. By about 2:30 p.m., the S&P 500 volatility index (the VIX, a widely watched gauge of expected market swings) was up 22.5 percent from the opening level, investors were moving into ten-year Treasury bonds in a flight to quality, and selling pressure had pushed the Dow Jones Industrial Average down about 2.5 percent. The major indices were already down over 4 percent from the prior day's close.

Second, the buyers were already thinning out. Buy-side depth in the E-Mini had fallen from nearly $6 billion early in the morning to $2.65 billion — a decline of 55 percent. SPY depth had fallen about 20 percent, from about $275 million to about $220 million. In plain terms: the cushion that normally absorbs large orders was already more than half gone in the futures market before the large sell program even started.

Third, the sell program was built to ignore exactly the signals that mattered. A customer with a very large order normally has choices: hand it to an intermediary, enter orders manually, or use an automated execution algorithm that can weigh price, time and volume. This trader's previous large sell program — in the twelve months before May 6, only two single-day sell programs of equal or larger size had been executed in the E-Mini, one of them by this same trader — had used a mix of manual trading and algorithms that considered price, time and volume, and had taken more than five hours to execute the first 75,000 contracts. On May 6, the chosen Sell Algorithm considered volume only: feed orders in at 9 percent of the previous minute's volume, without regard to price or time, until 75,000 contracts were sold. Why this matters: an algorithm that targets a share of volume has a dangerous feedback loop built into it. If other traders start trading frantically — for any reason — measured volume rises, and the algorithm reads that as room to sell faster, even if the extra volume is not real buying capacity at all.

An analogy: imagine draining a reservoir through a valve that opens wider whenever the outflow speeds up. On a calm day that works. On a day when everyone else is also yanking their valves open and the reservoir is already half empty, the valve does not protect the water level — it empties the tank.

The cushion disappearing: E-Mini buy-side depth Early morning — nearly $6 billion About 2:30 p.m. — $2.65 billion (down 55%) 2:45 p.m. — about $58 million (less than 1% of the morning) The animated third bar drains from the 2:30 level toward almost nothing — what the order book actually did between 2:32 and 2:45 p.m. as the sell program and the hot potato ran together. Depth is the buyers who are already there. When it is gone, price — not volume — is what moves.

What this shows: the three depth figures the joint report gives for the E-Mini — nearly $6 billion early in the day, $2.65 billion by about 2:30 p.m., and about $58 million at the low point. The market did not go from healthy to broken in one step; the cushion thinned first, then vanished.

What happened on May 6, 2010

All times are Eastern Time, as given in the joint report.

May 6, 2010 — the afternoon in Eastern Time 2:32 p.m. — a 75,000-contract sell program starts About $4.1 billion, run by an algorithm targeting 9% of volume, ignoring price and time 2:41–2:44 p.m. — high-frequency traders flip from absorbing to selling About 2,000 contracts sold while nearly 140,000 trade; the E-Mini falls about 3% in four minutes 2:45:13–2:45:27 p.m. — the hot potato Over 27,000 contracts pass between high-frequency traders; net buying is only about 200 2:45:28 p.m. — a five-second automatic pause CME Stop Logic halts the E-Mini; trading resumes at 2:45:33 and prices stabilise, then recover About 2:51 p.m. — the sell program finishes It kept executing on the way back up, completing the 75,000 contracts in about 20 minutes By about 3:00 p.m. — most prices are back Most securities revert to consensus values; indices close about 3% down on the day

What this shows: the whole event, from the start of the sell program to the recovery, took under half an hour — and the decisive fall took about four and a half minutes.

2:32 p.m. — the sell program starts. Against a backdrop of unusually high volatility and thinning liquidity, the large fundamental trader began selling 75,000 E-Mini contracts as a hedge to an existing equity position. The execution of this program was, the report notes, the largest net change in daily position of any trader in the E-Mini since the beginning of the year. The Sell Algorithm fed orders in targeting 9 percent of the trading volume calculated over the previous minute — and, crucially, without regard to price or time.

