Flash Crash of 2010
Prerequisites
- order-books — CLOB mechanics, price-time priority
- trading-fundamentals — market makers, liquidity, spreads
- matching-engine-system-design — how exchanges process orders
On May 6, 2010, the Dow Jones Industrial Average fell approximately 998 points (~9%) in about 36 minutes, briefly erasing nearly $1 trillion in market value, before recovering almost entirely within 20 minutes. Individual stocks briefly traded at prices of $0.01 or $100,000. It was the most violent intraday move in US equity market history.
The proximate cause
A large mutual fund (later identified as Waddell & Reed) initiated a sell program for E-mini S&P 500 futures contracts — approximately $4.1 billion in notional value. The program used a volume-weighted algorithm (VWAP — an execution algorithm that spreads orders across time proportional to historical volume patterns) that targeted a percentage of observed volume, without regard to price or time.
Under normal conditions, this size could be absorbed over hours. On May 6, the algorithm completed in approximately 20 minutes because the feedback loop amplified itself: as the algorithm sold, volume increased (HFTs responding), which caused the algorithm to sell faster (targeting a percentage of now-higher volume), which increased volume further.
Transmission mechanism: HFT hot-potato trading
High-frequency trading firms initially absorbed the sell flow — buying E-mini contracts from the algorithm. But HFT firms do not hold inventory; they immediately re-sold contracts to other HFTs in a “hot potato” pattern. Each firm held the position for seconds or less before passing it on. The SEC/CFTC joint report documented that HFTs traded over 27,000 E-mini contracts during the crash, but their net position change was approximately zero — they added no new liquidity, only recycled existing flow at increasingly distressed prices.
As prices fell, HFTs began withdrawing from the market entirely (pulling quotes from the order book). Market depth collapsed. The selling pressure, no longer absorbed by any resting liquidity, transmitted from futures to individual equities via index arbitrage (automated strategies that buy/sell ETFs and their underlying stocks to keep prices aligned with futures).
Impact
- Dow Jones fell ~998 points (9.2%) in 36 minutes
- ~$1 trillion in equity value briefly erased
- Individual stocks traded at absurd prices: Accenture at $0.01, Sotheby’s at $100,000
- Recovery was nearly complete within 20 minutes of the low
- Approximately 20,000 trades across 300+ securities were later cancelled (“broken”) by the exchanges under their clearly erroneous trade rules
Why existing protections failed
Pre-2010 circuit breakers were designed for a different era:
- Single-stock halts required a 10% move sustained for 5 minutes before triggering — far too slow for millisecond-scale events
- Market-wide breakers (inherited from the 1987 crash rules) triggered only at 10%/20%/30% declines, measured from the previous day’s close — the 9% drop didn’t reach the first threshold
- No individual stock price bands existed to prevent trades at $0.01 or $100,000
Regulatory aftermath
Limit Up / Limit Down (LULD) — effective 2013
LULD replaced the blunt single-stock halt rules with dynamic price bands:
- A reference price is calculated as the average transaction price over the preceding 5-minute window
- Price bands are established at +/-5% for Tier 1 securities (S&P 500, Russell 1000, selected ETPs) and +/-10% for Tier 2 during normal trading hours
- If a trade would occur outside the band, the exchange enters a Limit State — a 15-second pause during which only limit orders within the band are accepted
- If the Limit State is not resolved within 15 seconds, a 5-minute trading halt is triggered for that security
Market-Wide Circuit Breakers (revised 2013)
The thresholds were lowered and tied to the S&P 500 (not the Dow):
| Level | S&P 500 decline | Action |
|---|---|---|
| Level 1 | 7% | 15-minute halt (if before 3:25pm ET) |
| Level 2 | 13% | 15-minute halt (if before 3:25pm ET) |
| Level 3 | 20% | Halt for remainder of trading day |
Consolidated Audit Trail (CAT)
The SEC mandated a comprehensive trade surveillance system to reconstruct market events across all venues — the Flash Crash investigation took months partly because regulators had to manually stitch together data from dozens of independent exchange and broker-dealer systems.
Lessons for matching engine design
- Circuit breakers are per-instrument, not just market-wide. The matching engine must check price bands on every fill, not rely on external monitoring.
- Liquidity can vanish faster than any human can respond. The engine must handle transitions from deep books to empty books gracefully (reject market orders that would cross beyond bands).
- Feedback loops between algorithm types create emergent behavior. The VWAP algorithm and HFT market makers created a positive feedback loop that neither was designed for individually.
- “Broken trade” rules are a symptom of insufficient pre-trade controls. Cancelling trades after the fact damages market confidence. LULD prevents the problem at the source.
See also
- matching-engine-system-design — circuit breaker implementation in the matching engine
- exchange-fairness-and-access — how market structure contributed to the conditions enabling the crash
- knight-capital-2012 — another 2012 market structure failure with different root cause