Fragile Markets — Understanding the dynamic of a Flash Crash

Financial markets can be viewed as a cosmic web. Different asset classes/regions are interconnected via the complex system of market microstructure. As such, trying to explain/model the market's behavior using simple methods like stochastic process and normal distribution usually results in a huge surprise when markets exhibit abnormal behavior, and occurrences that are not supposed to happen in our lifetime (based on the normal distribution) keep on happening on a frequent basis.

One of the most fascinating phenomena is a Flash Crash. Flash Crash is defined as a rapid decline in a security price, followed by a significant price recovery shortly after. In a Gaussian (normally distributed) world flash crashes are virtually impossible, yet the market has experienced at least 9 crashes in the last 10 years (depends on how one defines a crash). The common definition of flash crash varies, but it’s considered as a tail event (usually more than 5 standard deviations move in a period of 15–30min). To put that in perspective let’s just think about human height. The average adult man's height (globally) is 5'9 ( 175cm) with a standard deviation of 4" (10.15cm). Now think how many adult men you know are taller than 7'5 (225cm) or shorter than 4'1 (124cm). You probably don’t know too many that fit that profile, which is why these occurrences are considered to be rare. To understand how events, which are considered to be extremely rare, keep on happening, and how they affect our trading, we need to dig deep into the market dynamics, the chain of events, and the ramifications of these events on the pricing of risk and volatility.

Since 2010 financial markets have experienced 9 major flash crashes. As noted above, an event is considered to be a “flash crash” under two parameters: the magnitude of the move, and the speed of the move. My definition of a flash crash is a move larger than 5 standard deviations in a period shorter than 30min (from the beginning of the move to the end of the rebound in price). Obviously, this definition is rather vague, but the general consensus among practitioners is that about 9 events over the last 10 years constitute as Flash Crash.

2010 Flash Crash

On May 6th, 2010, during a very nervous trading day (on the back of the Greek debt crisis), the Dow Jones dropped nearly 1,000 pts. (10%) and regained 600 pts. (6%) between 14:42 and 15:07 EST.

While there have been many speculations with regards to what actually happened that triggered the move (anything from fat-finger to large directional bets, and HFT market manipulation), the SEC/CFTC investigation revealed that the crash was most likely a result of a very large selling order of E-mini S&P500 futures by a mutual fund in a very illiquid and volatile market, which exhausted all available buyers, and HFT market-makers who withdrew their bid orders and joined as sellers, which exacerbated the move. To understand the magnitude of this event, we can just look at the price action of Procter & Gamble stock price, which plunged a staggering 36% and recouped the entire loss within 10 minutes.

May 6th 2010 . SPX 1-min price action

2013 — Crashed by a Tweet

The crash of April 2013 may have been forgotten by most, as it was less significant than other crashes, but it was- nonetheless -a wake-up call for many practitioners and policymakers.

On April 23, 2013, around 12 pm EST, a tweet on the AP Twitter account suggested that explosions at the White House had injured President Barack Obama.

This, obviously, was a hoax, as it appeared that the AP Twitter account was hacked. Nonetheless, it did cause a 1% immediate drop (and rebound) of the US equity market

April 23rd 2013 SPX Price action

This event emphasized how easy it is to manipulate the market using fake news, as many High-Frequency algorithms are based on NLP and non-numerical data (i.e. Twitter or other news sources)

The Treasury Flash Crash

On Oct 15, 2014, the US Treasury market experienced its most volatile day on record, after the 10-year US Treasury yield dropped 20bp and in a matter of 30-minutes around the opening of the US equity market. The yield recovered most of the losses throughout the rest of the session (closed 6bp lower on the day), but this was a major event, as the US Treasury market is considered to be a benchmark for risk-free assets, with a high degree of stability.

The price action around the extreme move in TY futures was absolutely chaotic. the US Treasury market, which is considered to be the deepest and most liquid market in the world, was trading in a vacuum, with no bids anywhere to be found(this is a market that exhibits the tightest bid-ask spread on normal days). Needless to say that the move in TY futures sent shockwaves across all other assets classes, as multi-strategy algorithms had to balance their positions.

Until now authorities have not found a direct cause for the move, however a combination of various factors might explain how this move could occur: Sour risk sentiment (following a global growth downgrade by the IMF, and big miss of retail sales that morning), a large short TY futures position by hedge funds, unusually low order-book depth, significant increase in latency due to high messaging traffic (due to orders cancelation and suspicious activity).

