Flash Crash

On the stock market, shares and other securities are traded, and their prices fluctuate constantly. Usually this happens slowly and in small steps. In a flash crash, something different occurs: prices plunge dramatically within just a few minutes, sometimes by several percent, sometimes by almost their entire value. Shortly afterward, they recover almost completely, as if nothing had happened. The name says it all: “flash” means a sudden burst of light, “crash” means a collapse. What’s special is that there is no bad news about a company behind it, but rather the technology of trading itself.

What a crash in minutes can do

The most famous case occurred on May 6, 2010, in the United States. The Dow Jones Index, an important average of major US stocks, lost nearly a thousand points in about five minutes. On paper, roughly one trillion dollars in company value briefly evaporated. Individual stocks traded for a single cent, even though they had cost dozens of dollars just moments before. After about twenty minutes, the spectacle was over.

For investors, this is no harmless numbers game. Many portfolios contain so-called stop-loss orders, automatic sell orders that trigger as soon as a price falls below a certain threshold. These are precisely the orders triggered during a flash crash. Anyone who sells this way realizes a real loss, even if the price returns to normal ten minutes later.

On top of that comes a trust problem. A market whose prices suddenly lose all connection to reality seems unpredictable. Regulatory authorities therefore study the topic intensively, and flash crashes are considered a textbook example of how automated systems, in aggregate, create risks that no one individually planned.

When algorithms drive each other into a spiral

Most stock trading today no longer runs through humans, but through computer programs. These programs, often called algorithms, follow fixed rules: if a price falls below a certain limit, sell. They react in fractions of a second, a thousand times faster than any human. In high-frequency trading, it’s even a matter of millionths of a second.

A flash crash arises when many such programs apply the same rule at the same time. A large sell order pushes the price down. This triggers the sell rule in other programs. These sales push the price down further, causing even more programs to sell. This is called a feedback loop: the effect amplifies its own cause.

This is made worse by a lack of buyers. Normally, there are market participants who constantly post buy and sell offers, thereby providing balance. Today, they too are programs. If they detect unusual conditions, they simply shut themselves down for self-protection. Then the price falls into a void, because no one remains on the other side. As a countermeasure, there are now circuit breakers, automatic trading halts that stop trading for minutes when price movements are too fast.

From the stock market to the crypto market

The term appears in financial news whenever an index or an individual stock collapses without any apparent reason. Currencies have already been affected, such as the British pound in October 2016, as well as government bonds. Flash crashes are especially common in cryptocurrencies, because trading volume there is lower and even medium-sized sales can trigger large price swings.

It’s important to distinguish this from a normal stock market crash. A real crash has a cause rooted in the economy, such as a financial crisis, and prices remain down for weeks. A flash crash is a technical event and is usually over within minutes. A common misconception is that flash crashes are caused by artificial intelligence. Most of the programs involved are simple rule-based systems with no capacity to learn whatsoever.

However, with the growing use of learning systems in trading, the topic is gaining new significance. Such models make decisions that are harder to retrace afterward than a fixed if-then rule. This is precisely why the flash crash is so often cited as a cautionary example in discussions about regulation.