Flash crashes explained

What is a flash crash?

A flash crash is when the price of a security – a currency - rapidly declines over a very short period of time before quickly entering a period of recovery.

Although some investors welcome it more than others, the importance of volatility in trading is indisputable. And, in the digital age whereby trading between humans is replaced by computers trading via algorithms aimed at profiting by making millions of automated orders at minuscule margins, the importance of volatility is growing. However, every so often this volatility turns what is regarded as normal fluctuations in the price of a security into sudden and rapid decline. A typical flash crash is over before most have even noticed it has happened at all, lasting just seconds or minutes (although some flash crashes have lasted longer).

Securities that plunge in price as a result of a flash crash typically recover the majority of their value as quickly as they lost it, although many fail to immediately recover all the lost value. With the speed of decline and recovery in mind, some consider a flash crash as nothing more than an extreme burst of volatility. However, the causes of previous flash crashes and the vast sums of investor’s cash that has been lost during them suggests something else entirely.

While a flash crash usually involves a sudden decline and recovery in price, it is worth noting that the same thing can happen the other way, with prices rapidly soaring in value before quickly giving back all or most of those gains. This is less common but one of the best examples would be in currencies: as they are traded in pairs, if the price of one currency plummets because of a flash crash then another will soar in price as a result.

What causes a flash crash?

There are several reasons why a flash crash can happen, and both humans and computers play their part.

How do humans cause flash crashes?

Sometimes human error plays its role with previous crashes being caused by accidental trading, when a trader or fund manager has unintentionally added an extra zero to their order or made an order at the wrong price, often referred to as a ‘fat-finger’ mistake.

Then there are deliberate attempts by traders to manipulate the market through an illegal method known as ‘spoofing’ (sometimes also known as ‘dynamic layering’), when someone places large sell orders at a price far from the current market value and then quickly cancels them before the security hits that price. This gives the illusion that there is a large sell-off happening and prompts others to begin selling too in fear the price will decline. This quickly sees an imbalance between the amount of orders to sell compared to buy, amplifying the fall in price. The person that placed the initial sell order also has orders to buy the same security at a value much less than the market value but cancels the order to sell the security before the security hits the price that would execute it. This means they can then buy the security at the bottom of the flash crash and sell it at a considerably higher price after it recovers – potentially allowing huge profits to be made in seconds.

How do computers cause flash crashes?

The growing role of computers in trading is also a major cause of flash crashes. Software glitches can sometimes mean market data is not effectively communicated between exchanges, which can mean inaccurate prices are applied to a security.

The rise of algorithmic and high-frequency trading has also exacerbated flash crashes in the past. This involves superfast computers trading at lightning speeds based on pre-programmed algorithms. For example, a security is trading at $100 and a high frequency trading system has an algorithm in place to automatically sell that security if the price hits $95 (to minimize potential losses) or if it hits $105 (to make a profit). This means that if the price of the security does experience a dramatic fall to the $95 level, however briefly, then swathes of automated sell orders can be triggered, which in turn pushes the price lower and continues to trigger more algorithms as the prices go down.

Interestingly, these same trading systems are also largely responsible for the subsequent recovery that follows a flash crash. For example, other algorithms have ordered their systems to purchase the security if it falls below $90 (because it is regarded as cheap), so as algorithms ordered to buy the stock start to be triggered the imbalance starts to even out again and the whole thing reverses itself. The price falls so low that buyers begin to outstrip sellers and the price recovers.

Although past flash crashes have had different causes, some similarities have been witnessed among most of them. For example, many flash crashes occur when there is thin trading volumes because low liquidity means large orders can then exacerbate price movements.

Flash crash examples

Below is a series of examples that demonstrate how flash crashes can impact different securities and shows how they are often caused by several drivers.

2010 flash crash: Dow Jones

The flash crash of the Dow Jones Industrial Average (DJIA) in May 2010 saw the index drop over 1000 points in just 10 minutes, which was the biggest drop of its kind on record at the time. While US indices dropped by as much as 10%, some individual stocks plunged by much larger amounts. Overall, the flash crash is thought to have wiped off $1 trillion in equity and while the DJIA recovered, it only managed to regain about 70% of the lost value by the end of the day – demonstrating the severe impact these events can have.

The spark of this particular crash came down to one British trader named Navinder Singh Sarao. Dubbed the ‘Hound of Hounslow’ and the ‘Flash Crash Trader’, Sarao was convicted after pleading guilty to charges of spoofing and market manipulation in 2016. The Securities and Exchange Commission (SEC) said the flash crash was caused by Sarao rapidly executing large sell orders of E-mini S&P 500 futures contracts through the Chicago Mercantile Exchange. Remnants of this trial continue today.

