Case: The 2010 Flash Crash

What happened, in order

The day had context before it had a crash. Equity markets were already nervous about European sovereign debt, and the index was down about **3%** in the early afternoon. The instrument of choice for hedging that fear was not a stock but the **E-mini S&P 500 futures contract**, the most liquid expression of the index in the world — thin spread, deep book, trades nearly around the clock. That is where the pressure was expressed. At **2:32 pm ET**, a large fundamental seller began executing a program of about **75,000 E-mini contracts**, roughly $4.1 billion of notional, through an algorithm set to a **9% volume participation limit and a time schedule**. That design choice is the heart of the case: the algorithm was not told to stop if the price moved against it, and it had no price limit and no volume floor. It was told to sell a quantity over a period, and it did exactly that, into a market that was already falling. The mechanics then compound. The program consumed the displayed bids in the E-mini — you can price that yourself in the lab below — and the index began to move faster. Firms that had been buying from the program stopped and joined it instead, so selling begat selling rather than meeting it. In the cash equities market, liquidity providers stepped away and, where their obligations required a quote, they entered **stub quotes**: displayed bids and offers placed absurdly far from the market precisely so that they would never trade. Buyers did return by about **3:07 pm** and most of the fall was recovered, but in the meantime equities had printed at impossible prices — a large multinational traded at **one cent**, others printed at six figures — and roughly **20,000 trades** far outside the prevailing range were later **broken** under the exchanges' clearly-erroneous rules. The timeline, in one column — Early afternoon: Index down about 3% on European debt fears · 2:32 pm: A sell program of about 75,000 E-mini contracts begins · 2:42 – 2:47 pm: Bids are consumed; index loses most of the day's remaining value ← · 2:47 pm: Trough: the index is down more than 8% · 3:07 pm: Most of the fall has been recovered · Close: Down about 3.2% — and ~20,000 trades later broken

Why it happened: three mechanisms, no villain

The official investigation identified the program as the trigger and the market's own structure as the amplifier, and the distinction is the whole lesson. Three mechanisms did the damage, and every one of them was legal, rational and in place the day before. **First, the execution rule.** The seller chose an algorithm that would keep selling on schedule regardless of price or volume. That is a defensible instruction for a large risk transfer — you are not supposed to let the market know you care — and it is catastrophic in a falling market, because the algorithm becomes a machine that supplies selling to a price that is already going down. Modern practice would use volume-aware or price-limited execution (M17); the discipline is the point, not the automation. The report is explicit that this was a legitimate hedging decision, badly executed rather than maliciously. **Second, the feedback among liquidity providers.** In the minutes after 2:32, high-frequency firms and market makers that had been net *buyers* from the program reversed and became net *sellers* — the report describes a "hot potato" effect as positions were passed among firms faster than they could be cleared. This is not conspiracy; it is inventory and adverse-selection risk responding exactly as M10 describes. A market maker who believes the price is falling has no obligation to keep buying, and the faster the market falls the more expensive every fill becomes. The very firms that provide the liquidity in calm markets are the ones who must withdraw when the information content of the flow spikes, which is the structural weakness: liquidity is a service that is available precisely when you do not need it. **Third, the stubs.** When liquidity is displayed with an obligation to *quote* but no obligation to *fill*, the book can be full and empty at the same time. The stub quotes are the visible symptom: bids at a penny and offers at $100,000, placed there to satisfy a quoting requirement without risking a trade. A market order in that book does not fail; it fills at whatever is at the other end, which is why prints at one cent happened and why the exchange rules for busting trades had to be invoked afterwards. The market was working as designed — the design was the problem. The tempting lesson is "computers did it". Every participant here was a firm following its own incentives under rules that permitted them. The failure was that the rules required quotes without requiring anyone to buy.

What changed, and what did not

The regulatory response came fast and remains an unusually direct line from event to rule. Within weeks the exchanges adopted **single-stock circuit breakers**, and in **2012** the **Limit Up-Limit Down** plan replaced them with the bands you learned at M12: a stock whose price leaves its band for 15 seconds enters a limit state and then pauses for five minutes, which caps the damage a single order can do to a single name. In **2013** the **market-wide circuit breakers** were re-based from the Dow to the **S&P 500** and set at 7%, 13% and 20% — worth noticing, because on May 6 the thresholds in force were 10%, 20% and 30% on the Dow and the peak decline of about 9% never tripped any of them, so the guardrails existed and were irrelevant. **Clearly erroneous** execution rules were tightened to define how far outside the market a print must be to be cancelled, and the reporting and audit trail around large orders improved. What did not change is the underlying property: **displayed liquidity is not committed liquidity**. LULD makes an extreme print less likely to persist and buys time for information to arrive; it does not make anyone buy. The same shape has recurred — the August 2015 ETF dislocation, when thousands of U.S. names opened with prices that did not reflect value, and **March 2020**, when Treasury and corporate bond ETFs traded far from their net asset values because the underlying bonds had stopped trading (M16). In each case the market's plumbing worked and the prices were simply uninformative for a while. The durable lessons are the ones that generalise beyond any one event. Liquidity is a **service provided voluntarily by risk-taking firms**, and its availability collapses exactly when the market needs it most. **Your order type decides what you experience**: a market order or a resting stop converts a missing bid into a bad fill, while a limit order simply does not trade. And **guardrails change the timescale of a mistake, not the price**: a halt gives everyone a few minutes, which is why the order you send into a reopening (M12) is the one that decides your outcome. One more thing the report settles: retail orders were not the cause. The selling came from a professional hedging program and the amplification from professional liquidity providers. When the next crisis is blamed on retail flow, the mechanism should be demonstrated rather than asserted.

Sources

• The official narrative of May 6, 2010, minute by minuteCFTC & SEC — Findings Regarding the Market Events of May 6, 2010 (Joint Staff Report)
• Clearly erroneous execution rules and the trades that were cancelledSEC / FINRA — clearly erroneous execution rules
• Limit up-limit down and the single-stock circuit breakers that followedSEC — LULD Plan and the 2010 single-stock circuit breaker pilot
• Market-wide circuit breakers re-based onto the S&P 500SEC — amendments to Rule 80B (2013)
• Later liquidity events with the same shapeSEC / BIS — liquidity dislocations of August 2015 and March 2020