High Frequency Trading (HFT)



Recently, trading speeds on stock markets have increased dramatically as technology has been brought to bear on decision making as well as the matching and execution of trades. High speed computers can now gather data, make a decision by applying complex algorithms and execute a trade, all in much less than 1,000th of the time taken to blink an eye.

The growth in HFT over the past 6 years has been extraordinary, and HFT trades now make up more than 50% of all share transactions in the USA.

This article by Charles Duhigg published by The New York Times on July 23, 2009 provides a good example of how HFT operates.

“Stock Traders Find Speed Pays, in Milliseconds.”

“It was July 15, and Intel, the computer chip giant, had reporting robust earnings the night before. Some investors, smelling opportunity, set out to buy shares in the semiconductor company Broadcom. (Their activities were described by an investor at a major Wall Street firm who spoke on the condition of anonymity to protect his job.) The slower traders faced a quandary: If they sought to buy a large number of shares at once, they would tip their hand and risk driving up Broadcom’s price. So, as is often the case on Wall Street, they divided their orders into dozens of small batches, hoping to cover their tracks. One second after the market opened, shares of Broadcom started changing hands at $26.20.

The slower traders began issuing buy orders. But rather than being shown to all potential sellers at the same time, some of those orders were most likely routed to a collection of high-frequency traders for just 30 milliseconds — 0.03 seconds — in what are known as flash orders. While markets are supposed to ensure transparency by showing orders to everyone simultaneously, a loophole in regulations allows marketplaces like NASDAQ to show traders some orders ahead of everyone else in exchange for a fee.

In less than half a second, high-frequency traders gained a valuable insight: the hunger for Broadcom was growing. Their computers began buying up Broadcom shares and then reselling them to the slower investors at higher prices. The overall price of Broadcom began to rise.

Soon, thousands of orders began flooding the markets as high-frequency software went into high gear. Automatic programs began issuing and cancelling tiny orders within milliseconds to determine how much the slower traders were willing to pay. The high-frequency computers quickly determined that some investors’ upper limit was $26.40. The price shot to $26.39, and high-frequency programs began offering to sell hundreds of thousands of shares.

The result is that the slower-moving investors paid $1.4 million for about 56,000 shares or $7,800 more than if they had been able to move as quickly as the high-frequency traders.
Multiply such trades across thousands of stocks a day, and the profits are substantial. High-frequency traders generated about $21 billion in profits last year, the Tabb Group, a research firm, estimates.”


In 1997 the NYSE stopped quoting stocks in eighths of a dollar and moved to increments of 1 cent. This reduced the revenue the market makers earn from the bid/offer spread, so the exchange began paying rebates to high-frequency brokerages if they bought shares at the best public prices.  This provided a financial incentive for the development of the high frequency, high volume, low margin approach that characterises HFT.

Meanwhile, HFT was supported by significant upgrades to trading systems in many exchanges, designed to dramatically cut transactions times.  Going even further, the extremely short period of time that it takes for a piece of information to travel from one computer to another in a network, which is known as “latency”, was shortened even more as exchanges began to rent space right next to the trading platforms in their own data centres. In consequence, many firms can now process an order in less than 3 millionths of a second.

Australia and Asia have been relative latecomers to HFT. Hong Kong is making a significant investment in the required technology, and our part of the world looks like the next region for explosive growth in this area.

It looks as though this growth will continue, with HFT not only spreading geographically but to other markets as well. HFT is now well established and growing in foreign exchange, futures and options markets, and bond markets are also joining in.


Low latency infrastructure now looks for improvements in the millionths of a second range, and “extremely low latency” is the expression used for the best of it. Low latency provides information, including about competing bids and offers, microseconds faster than higher latency systems.

In Tseung Kwan in Hong Kong, work is underway on a data centre where stocks, futures, options and currencies can be traded on computers metres away from Hong Kong Exchanges’ own systems which are used to handle trades. The cost will be high and milliseconds will be saved, but in the world of HFT, milliseconds are precious.

