An explanation of the principles and function of high-frequency trading

According to the latest estimates, high-frequency traders are responsible for up to seventy percent of all trades in American stock markets. On the ‘old continent’, the ratio of these ‘cheetahs of the financial world’ is approximately forty percent.

High-frequency trading is a subsection of algorithmic trading, meaning trading using computers/algorithms. In the ‘slow’ version, computer programs receive instructions from traders and seek to execute them in the best possible way in the market. Let’s say an investor wants to purchase 100,000 shares of Microsoft. Rather than making the purchase at once, the investor submits an instruction to the algorithm that will identify the best possible methods to buy the shares according to preset data. For example, it may distribute the purchase across thousands of smaller trades, so that the demand doesn’t push the price too high. This is called an algorithmic execution. These algorithms are readily available to individual traders as well.

A subsection of thus-defined algorithmic trading is algorithmic decision-making. Here, algorithms evaluate the circumstances based on predefined instructions and, in accordance with the targeted strategy, send the resulting instructions to the exchange. There are two differences as opposed to the previous situation. While in algorithmic execution, a human determines the instructions, in algorithmic decision-making the instructions are issued directly by the computer. The second difference is the fact that in the case of computer decision-making it is not clear ahead of time in which position the trader will find himself after a certain period of time elapses. While in the first described scenario, the result is 100,000 Microsoft shares in the trader’s account, in case of algorithmic decision-making his final position is a result of a calculation of ‘the best decisions’ the algorithm was able to make at the given moment within the strategy established by the trader.

High-frequency trading falls under algorithmic trading but, as its name suggests, the frequency of instructions sent (including cancellations) is high. However, even in HFT the computers don’t execute all activities on behalf of the trader. Strategy planning (in RSJ Securities’s case market making) and the evaluation of its success continue to be the responsibility of people. So do the continual monitoring, calibration and development of trading programs.

In order to understand the reasons behind its creation, we must realize that HFT represents a next evolutionary level of trading in financial markets.  

Initially, it was all simple. For example the famous Rothschild banking family based its success, among other things, on trading in gold. They noticed that its price fluctuated in various European cities. This knowledge became significant when bankers were able to trade at different markets at the same time or when they could compare gold prices faster than their competitors. Rather than the Rothschilds traveling from place to place, they developed a network of fast couriers, supplemented by carrier pigeons. Thanks to this fast messaging, Rothschild family branches in European cities knew when to buy and when to sell. Pigeons and horses were subsequently surpassed by telegraph and then by telephone. In the end, the internet was invented and, with it, the ability to transfer information and respond in fractions of seconds. But this is too large a task for the human brain, so the process was passed on to computers. High-frequency algorithmic trading was developed, whereby computers trade at dizzying speeds and almost entirely on their own.

Although the media likes to highlight the speed of HFT (“superfast traders”), speed itself is no business strategy, as is, for example, arbitrage or market making. Speed alone doesn’t assure trading success. Besides speed, high-frequency traders must depend upon their intelligence and market knowledge – they certainly can’t senselessly trade in financial markets. Behind their successes are hundreds of hours of constant evaluation of market situations, designing algorithms, transposing them into computer programs and building technology infrastructure.

Just as speedy yachts may be used for many activities (from trading to smuggling to piracy), we use high-frequency trading and the technological infrastructure it’s based upon to pursue an entire palette of various strategies. Therefore, there is not much point in viewing all high-frequency traders through the same lens. This is further supported by the fact that it’s not even easy to define HFT. The expression ‘high-frequency trading’ is about as accurate as ‘fast steamboat.’ We may imagine the type of technology that propels it, but only sense its speed. And we know nothing at all about the purpose of its voyage.

Algorithmic trading, including HFT, often focuses on arbitrage – which means taking advantage of differences in prices. That’s the same strategy as was used by the Rothschild family. The truth is that the price of the same asset should be the same in all markets, after discounting transaction costs such as transportation.