The first phase — absorption. At first, the market did its job. The selling was initially absorbed by high-frequency traders (HFTs — firms whose automated systems trade very rapidly, usually holding positions only briefly) and other intermediaries in the futures market, by fundamental buyers, and by cross-market arbitrageurs, who bought E-Mini contracts and simultaneously sold SPY or individual stocks to stay balanced. HFTs built up a temporary net long position of about 3,300 contracts. That number matters, because the report explains that HFTs typically trade a very large number of contracts without accumulating an aggregate inventory beyond about three to four thousand contracts in either direction. The shock absorbers had a hard limit, and they were already near it.

2:41–2:44 p.m. — the absorbers become sellers. Between 2:41 and 2:44 p.m., HFTs aggressively sold about 2,000 contracts to reduce those temporary long positions. At the same time they traded nearly 140,000 contracts — over 33 percent of total volume. And here the feedback loop closed: the Sell Algorithm responded to that increased volume by increasing the rate at which it fed orders into the market, even though the orders it had already sent were arguably not yet fully absorbed. The combined selling pressure from the Sell Algorithm, HFTs and other traders drove the E-Mini down approximately 3 percent in just four minutes. Cross-market arbitrageurs who did buy the E-Mini simultaneously sold equivalent amounts in the stock market, driving SPY down approximately 3 percent in the same window.

2:45:13–2:45:27 p.m. — the hot potato. Still lacking sufficient demand from fundamental buyers or arbitrageurs, HFTs began to quickly buy and then resell contracts to each other, generating what the report calls a "hot-potato" volume effect, as the same positions were rapidly passed back and forth. In those fourteen seconds, HFTs traded over 27,000 contracts — about 49 percent of total trading volume — while buying only about 200 additional contracts net. Almost half the market's activity was the same risk being handed around, adding volume without adding buyers. Buy-side depth in the E-Mini fell to about $58 million, less than 1 percent of its morning level, and the E-Mini price dropped a further 1.7 percent in those fifteen seconds to its intraday low of 1056. Measured across the four and a half minutes from 2:41 p.m. to 2:45:27 p.m., the E-Mini had fallen by more than 5 percent and SPY by over 6 percent. Between 2:32 and 2:45 p.m., the Sell Algorithm alone had sold about 35,000 of its 75,000 contracts, worth about $1.9 billion.

2:45:28 p.m. — five seconds that mattered. At 2:45:28 p.m., trading in the E-Mini was paused for five seconds when the CME's Stop Logic Functionality was triggered, in order to prevent a cascade of further price declines. Stop Logic is an automatic pause built into the futures exchange — the closest thing this story has to a circuit breaker actually firing. In that short pause, sell-side pressure partly eased and buy-side interest increased. When trading resumed at 2:45:33 p.m., prices stabilised, and shortly afterwards the E-Mini began to recover, followed by SPY. The Sell Algorithm continued executing until about 2:51 p.m. — now into rapidly rising prices — and completed its program in about 20 minutes in total.

At the same time, in individual stocks — the second liquidity crisis. The report describes a second, linked crisis in single stocks at about 2:45 p.m. Based on interviews, many liquidity providers' automated systems temporarily paused by design when prices moved beyond pre-set thresholds, so humans could assess conditions. Participants then weighed whether the moves might be bad data, what the moves did to their risk and position limits and their profit and loss, and whether trades might later be broken. On their individual assessments, some market makers widened their quoted spreads, others reduced the liquidity they offered, and a significant number withdrew completely; some fell back to manual trading but could only cover a subset of securities against a nearly ten-fold increase in volume. Retail stop-loss orders (standing instructions to sell if a price falls to a set level) were triggered into that thinning market. The result was trades at irrational prices: as low as one penny and as high as $100,000. Those extreme prints came from stub quotes — placeholder quotes that market makers (or exchanges on their behalf) post far away from the current market to satisfy their obligation to quote both a buy and a sell price continuously, even when they have withdrawn from active trading. A stub quote is never meant to be traded at. On May 6, with real quotes gone, some of them were.