Oct 15th 2014. US 10y Treasury yield move

EUR/CHF Floor removal

On Jan 15, 2015, the SNB decided to abandon the EUR/CHF 1.20 floor (which was placed about three years earlier). The SNB decision to stop defending the 1.20 floor sent the EUR/CHF almost 30% (with some stop orders triggered at 40% decline) in a matter of seconds, as no bids appeared until the SNB had to step in and provide liquidity to the market. It took the market about 15min to recover 2/3 of the plunge, however, the price action remained highly volatile throughout the rest of the day.

Jan 15th 2015. EUR/CHF price change

This extraordinary move in CHF (and other currencies) created huge losses for brokers and banks, as clients were stopped out with enormous slippage. When they were liquidated, the brokers had to absorb losses, as the clients’ accounts were empty. The main casualties of the SNB move were FX retail brokers (FXCM being the largest one), who had a very large position which they could not unwind with the market being extremely volatile.

Post Renminbi devaluation

On Monday, Aug 24 2015 the US equity market experienced one of its worst crashes on record. After very volatile trading sessions in Asia and Europe (with the Shanghai Composite index plunging over 8%), the US equity market opened with a 5.5% gap lower on extremely thin liquidity. The market did rebound throughout the session, however, it turned lower to end the session near the day’s low.

Aug 24th, 2015. SPX price change

Post Brexit GBP crash

On Oct 7, 2016, the Sterling depreciated about 6% against the USD in early Asian trading (AKA — The FX witching hour), before quickly retracing most of the move.

Oct 7th, 2016. GBP/USD price move

While the general negative GBP sentiment was governed by negative Brexit headlines, it wasn’t the main reason for the abrupt move in spot. The BIS report following the evet didn’t find a clear catalyst for the move, but rather a combination of negative sentiment paired with illiquid markets, large order imbalance, and dealers being “short gamma”. Some suspicions were made with regards to the ‘fat finger’ trade, but could not be proven.

USDJPY 2011, 2019 Flash Crashes

If there is a currency pair that is highly vulnerable to abrupt moves such as Flash Crash it’s the USDJPY. Historically speaking, the Japanese Yen has always been a risk proxy currency, as it’s considered to be a “safe haven” asset (like US Treasury bonds). Over the years, as the US-Japan long end rate differential tighten the correlation has deteriorated, nonetheless, investors tend to flee to the JPY once risk sentiment turns sour.

In the aftermath of the 2011 Fukushima nuclear disaster, on Mar 17th, 2011 the USDJPY dropped 4.5% during the twilight hours between the US and Asian trading session. Later the USDJPY recovered the entire move and finished the trading day nearly flat.

The most recent USDJPY flash crash happened on Jan 3rd, 2019, as a combination of sour risk sentiment and unusually large sell order (probably trigger/stop-loss related) in AUDJPY sent the USDJPY tumbling 4% (and the AUDJPY 6.5%) lower in a matter of 5 minutes, before recovering most of the losses.

USDJPY flash crash dynamic (based on 2011,2019)

2020 — Oil turns negative

On Apr 20th, 2020 the WTI front month future (which was expiring that day) dropped a staggering $50, to trade negative for the first time ever, as traders fleet to the exit, trying to roll their long WTI futures at any cost (as oil storages were at full capacity and no one wanted to be delivered physical oil). This move, which was partially affected by the USO ETF holdings, was a game-changer when it comes to volatility and option pricing in commodities (especially energy products), as pricing models had to account for negative securities’ price (and model assets that can alter price sign).

Apr 20th, 2020. WTI Active Future price change

A recipe for a Molotov Cocktail

As we can see, these extraordinary moves are not “once in a lifetime” events (as we got a flash crash every year on average, over the last 10 years). To understand the dynamics of a flash crash we first need to understand the market microstructure, as the combination of various factors can create a fertile ground for extreme moves.

Banking Regulation

Since the 2008 GFC bank risk limits were reduced substantially. Banking regulations like Dodd-Frank and Basel Accords require banks to hold higher Tier1 Capital (high-quality liquid assets) against their exposures. This more conservative liquidity requirement made banks and dealers warehouse less risk (which means they have very little appetite to provide liquidity — they are not in the warehousing business, they are in the moving business)

Increase in passive investment

Recent years have brought a sharp increase in leverage retail investing via a margin account. According to recent estimates, retail trading accounts for about 25% of the daily trading volume in US equities. As these accounts are leveraged accounts, in case of a large move in stock prices they are likely to be called on margin, which means that unless they can meet the margin requirements these positions will be liquidated, and exaggerate the move.

Algorithmic Trading

The algorithmic trading space, which accounts for about 80% of the order flow in financial markets, plays two significant roles, both as a liquidity provider and as executor of algorithmic-driven decision models.