And, while not responsible for the initial turmoil, the rapid decline in price triggered large numbers of automated trading to take place as prices broke through pre-determined thresholds. As the majority of trading is done through automated programs, most high frequency traders end up trading with other high frequency traders, all of which have their own orders and limits in place. This means when those high frequency trading orders were triggered by Sarao’s fraudulent sell orders, it went on to trigger orders from other high frequency traders – causing a downward spiral.

While regulators had already established the need to set limits on how low a security could go over a certain timeframe before having to intervene (such as suspending trade in a stock), it was only after the 2010 flash crash – which, while plunging many stocks lower also saw some soar at unbelievable rates – that new rules were introduced to set how far a security could rise in a short period of time.

2014 flash crash: US bonds

The flash crash in US Treasury bonds – known as the ‘Great Treasury Flash Crash’ – occurred in October 2014 and debate over the primary cause is still debated today. In just 12 minutes, the yield on the ten-year US Treasury bond managed to lose and then recover 1.6% and was the largest decline in a single day since 2009.

US regulators released a shallow report into the incident less than a year later which, to the frustration of some, provided a few answers but ultimately failed to attribute the event to a single cause. Some point to the fact that was twice the amount of trading volume than usual, combined with tighter liquidity as there was considerably fewer bonds for sale than usual.

However, much of the blame has been put down to high frequency traders. The price of bonds was experiencing a normal rise due to demand taking precedent over supply ahead of the flash crash. That rising price threatened to trigger the pre-determined orders of high frequency traders that had set instructions to automatically sell their bonds when the price was high enough to make a nice profit. And, while many of those orders were not ultimately triggered, they did start to become visible to other traders to paint a picture that there was a growing number of people eager to sell their bonds, which then started to reverse the price trend to push them lower once again. That, in turn, made the same thing repeat once again as more algorithmic orders kicked-in as the price spiraled lower.

The Great Treasury Flash Crash shows that high frequency and algorithmic trading is all based around chasing momentum. The report found many of the trades that were completed during the flash crash were between one high frequency trader and another, and even some of them trading with themselves. The volatility in prices meant both orders to buy and sell were triggered and, driven by programming rather than common sense, this meant high frequency traders were responsible for a large amount of the selling when the price went down and the buying when the price started to rise again. What is particularly interesting is the fact that regulators found the flash crash largely sorted itself out, suggesting the turmoil caused by algorithmic trading sorted itself out in the end.

2015 flash crash: FTSE 100

The FTSE 100 index experienced a flash crash in September 2015 when it experienced a sudden drop of over 1%. While it quickly recovered, the £15 billion or so in losses the sharp drop in prices triggered automatic trading suspensions of at least nine major constituents – including HSBC and Royal Dutch Shell.

Once again, no firm cause has been identified but it is largely accepted that the flash crash was caused by a ‘fat-finger’ mistake combined with tight liquidity.

2015 flash crash: Dow Jones (again)

The DJIA suffered yet another flash crash in August 2015 when the Dow plunged around 1100 points within the first five minutes of the trading day. This once again triggered suspensions as stock prices became too volatile, which in turn made it difficult for exchanges to correctly price indices that those stocks were included in. The S&P 500 plunged 5% within minutes of the open but managed to largely recover its losses by the middle of the day. Troubles were largely isolated to US equities listed on the New York Stock Exchange (NYSE).

Unlike prior flash crashes, this one seemed to have more innocent beginnings. There had already been a sell off in stocks on the Thursday and Friday before the flash crash on the following Monday, which some argue left investors wary of the over the weekend. Plus, Asian markets, which open before the US, plunged when trading began on the Monday and US investors followed suit later in the day. Ultimately, this culminated in a large imbalance as orders to sell outstripped those to buy, pushing prices lower. The lack of bids and the severe volatility meant trading in many stocks was suspended, which in turn made it difficult to judge the fair value of any indices or exchange-traded funds (ETFs) that they were included in.

2016 flash crash: GBP/USD

The pound dropped a staggering 6% against the dollar during overnight trade (for London investors anyway) in October 2016 from over $1.26 to as low as $1.14 before recovering and leveling back out at around $1.24 within hours, again showing that many securities fail to immediately recover the losses from flash crashes. Again, a single cause has not been identified. Some believe it was a fat-finger mistake, but others have a much more interesting theory that places the blame on algorithmic trading. Interestingly, one claim suggests a newer, more experimental (and less precise) algorithm that acts on news headlines and social media was to blame.

