The 2010 Flash Crash: an algorithm that sold faster the more trading it helped cause

transcript

A computer was told to sell 4 point 1 billion dollars. Price didn't matter. The more the market churned, the faster it sold.

6 May 2010, 2:32 p.m. A big fund starts selling 75,000 futures contracts. Its algorithm has one rule. Sell 9 percent of last minute's volume. Price? Ignored. Time? Ignored.

High-frequency traders likely buy the first orders. Then sell them on. They trade fast with each other. Volume jumps. Here's the trap. More volume tells the algorithm: sell faster. But real buyers haven't absorbed the earlier orders yet. Prices slide.

In 14 seconds, fast traders swap over 27,000 contracts. They actually buy about 200 more. A game of hot potato. Waiting buy orders shrink to under 1 percent of the morning's.

From 2:41 p.m., the futures plunge. By 2:45:27, they're down more than 5 percent. Then the twist. The exchange pauses trading. For five seconds. Selling eases. Buyers return. Prices start to recover.

In stocks, market makers' systems paused. Many firms pulled out. Over 20,000 trades went through 60 percent or more off. Some at a penny. Some at 100,000 dollars. All cancelled later.

That fund once took over 5 hours for a sale this size. This time, about 20 minutes. A loop that reacts to its own noise runs away. Give it limits. And a pause.

Full story in the article.