As liquidity providers, HFT algorithms are selectively showing liquidity, but once markets exhibit “unfriendly” behavior for these algorithms, they will withdraw their bids/offers. Furthermore, some HFT algorithms are engaging in spoofing, which means that they will place and withdraw orders to manipulate prices.

As executors, HFT algorithms will execute orders in multiple assets across different asset classes, which means that these algorithms interconnect the different markets, and make the US equity/Treasury market the epicenter of any abnormal move (as they are the deepest and most liquid markets).

Flash Crash dynamic explained

As we understand by now, flash events should not be viewed as a standalone event (or “Black Swan”), but rather a result of a complex interaction between different participants (HFT, market makers, pension funds, etc.…) on a very fragile ground (regulatory constraints on banks’ risk limits). This complex market microstructure can create a feedback loop and explosions in volatility (i.e. abrupt spot moves).

In recent years there has been an attempt to simulate flash crashes using agent-based modeling (ABM). ABM, in short, is a class of computational models that try to simulate complex systems. Unlike Monte-Carlo simulation that relies on random/stochastic process, ABM models the interaction between the different agents (participants), based on their utility function. What makes ABM a powerful tool is its ability to identify possible “explosive” interactions between agents, which could lead to a feedback loop, and eventually a volatile move.

If we try to use ABM logic we can start to understand how the combination of fragile environment and complex interaction between different types of participants can cause these abrupt moves. A paper published in 2014 was able to simulate the May 2010’s crash under the so-called “hot-potato” effect, where HFT volume spikes with agents trying to dump their inventories as quickly as possible. This abrupt spike in trading volumes creates the feedback loop that usually takes the market into a spiral of a quick drop in price.

Monetizing a Flash Crash

Although considered to be tail events, flash crashes, due to their rapid nature (of a price drop and mean reversion), are very difficult to capture using tail hedges. As traditional tail hedge is involved owning low delta options, which are usually costly, and involved with ongoing P&L bleed, even if one captures a flash crash he/she will not be able to make up for the drawdown of the strategy.

Let’s review a theoretical strategy to capture the outsized move.

The strategy that we will use is a very basic strategy — At the end of each business day (at the NY close) we will initiate two take-profit orders in different assets (FX, US Rates, US equity indices) with a distance of 5 standard deviations from the market, based on the implied overnight move (for FX we will use the overnight ATM vol, for a listed product we will use the nearest expiry ATM vol).

For example:

USDJPY spot= 104,

o/n ATM vol =5.5% , implied move = 0.35% ->

Orders’ distance from spot =+- 1.7%

Orders that get filled are closed at the NY close (i.e., trades are not carried overnight).

For the sake of simplicity, we will build a basket of 7 assets ( G10 FX , Asia EMFX, US Equities).


Since Jan 1st, 2010 the strategy returned 50% with a 3.98 Sharpe ratio, and a Max Drawdown of 3.15%. In total 63 trades were executed.

Obviously, as we are trading outright spot, we are exposed for a continuation of the move, that said, we chose to place the orders 5stdev from the market, which makes the likelihood extremely low.

There is a lot to be explored and learned from events of a market crash, as these events become more frequent, and should affect the pricing of risk (and volatility). Understanding the environment, dynamics, and interaction between the different agents in the market will help practitioners reveal potential weaknesses in their portfolios, and perhaps allow them to profit from these short-yet-extreme periods of volatility.

Feel free to share your thoughts and comments

Twitter: Harel Jacobson


An Agent-Based Model of the Flash Crash of May 6, 2010, with Policy Implications

The October 2014 United States Treasury bond flash crash and the contributory effect of mini flash crashes




Global Volatility Trading. Python addict. Bloomberg Junkie. Amateur Boxer and boxing coach (RSB cert.)!No investment advice!

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

Towards a New Model: A Potential Solution to Refugee Integration and Empowerment

The FED keeps interest rate unchanged

Nobody is making fun of deficit hawks anymore

Investors see farmed meat as a high risk sector

Colorado wage theft bill headed to governor’s desk

Extending HECS-style loans during the COVID-19 crisis is a bad idea

Trading Basics 101

How can the economy look like post-Covid-19?

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Harel Jacobson

Harel Jacobson

Global Volatility Trading. Python addict. Bloomberg Junkie. Amateur Boxer and boxing coach (RSB cert.)!No investment advice!

More from Medium

Examining the Drawdown Risk of Sector ETFs —  2022

Asset Allocation — Applications of Markowitz Techniques -Part 1

The Math of buying Index Fund

Value at Risk for your Household