⚡ When Algorithms Broke the Market: Flash Crashes and What They Reveal
On May 6, 2010, at approximately 2:32 PM Eastern Time, the US stock market lost nearly 9% of its value in less than five minutes. The Dow Jones Industrial Average dropped 998.50 points — at the time, the largest single-day point decline in its history. Shares of major, solvent companies traded at pennies. Accenture, a global consulting firm with billions in revenue, saw its stock price fall from roughly $40 to one cent. Shares of Sotheby’s, the auction house, traded at $100,000 in a single print. Apple briefly traded at under $230 after opening above $250. Then, just as suddenly as it began, the market snapped back. Within twenty minutes, most prices had recovered to near their pre-crash levels. Nearly a trillion dollars in market value had evaporated and returned in less than half an hour. The official investigation by the SEC and CFTC took months. The final report pointed to a confluence of factors: a large sell order in the E-mini S&P 500 futures contract executed by a mutual fund using an aggressive algorithm, a subsequent liquidity vacuum as high-frequency traders withdrew from the market, and a fragmented regulatory structure that had no mechanism to halt the cascade across different exchanges. Five years later, the Department of Justice and the CFTC filed criminal and civil charges against a single trader in London — Navinder Singh Sarao — alleging that his “spoofing” activity had contributed to the crash. Sarao, operating from his parents’ house in Hounslow, had used a customized trading program to place large sell orders he never intended to execute, creating the illusion of selling pressure, then canceling the orders and buying into the resulting price decline. The Flash Crash was not an isolated event. It was a preview. 🎯 What Actually Happened: A Step-by-Step Breakdown The SEC-CFTC report identified a sequence of events that, in hindsight, reads like a blueprint for modern market fragility. Step 1: The Large Sell Order At 2:32 PM, a mutual fund initiated a sell program for 75,000 E-mini S&P 500 futures contracts, worth approximately $4.1 billion. The algorithm used to execute this trade was designed to target a volume participation rate — it would sell at a rate equal to 9% of the total market volume, regardless of price. Under normal conditions, this algorithm would have distributed the sell order over several hours, and the market would have absorbed it. On May 6, conditions were not normal. Step 2: High-Frequency Traders Absorb, Then Flee Initially, high-frequency trading firms bought the contracts being sold by the mutual fund’s algorithm. This is the normal function of HFT in the market — providing liquidity by taking the other side of large orders. But these firms do not hold positions. They buy and then immediately resell to someone else. On May 6, the “someone else” was another HFT firm, which did the same thing. The contracts began to “hot potato” between HFT firms, with each firm holding the position for fractions of a second before passing it on. When the selling pressure continued and prices began to fall, the HFT firms did what their algorithms are programmed to do when risk exceeds thresholds: they stopped buying. Some withdrew entirely from the market. Liquidity evaporated. Step 3: The Liquidity Vacuum With HFT firms gone, the only remaining buyers were fundamental traders — humans and long-term algorithms that evaluate price relative to value. But prices were moving so fast that these systems could not keep up. Stale quotes, system delays, and the sheer speed of the decline meant that buy orders that would normally have provided a floor simply did not execute. At 2:45 PM, the E-mini S&P futures had fallen over 5% in thirteen minutes. The decline then accelerated into the equities markets. Step 4: The Broken Trades As individual stocks crashed, a surreal sequence of prints appeared on the tape: Over 20,000 trades across more than 300 securities were later canceled by the exchanges, deemed “clearly erroneous” — a euphemism for “the price at which these trades executed had no relationship to any rational assessment of value.” Step 5: The Snapback At approximately 2:46 PM, buying returned. Prices rebounded almost as quickly as they had fallen. By 3:00 PM, the Dow had recovered roughly 700 of the 998 points it had lost. The market closed down 3.2% on the day — a bad day, but not a catastrophic one. The catastrophe had been compressed into minutes. 👤 The Hound of Hounslow: Sarao and Spoofing Five years after the crash, investigators uncovered something the original SEC-CFTC report had missed. Navinder Singh Sarao was a self-taught trader operating from his parents’ home in Hounslow, West London. Using off-the-shelf trading software that he had modified himself, Sarao employed a technique called spoofing — placing large orders with no intention of executing them, creating a false impression of supply or demand, and then trading on the resulting price movement. His method, called dynamic layering, worked like this: The CFTC complaint alleged that Sarao used this technique on May 6, 2010, and that his spoof orders contributed to the liquidity imbalance that triggered the crash. On that day alone, Sarao allegedly made approximately $879,000 in profits. Sarao was eventually extradited to the United States, pleaded guilty to one count of electronic fraud and one count of spoofing, and was sentenced to one year of home detention. He was also ordered to forfeit $12.8 million. The remarkable thing about Sarao’s operation was not its sophistication. It was its simplicity. A single trader, with modified software and a home internet connection, had contributed to a trillion-dollar market event. 📉 The Other Flash Crashes The 2010 Flash Crash was the most famous, but it was not the last. The years since have produced a steady drumbeat of similar events, each revealing a different vulnerability in the algorithmic market structure. The Treasury Flash Crash — October 15, 2014 The US Treasury market — the deepest, most liquid bond market in the world — experienced