The concept is referred to as co-location and it is being adopted widely. In Australia, the ASX plans to dramatically expand its co-location services with a new $32 million data centre which is due to be completed in August 2011.

Financial market news is also now being formatted by firms such as Bloomberg so that it can be delivered to, and read and analysed by, computers almost instantaneously after release.  Beyond even that, algorithms are now being designed to interpret stories and to make judgements about the likely effect of those stories on market sentiment.
In 2008 Dow Jones ran ads in the Wall street journal, proud that they had managed to report an interest rate cut by the Bank of England 2 seconds faster than their competitors.  In the world of HFT, 2 seconds is a very long time.


Algorithmic trading can be applied to virtually any investment strategy, including pure speculation and trend following, passive benchmarking to replicate an index’s return or  exotically named techniques such as “Stealth”, “Iceberg”, “Dagger”, “Guerrilla”, “Sniper” and “Sniffer.

Typical HFT strategies involve a mix of high turnover of capital, very short holding times, multiple trades each day, very small returns per trade and all positions being closed out at the end of every day.

“Latency arbitrage” is a contentious strategy. It exploits knowledge that the trading system of a particular exchange is about to slow down under a processing load, providing a trader with an opportunity to set up a buy or sell order in advance. The method of ensuring that the processing load is then actually experienced is called “quote stuffing”.

Algorithmic trading strategies often attempt to reduce costs by breaking large orders into several smaller ones which are then placed into the market progressively, a method known as “iceberging”. The algorithm called Stealth tries to find the large hidden orders that hide behind tiny obvious ones (icebergs) and to profit by subverting the intent of the iceberger. In this case, we have a science fiction-like battle of wits between two computers.

Some strategies are called “gaming” because they depend on the programming skills of other traders. “Dark pools” are alternative market-places where trading is anonymous, and most orders are “iceberged”. “Sharks” “ping” small market orders into dark pools, concluding if they are filled that they have discovered an “iceberg”.

Flash Orders

Some markets allow HFT traders to look at orders for very short periods of time (of the order of 30 milliseconds) before they are shown to everyone else. This is enough time, as we have seen, for traders to conduct a transaction and turn a profit by very quickly trading shares they know will soon be in high demand. The aim is to earn small amounts per trade on very large numbers of trades, sometimes millions of times a day.

The markets defend this practice as providing liquidity. Others claim that it allows one trader to probe the market with tiny orders that are immediately cancelled to provide an opportunity others don’t have of gaining insight into the other side’s willingness to pay. On the face of it, it looks like an unfair advantage.

A turning point seems to have occurred on May 6 in the USA, when an event took place which became known as the “flash crash”. Rapidly delivered, computer generated orders were widely held to be responsible for sending the Dow Jones Industrial Average down by 1,000 points in 20 minutes. This has led many commentators to question whether there is any inherent difference between flash orders and the misbehaviour called “front-running”.

Nanex has plotted the flows of orders from HFT traders, and they form very distinctive and unusual patterns which have been given names such as “Bandsaw II” and the “Boston Zapper”.

Polarised Opinion

HFT has supporters and detractors, but the growth seems to roll on in disregard of the arguments for and against.

Fully automated markets such as NASDAQ, Direct Edge and BATS, in the US, have prospered at the expense of less automated markets such as the NYSE, providing a warning to other world markets of the dangers of being left behind.  The expansion of volumes and contraction of margins has led to economies of scale, in turn generating pressure for lower commissions and fees, and mergers or consolidation of financial exchanges.

Finally, HFT seems to benefit from and also cause, in a loop, fragmentation of markets.  Fragmentation produces diverse trading venues with slightly different trading systems, speeds and fee schedules, providing opportunities for traders to exploit the differences through their computer algorithms.

Perhaps the clearest evidence of the changes and fragmentation that dark pools, and platforms like Chi-X Europe are causing is the fact the London Stock Exchange now accounts for only 55% of trading in the stocks that comprise the FTSE 100 index. This should give the ASX and all market participants pause for thought as Australia prepares to ramp up its involvement in this area and alternative exchanges to the ASX are introduced.

You must be logged in to post or view comments.