Remaining in the Rothschild’s time periods, if I buy gold in Prague for 1,000 thalers, a ticket for a coach to Vienna for another ten and I can sell the same amount of gold there for 1,500 thalers, I have discovered an excellent opportunity for arbitrage – I’ll make 490 thalers in one trip. Yet, with every such journey (either mine or my competitors’ who also noticed the opportunity) the demand for gold will increase in Prague, while in Vienna its offer will rise. The prices will become ever closer and, after a while, a trip to the Austrian metropolis will no longer make money.

In case of arbitrage, speed is an important factor. Who lingers will lose the opportunity to profit. For this reason, traders using HFT use the latest communication technologies and place their computers directly at the stock markets (co-location).

Unjustified differences in prices for the same assets are considered market inefficiencies. The example described above could be uncovered relatively easily. It’s harder to make money on inefficiencies caused by varied prices of commodities and financial products that are different, but mutually dependent in terms of price. And that’s most of them these days. For example, if the exchange rate of the Czech Crown against the Euro changes, it should reflect in the price of the option that entitles me to buy Euros in a month. A number of arbitrageurs (that’s the name for the traders using such differences) focus on questions regarding how the prices of different assets affect each other, or on the basic problem of whether the assets are correctly valued. 

A typical and logical characteristic of market inefficiencies is that, over time, they level and disappear. Therefore, traders must focus on various new ones. Usually these are ever smaller and among the smallest of them we are no longer within the realm of certainty, but in the kingdom of likeliness. While in the past it was possible to trade with a significant level of certainty, today’s markets change so quickly that we can only talk about the likeliness of a trade being successful. And if we are to earn something even on the small inefficiencies, which means to increase the likeliness of making money, we must repeat the trade as often as possible. This leads arbitrageurs to often trade in large volumes or in high frequency.

We can imagine this as a bet with tossing a coin. While the market expects that heads will fall in half of those tosses, we found that in reality (for example due to an irregular shape of the coin) heads will fall more often. If they fell with ninety-percent likeliness, you would have to bet on very few tosses in order to be sure of winning. 

However, if the market inefficiency is small, we have to repeat the tossing considerably more often in order to achieve the likeliness of success. Should heads fall with a likeliness of 50.5 %, we would achieve the likeliness of 99.921% profit only upon one hundred thousand tosses. At the same time, there is the risk of whether we discovered the inefficiency correctly. If in fact heads would fall less often than tails, we could lose a lot of money. For this reason, high-frequency companies should trade with their own funds and not the money of their clients.

Another widespread strategy is market making, where the trader lists on both sides. This means that they offer to both sell and buy a financial asset. This way the trader enables other participants in the market to execute trades. This business model is similar to currency exchange offices. Market makers make their profit on the difference between purchase and sale price, called spread. They are also often rewarded by the stock exchange that values their activity, because it raises its attractiveness. This means that there is always someone within the market who will purchase or sell assets (market makers sometimes specifically commit to this contractually). Additionally, the prices are more advantageous. But this strategy also presents many challenges. In order for market makers to avoid becoming a target for arbitrageurs, they must also use smart algorithms to uncover market inefficiencies and be sufficiently fast.

Because high-frequency trading is not a strategy in itself, it’s not very useful to evaluate the benefits of HFT as a whole. It’s more efficient to focus on its individual strategies. We saw that arbitrageurs could be seen as ‘hunters of inefficiencies.’ By their activity, markets become more efficient – the prices of assets better reflect reality and reflect it more accurately. Unfounded differences between the prices of one asset in different markets disappear and the activity of arbitrageurs also leads to the adjustment of prices of assets that are diverse, yet mutually dependent in terms of price.

Algorithmic trading is directly related to those advancements in electronics at stock exchanges that geographically connect different markets and enable their greater competitiveness. It is therefore very easy to trade gold in both Prague and Vienna.

