In the 2010 Flash Crash, an algorithm sold faster the more trading it helped cause.

6 May 2010 was already a nervous day in US markets. At 2:32 p.m., a large mutual fund complex began selling 75,000 E-Mini S&P 500 futures contracts, worth about $4.1 billion, as a hedge. Its automated sell algorithm was set to sell 9% of the previous minute's trading volume, "but without regard to price or time".

High-frequency traders (HFTs) were the likely buyers of the first orders. They then sold to trim their positions, trading rapidly with each other. Volume jumped, so the algorithm sold faster, though its earlier orders were arguably not yet absorbed by real buyers. Between 2:45:13 and 2:45:27, HFTs traded over 27,000 contracts, about half of all volume, while buying only about 200 more: a "hot-potato" effect. Buy orders waiting in the E-Mini fell to about $58 million, under 1% of that morning's level. From 2:41 to 2:45:27, the E-Mini fell more than 5%.

At 2:45:28, the CME's Stop Logic paused E-Mini trading for five seconds. Selling eased, buyers came back, and prices began to recover.

Individual stocks had their own crisis. Many market makers' systems paused themselves, and many firms pulled out. Orders to sell "at the market" hit stub quotes, placeholder prices far from reality. Between 2:40 and 3:00 p.m., over 20,000 trades in more than 300 securities went through 60% or more away from their 2:40 prices, some at a penny, some at $100,000. They were cancelled after the close.

The same trader once took more than 5 hours to sell its first 75,000 contracts, watching price and time too. On 6 May, the volume-only algorithm took about 20 minutes.

Why it matters: A loop that reacts to a signal it creates can run away; high volume, the report says, "is not necessarily a reliable indicator" of liquidity. Limit what automation may do, and build in a pause.