How the pressure spread: futures first, stocks second SELL ALGORITHM 75,000 contracts, targeting 9% of volume, ignoring price E-MINI FUTURES Down about 3% in four minutes, then over 5% total HOT POTATO + ARBITRAGE HFTs pass positions back and forth; arbitrageurs sell SPY SPY AND STOCKS SPY down over 6% at the low; providers pause or withdraw STUB-QUOTE TRADES Placeholder quotes get hit: trades at $0.01 and $100,000 No link in this chain was an attack or a single broken firm. Each step was a system doing what it was designed to do — into a market that had run out of buyers willing to stand in the way.

What this shows: the propagation path the report documents — selling that started in one futures contract reached individual stocks in minutes, through the very arbitrage links that normally keep prices aligned.

The damage

There is no single loss figure for May 6, and this page will not invent one. The damage the report documents is specific, and it is mostly about trust in prices.

The broken trades. Between 2:40 and 3:00 p.m., approximately 2 billion shares traded, worth more than $56 billion. Over 98 percent of those shares executed within 10 percent of their 2:40 p.m. value. But over 20,000 trades — many of them based on retail customer orders — across more than 300 securities, including many ETFs, executed at prices 60 percent or more away from their 2:40 p.m. prices. After the close, the exchanges and FINRA (the Financial Industry Regulatory Authority, the broker-dealer regulator) met and jointly agreed to cancel — "break" — all such trades under their clearly erroneous trade rules. If you traded that afternoon, whether your trade stood depended on a rulebook threshold decided after the fact. That uncertainty is itself part of the damage: participants told investigators that the possibility of broken trades was one of the things they had to weigh, in real time, while deciding whether to keep providing liquidity.

The vanishing prices. Many of the almost 8,000 individual stocks and ETFs traded that day fell 5, 10 or even 15 percent before recovering most or all of their losses. Some moved far more severely, both up and down, executing at a penny or less, or as high as $100,000 a share on stub quotes, before returning to their pre-crash levels. For anyone whose stop-loss order, margin check or risk limit referenced those prices, the fact that the prices were "wrong" for only a few minutes did not make them harmless.

The close. By the end of the day the major futures and equity indices had recovered to close at losses of about 3 percent from the prior day — a bad day, but not the catastrophe the intraday chart had briefly shown. And one regulatory fact frames the whole event: the existing market-wide circuit breakers — the automatic halts meant to stop a market-wide fall — were not triggered at all on May 6. The safety system calibrated for the previous era's crashes watched a 5-to-6-percent fall in minutes and did nothing, because its thresholds and reference points were not built for this speed.

How they recovered

Step one: the five-second pause. The immediate recovery began with the CME Stop Logic pause at 2:45:28 p.m. Five seconds is nothing to a human. To the feedback loop it was everything: selling pressure eased, buy-side interest returned, and when trading resumed at 2:45:33 p.m. prices stabilised and then recovered, with SPY following the E-Mini back up. The lesson the industry took from those five seconds — that a short, automatic pause can break a self-reinforcing fall — shapes the fix section below.

Step two: participants checked their systems and came back. The severe dislocations in individual securities were fleeting, the report says, because as participants had time to react and verify the integrity of their data and systems, buy-side and sell-side interest returned and orderly price discovery began to function again. By approximately 3:00 p.m., most securities had reverted to trading at prices reflecting true consensus values. Note what actually restored confidence: not an announcement, but verification — traders convinced themselves their data was real and the fall had no hidden catastrophic cause, and only then stepped back in.

Step three: the broken trades were unwound by agreement. After the close, the exchanges and FINRA jointly broke the more than 20,000 trades executed at prices 60 percent or more from their 2:40 p.m. values, under the clearly erroneous rules. That cleaned up the most extreme outcomes — and, as the fix section shows, regulators then moved to make those break rules clearer in advance, so the next event would not require an after-the-fact meeting to decide which trades counted.

Step four: the investigation itself. The staffs of the CFTC and SEC reconstructed the afternoon from market data and interviews and published their findings on September 30, 2010, building on a preliminary report of May 18, 2010. The reconstruction is the reason this case can be told with second-level timestamps at all — and the difficulty of assembling it fed directly into later proposals for a consolidated audit trail across markets.