The strategy of market making brings benefits not just to market makers but also to all participants. Because market makers enable other exchange participants to trade practically any time, it is easier to buy and sell assets. In professional language, we say that the markets become more liquid. This increased liquidity also presents itself by lower spreads, those differences between purchase and sale price. Whoever sells, gets more, whoever buys, pays less. 

We observe the fact that more liquid assets have smaller spreads every day by a glance at the currency exchange-rate list. If the exchange office does not uniformly set the width of the spread (for example, Czech banks often do this), we easily see that frequently traded currencies, such as Euro, US Dollar or British Pound, have much narrower spreads. On the other hand, ‘exotic’ currencies such as the Ukrainian Hryvnia or Croatian Kuna that the exchange office cannot easily trade off or obtain have much wider relative spreads, as expressed in the percentages of the median exchange rate. Table 1 shows the actual exchange rates of a single exchange office in Prague. The right-hand column shows the relative width of the spread in percentages – while the Euro has a relative spread of 2.37%, in case of the Hungarian Forint it’s over 42%.

The dependency of the width of the spread on liquidity is also shown in another exchange rate listing. This exchange office lists exchange rates both for foreign currency exchanges in cash and cashless. Because selling bank notes is much more difficult for an exchange office, it charges much higher prices. The spreads for all the listed currencies are therefore wider for foreign currency/cash operations (see Table 2). In the case of this exchange office, the spreads for more frequently traded currencies are narrower.

It’s easy to see that narrower spreads are more beneficial to the customer. Let’s say we want to spend a weekend in Budapest. If we were to buy Hungarian Forints at the first exchange office, we would get approximately 108,000 Forints for 10,000 Czech Crowns. After returning to Prague we find out, that we only spent 80,000 Forints and have 28,000 left. If we exchange this sum back into Czech Crowns, we will receive 1,680 Czech Crowns.

How much would we have left after weekend in Budapest, if the spread in case of Forints was as narrow as the spread in Euro? If the median exchange rate remained the same (7.625 CZK for 100 HUF), then a spread of 2.37% means an exchange rate of 7.53–7.72 CZK/HUF. For our ten thousand Crowns we would suddenly receive 129,500 Forints (that is twenty thousand more than in the previous situation). After spending 80,000 Forints in Hungary, we would have 49,500 left. By exchanging them to Crowns, we would get back 3,730 CZK. We would be better off by an entire 2,050 CZK (a fifth of the amount that we initially brought to the exchange office). And all this thanks to narrower spreads. The median value of the Forint against the Czech Crown didn’t change. 

Clients of currency exchange offices reap the benefits of narrow spreads and so do the clients of market makers in financial markets, although they wrinkle the brow of the market makers. On one hand, narrow spreads attract clients – would you prefer to go to an exchange office offering Euro at 25.00–25.60 to Czech crown or to one with a range of 24.50–26.00? On the other hand, for market makers, the narrow spreads mean that they must strategically and quickly respond to every change in the market, and actually to every piece of news that might influence that market. The reasons can again be explained on the example of a currency exchange office. This time we will look at it through the eyes of its owner. 

Let’s say, that in the morning we found out that the value of the Ukrainian hryvnia is 2.25 Czech crowns. Because the Ukrainian hryvnia is less liquid, we establish a relatively wide spread of 1.85–2.65, which represents a relative spread of nearly 36% (that is 18% to each side). At noon, we find out that the latest macroeconomic data show that the Ukrainian economy is becoming troublesome. The value of the hryvnia is beginning to drop. But if the drop isn’t significant (specifically if it doesn’t exceed the 18%), we needn’t change our exchange rate at all, because the actual price of the hryvnia will be within our spread. If, for example, it drops to 2 Czech crowns (by 11%), we still make 15 Hellers because our purchase price remains 1.85 Czech crowns. This is considerably less than the original 40 Hellers, but we are still in plus.

