Algorithms and high-frequency trading
An algorithm with no price limit can crush a thin book: hence limits and circuit breakers.
Lesson 710 min
By the end of this lesson
- Distinguish algorithmic trading, high-frequency trading and simple order routing under MiFID II
- Explain the part played by the price-blind instruction and the halt on 6 May 2010
- Calculate the proceeds of a market sale that crosses several levels of the book
In this lesson3
An order is no longer necessarily decided by a human: it can be decided by a program. An algorithm is neither good nor bad: it carries out the instruction it is given, very fast and without second thoughts. What you will learn in this lesson: an algorithm that looks only at quantity can crush a thin book within moments, and that is why regulation imposes limits on firms and circuit breakers on trading venues.
What the law says
The European MiFID II directive gives two definitions (Directive 2014/65/EU, art. 4(1), points 39 and 40).
Algorithmic trading
Trading in which a computer algorithm automatically determines the parameters of orders (whether to send them, when, at what price, for what quantity, how to manage them afterwards), with limited or no human intervention. A system that only routes or confirms orders is not included.
High-frequency trading
Algorithmic trading that combines three features: infrastructure designed to minimise latency (co-location, proximity hosting or high-speed direct electronic access), no human intervention for individual orders, and high intraday message rates (orders, quotes, cancellations).
Two safeguards follow. A firm engaged in algorithmic trading must have trading thresholds and limits and prevent erroneous orders from being sent (art. 17(1)). A regulated market must be able to halt or constrain trading if prices move sharply over a short period and, in exceptional cases, cancel or correct a transaction (art. 48(5)).
From memory: what are the three features that make algorithmic trading high-frequency, according to MiFID II?
6 May 2010
The joint SEC and CFTC report tells of an afternoon when all of this came into play (SEC and CFTC, “Findings regarding the market events of May 6, 2010”, 30 September 2010):
- The major US indices, already down more than 4%, fell a further 5 to 6% within minutes, then rebounded almost as quickly.
- At 2:32 p.m., a large investor, a mutual fund complex, started selling 75,000 E-Mini futures contracts, about 4.1 billion dollars. Its algorithm targeted 9% of the previous minute's volume, without regard to price or time. It sold in 20 minutes; a comparable sale had taken more than 5 hours.
- Between 2:45:13 and 2:45:27, high-frequency firms traded more than 27,000 contracts among themselves, 49% of volume, while buying only about 200 contracts net: the “hot potato”. Buy-side depth fell below 1% of its morning level.
- At 2:45:28, an automatic CME mechanism paused the contract for 5 seconds. Prices stabilised, then recovered.
- In equities, more than 20,000 trades in more than 300 securities were executed more than 60% away from their value moments earlier, some at one cent or at 100,000 dollars: market makers that had withdrawn left fictitious prices, far from the market, to meet their obligation to quote both sides.
The report's lesson is not “algorithms are dangerous”, but: an instruction with no price limit, in a market already under stress, removes liquidity faster than it comes back. The economics section returns to the great crashes (Crashes).
The life of an algorithm that sells without looking at price
A fictional fund manager gives his algorithm a sale of 150 shares in slices of 30, at market. The book is thin below the first level. The portal's engine computes each slice; the venue halts trading if the price falls more than 5% below the reference.
An instruction without a price
A fund manager must sell 150 shares. His algorithm sends 5 slices of 30 at market, one after the other, with no price limit.
The same algorithm, with or without a limitFictional
Fictional book, buy side: 30 at 9,950, 30 at 9,900, 20 at 9,800, 20 at 9,600, 20 at 9,300. Reference 10,000. The algorithm sells slices of 30 at market.
- Slice 1. 30 × 9,950 = 298,500 XAF.
- Slice 2. 30 × 9,900 = 297,000 XAF.
- Slice 3. 20 × 9,800 + 10 × 9,600 = 196,000 + 96,000 = 292,000 XAF.
- Slice 4. 10 × 9,600 + 20 × 9,300 = 96,000 + 186,000 = 282,000 XAF. The last price, 9,300, is 7% below the reference: the venue halts. 120 shares sold for 1,169,500 XAF, i.e. 9,746 XAF on average.
- With a limit at 9,800. The algorithm sells only 30 + 30 + 20 = 80 shares, for 298,500 + 297,000 + 196,000 = 791,500 XAF, i.e. 9,894 XAF on average. It keeps 70 shares and waits for buyers to return.
Without looking back: in the example, what stopped the unlimited sale, and what could have stopped it earlier?
On the BVMAC, trading is by call auction, once a day at 11:00, according to the published hours: the race for speed does not have the same meaning there. Whether algorithmic or high-frequency trading takes place there is not confirmed. Price limits are set by BVMAC instruction (BVMAC general regulation 2025, art. 133); the student track presents them (Volatility and price limits).
Your turn
Read next: SEC et CFTC, « Findings regarding the market events of May 6, 2010 »
A question? Ask your teacher, or an AI assistant, giving it the title of this lesson.