The fix: technology and controls that came out of it

The joint report itself was deliberately narrow: it drew lessons and noted that the staffs were working with the markets to consider recalibrating the market-wide circuit breakers. The concrete controls came from the SEC and the exchanges afterwards, and they map one-to-one onto the failure points above. They are sourced here to the SEC's own releases, not to the 2010 report.

Single-stock circuit breakers first, then limit up-limit down. In June 2010, weeks after the event, the SEC approved on a pilot basis single-stock circuit breakers — five-minute trading pauses when an individual stock moved sharply within five minutes — initially for S&P 500 stocks and later extended. On May 31, 2012, the SEC approved their replacement: a limit up-limit down mechanism that prevents trades in an individual listed stock from occurring outside a price band set as a percentage above and below the stock's average price over the immediately preceding five minutes — 5 percent for stocks in the S&P 500 and Russell 1000 indices and certain exchange-traded products, 10 percent for other listed securities, with the bands doubled during opening and closing periods and wider bands for securities priced at $3 or less. If trading cannot occur within the band for more than 15 seconds, a five-minute trading pause follows. Every trading centre — exchanges, alternative venues and broker-dealers trading internally — had to build policies and procedures to prevent trades outside the bands and to honour the pauses. The exchanges and FINRA were to implement the changes by February 4, 2013, for a one-year pilot. In engineering terms: the stub-quote trade at $100,000 becomes mechanically impossible, because no trade may print outside the band at all.

Market-wide circuit breakers, recalibrated for speed. The same 2012 approval updated the market-wide halts that had sat unused on May 6: the decline thresholds needed to trigger a halt were reduced to 7, 13 and 20 percent from the previous 10, 20 and 30 percent; halts that do not close the market were shortened to 15 minutes from 30, 60 or 120 minutes; the trigger reference became the broader S&P 500 Index instead of the Dow Jones Industrial Average; and the thresholds are recalculated daily instead of quarterly, with a simpler two-period structure around 3:25 p.m. The design trade is explicit: lower thresholds and shorter halts, tuned to a market that can fall 5 percent in four minutes and recover in fifteen.

Stub quotes effectively prohibited, break rules clarified. The SEC's 2012 release lists the companion measures taken in the wake of May 6: new exchange and FINRA rules clarifying how and when erroneous trades would be broken, and new rules strengthening minimum quoting standards for market makers that effectively prohibit stub quotes in U.S. equity markets. A placeholder quote far from the market is no longer an acceptable way to satisfy a quoting obligation.

Controls at the door, and a record of who is large. The same list names a rule requiring broker-dealers to have risk controls in place before providing customers with access to the market, and a large-trader reporting system to improve the regulators' ability to identify large participants and collect their trading information. The SEC staff also considered a consolidated audit trail to track orders and trades across the whole market system — the direct answer to an investigation that had to piece one afternoon together from fragmented records.

The lesson the report itself insisted on. Underneath the rule changes sit the report's own two sentences, which deserve quoting in substance: the automated execution of a large sell order can trigger extreme price movements under stressed conditions, especially if the algorithm does not take prices into account; and, especially in times of significant volatility, high trading volume is not necessarily a reliable indicator of market liquidity. Every control above is machinery built around those two observations — price-aware execution, bands that reference recent prices, and pauses that fire on price movement rather than on volume.

What this means for you

HFT / low-latency engineers

This is your case. Three specifics should sting. First, the inventory limit: HFTs absorbed the selling only until their aggregate long position reached roughly the 3,000-to-4,000-contract band they normally allow themselves, and then the absorbers became aggressive sellers within the same three minutes — a risk limit doing exactly what it was designed to do, at exactly the moment the market most needed the opposite. Know what your limits do to the system, not just to your book. Second, the feedback signal: the Sell Algorithm's volume target is the purest documented example of a control loop measuring the wrong variable — activity instead of capacity. Any execution or throttling logic you build that keys off throughput without a price, depth or loss guard has the same shape. Third, the pause worked: five seconds of Stop Logic did what no participant's individual risk check could. Design your own pauses — rate limiters, kill switches, self-imposed halts on implausible prices — to be automatic, short and survivable, and test them the way the CME's was tested: by firing in the real event.