The major benefit of using computers to trade is the speed at which they can do it and the fact they can trade without fear of falling foul of human sentiment or emotion. However, speculation that some algorithms are acting on data that is less black and white than the numbers they usually use to operate means these computers are trying to trade on human emotion that they don’t have. In this case, one suggestion is that a rogue algorithm reacted to comments from the then French President Francois Hollande about giving UK Prime Minister Theresa May a hard Brexit, which prompted a large sell order that pushed the price low enough to trigger further pre-determined sell orders from other computers.

One of the reasons an algorithm has been labeled as the main culprit is because the flash crash occurred overnight when only markets in Asia, Australia and New Zealand were open rather than the major hubs in the UK or US. Plus, that means there was lower liquidity in GBP/USD than usual, exacerbating the price movements.

2017 flash crash: ethereum

Those that have dipped a toe into the world of cryptocurrencies will be no stranger to extreme volatility, and it is therefore unsurprising that some have already suffered from flash crashes during their relatively sort lives. In the middle of 2017, the price of ethereum on the now-defunct exchange GDAX managed to plummet from $319 to just 10 cents in a matter of seconds. Unlike many securities after a flash crash, ethereum managed to recover all of those losses and more during the same day.

At the time, GDAX attributed the flash crash to a multi-million dollar sell order pushing the price lower, again triggering hundreds of other sell orders, enough to nearly wipe the entire value of the cryptocurrency altogether.

2017 flash crash: Precious metals futures

Silver futures experienced a flash crash in July 2017 when the price of contracts due for September delivery plunged 11% around $16.15 to $14.35 per troy ounce. It occurred when US and European markets were closed, so thin trading from Asia exacerbating trading algorithms was largely blamed.

Silver futures recovered most of the lost value within hours. CME Group, which runs the Nymex exchange the futures trade on, said its markets had 'worked as designed' as the event activated its 'velocity logic' that paused trading in the market for 10 seconds to allow liquidity to return to the market.

2019 flash crash: USD/JPY and AUD/USD

One of the most recent flash crashes occurred in January 2019 and impacted foreign exchange markets. The event is thought to have been triggered by a statement from Apple that pointed toward a weakening Chinese economy which prompted traders to sell out of riskier currencies, like those of emerging markets and the Australian dollar. China is a key trading partner for countries like Australia, so any deterioration there is often swiftly felt by others. AUD benefits from good news emanating from China but when investors feel nervous about the future of the Chinese economy then they often flock to the safe haven play in Asia: the yen (JPY). As investors ploughed their money into JPY, this unwinded the JPY carry trade.

The consequence of this was a dramatic fall in AUD/JPY, which fell by as much as 7% in a matter of minutes, and as is the nature of foreign exchange, had a knock-on effect. This also meant JPY was stronger against other currencies, including the dollar, too.

Once again, this flash crash largely occurred when most markets were closed and liquidity was thin on the ground. The low trading volumes were exaggerated further because it occurred when Japan was on a bank holiday.

How can flash crashes be prevented?

Flash crashes are a phenomenon that is not fully understood. While it is clear human error can create the required spark, it is the computerized systems increasingly used to trade securities that ignite flash crashes. One of the characteristics of a flash crash is that there is a sharp price movement when there is no fundamental reason for such extreme volatility. Plus, the near light speed at which they can happen shows the crash, and often the subsequent recovery, is driven by high frequency traders using algorithms.

It also seems clear that the lack of human participation, when major markets are closed and liquidity is low, increases the role of algorithmic traders. The fact most of these computers trade with one another (and themselves) means one fat finger or incorrect bit of programming of one algorithm often triggers another algorithm, which triggers another and so on.

But the lack of true understanding about flash crashes means we are far from finding a solution that eradicates them altogether, demonstrated by the fact they keep happening regardless of what measures have been introduced by exchanges and others. The reaction from the CME Group following the flash crash in silver futures suggest two interesting points. Firstly, you need safeguard algorithms to counter trading algorithms, meaning more computers to manage the computers that trade. Secondly, if the systems did their job but the flash crash still occurred then it shows safeguards put in place are about reacting to (and minimizing the damage caused by) a flash crash rather than preventing them. One of the most popular measures introduced by exchanges such as the NYSE are circuit breakers, which halt trading when automated systems recognize a flash crash is occurring until buy and sell orders can be evenly matched up and trading can resume as normal.

The problem boils down to market structure. The dynamic of trading between two humans is vastly different to the dynamic of trade between two computer systems, with the former driven by emotion and sentiment and only capable of running for so many hours in the day, and the latter driven by technical forces and able to operate so long as a market is open. Human error often lays the ground work for a flash crash but it is computers that make it happen, implying a flaw in the relationship between human-computer trading. And yet, it is only humans that pay the price.