Enterprise network engineers

Your version of this loop runs in every automated system that reacts to its own output: a monitoring tool that restarts a service, which floods the logs, which triggers more restarts; a load balancer health check that removes nodes as load rises, concentrating traffic on the survivors until they fail the check too. May 6 is the canonical argument for three habits: make automation price-aware — in your terms, make it state-aware, reacting to the health of the thing it is acting on and not only to the rate of requests; put a band around automated action, the way limit up-limit down puts a band around price, so no single automated decision can travel unlimited distance from the last known-good state; and rehearse the five-second pause — a deliberate, automatic halt that gives humans and systems time to verify data before resuming, instead of every component improvising its own withdrawal at once.

Hyperscaler engineers

You run the cross-market problem at fleet scale: tightly coupled services where an arbitrage-like mechanism — retries, rebalancing, failover — keeps two things in step in normal times and transmits the shock between them under stress. The hot potato is your retry storm: the same request passed rapidly between healthy-looking nodes, generating enormous measured activity and almost no net progress, while every dashboard shows volume and calls it throughput. The report's sentence translates directly: high request volume is not a reliable indicator of capacity. Measure net completions and remaining headroom (your depth), shed load before the cushion reaches 1 percent, and remember the market-wide breakers that never fired — a protection calibrated for the previous generation's failure speed is decoration. Recalculate your thresholds against the fastest failure you can actually have, not the slowest you have historically seen.

Try it yourself

Two free Packet Path tools build the instincts this case rewards. The Microburst Buffer Planner shows how a sudden burst overwhelms a finite buffer in microseconds — the network twin of an order book whose depth looks ample until a burst arrives faster than it can drain. Run a burst through it and watch how quickly "plenty of capacity on average" becomes "nothing left at the moment that matters," which is precisely the $6 billion-to-$58 million story told in packets. Then use the Tick-to-Trade Budget Calculator to lay out where time actually goes in a trading path — feed, decision, order, exchange — because May 6 was decided in windows of five and fifteen seconds: understanding your own latency budget is how you know whether your pause, your limit check or your kill switch can possibly fire inside the window where it would still help.

Sources

Primary source — U.S. Commodity Futures Trading Commission and U.S. Securities and Exchange Commission staffs, Findings Regarding the Market Events of May 6, 2010, Report of the Staffs of the CFTC and SEC to the Joint Advisory Committee on Emerging Regulatory Issues (September 30, 2010). All incident mechanics, times, depth figures, contract counts, trade counts and lessons on this page are taken from this report, including its statement that the Commissions expressed no view regarding its analysis, findings or conclusions. Official posting referenced in the report: https://www.sec.gov/spotlight/sec-cftcjointcommittee.shtml — full report text: http://www.sec.gov/news/studies/2010/marketevents-report.pdf

U.S. Securities and Exchange Commission, press release 2012-107, SEC Approves Proposals to Address Extraordinary Volatility in Individual Stocks and Broader Stock Market (May 31, 2012) — the limit up-limit down mechanism (price bands, 5% and 10% levels, the 15-second / five-minute pause rule), the updated market-wide circuit breakers (7%, 13% and 20% thresholds, 15-minute halts, S&P 500 reference, daily recalculation, implementation by February 4, 2013), and the list of additional post-May 6 measures (erroneous-trade rule clarification, minimum quoting standards effectively prohibiting stub quotes, market-access risk controls, large-trader reporting, and the consolidated audit trail under consideration): https://www.sec.gov/newsroom/press-releases/2012-2012-107htm

U.S. Securities and Exchange Commission, Investor Bulletin: New Measures to Address Market Volatility (Office of Investor Education and Advocacy) — the June 2010 pilot approval of single-stock circuit breakers after May 6, 2010 (five-minute pauses on sharp five-minute moves, first applied to S&P 500 stocks and later extended to all National Market System securities), and their replacement by limit up-limit down: https://www.sec.gov/investor/alerts/circuitbreakersbulletin.htm

No other sources are used for facts on this page. The joint report does not name the mutual fund complex that ran the sell program, and this page does not name it either. Later, separate investigations that are not part of the 2010 joint report are not presented here as causes found by that report.

Key takeaways

Keep reading