But how would a similar drop reflect in the euro? In the morning, we established the value of euro to be 25.30 CZK. Then we stipulated a spread of 25.00–25.60, meaning 1.2% to each side. We can’t afford a wider spread or our exchange rates wouldn’t be competitive. If by noon the euro dropped by 10% to 22.77 CZK and we wouldn’t change our exchange rate, we would already be off. Other exchange offices that had the exact same exchange rate in the morning, but adjusted it at noon, would now have a spread of 22.50–23.04 (while keeping the same spread setting). There would be nothing simpler to buy euro for approximately 23 CZK in a competitor’s exchange office and immediately sell it to us for 25 CZK. 

Perhaps nobody would notice such a thing on the streets of Prague, but todays financial markets would literally wipe away the poor devil with old exchange rates. The markets are fast, merciless and many ferocious traders focus on liquid assets, constantly seeking out incorrect out-of-date prices. Now, let’s bear in mind that in case of highly liquid assets (for example futures exchange rate of EUR/USD) the relative width of the spread, this cushion that protects the market makers from market fluctuations, is only 0.000075%.

The onset of algorithmic and high-frequency trading brought about many changes the consequences of which are not easy to evaluate. New sharks in the water undoubtedly caused stock exchange trades to be smaller on average. 

Several directions of criticism sprung up. The first reproach: HTF critics dislike the frequent cancellation of instructions sent to the exchange. It is not uncommon that the electronic trader sends an instruction that they would like to purchase, for example, a Facebook share for 35.00 USD. Not even a second later, the system cancels the instruction – now it would only purchase one share for 34.99 USD. Critics complain that traders often do this just to make trading more difficult for their competitors with this ‘noise’. It’s similar to a situation in which we might take a drum to the classic stock exchange floor and started mindlessly drumming. If the listing is cancelled with the intent to confuse other players in the markets, then no doubt the sinners should be punished. However, high-frequency traders often behave this way because situations in the market change rapidly. If they failed to change their listings, they would become easy prey for their competitors and begin losing money in their trades. If we were to forbid traders to randomly change listings, it would be similar to forcing currency exchange offices to only change their exchange rates on Mondays. Out of the fear of setting the exchange rate wrong for the whole week, they would adjust them to the detriment of their clients – thickening their cushion in the form of a wider spread.

Similarly to the ‘traders with drums,’ companies that attempt to manipulate the market give a bad reputation to algorithmic traders, causing the second direction of criticism. For example, a technique called mystifying is based on submitting instructions to the exchange that divert the market for a moment. However, these instructions are never processed. They only serve the manipulator to get the best possible prices for an opposite trade. Again, these are different strategies that traders may use, although they shouldn’t. Regulators should certainly focus on eliminating such practices. Moreover, such fishy practices have existed ever since the foundation of stock exchanges – they are certainly nothing new with the onset of algorithmic trading.

A similar situation is in case of front-running, which is also forbidden. In ‘classic’ front-running, an investor called their broker on the trading floor and requested that, for example, the broker purchase some shares. The broker didn’t hesitate and first bought some himself. Subsequently, after executing the investor’s instruction, he would immediately sell his own, because in the meantime the price went up due to increased demand. Some critics point out--and this is the third main reproach--that algorithmic trading enables front-running in a modern cloak. In their view, high-frequency traders are able to predict what others will do at the exchange. But in fact they can only guess as to what will happen in the future. Again, this is nothing new, only a different form for something classic stock-exchange brokers achieved by observing the behavior of their competitors.

The fourth criticism of HFT is usually that humans are not involved and everything happens so quickly that it can’t be effectively controlled. However, the allegation regarding the absence of control is untrue. Everyone who doesn’t want to lose their pants must apply an entire range of mechanisms to prevent the ‘algorithm going crazy’ and causing damage to the market. Because such losses usually go to the debit of the owner of such a ‘crazy’ algorithm (in the summer of 2012, Knight Capital lost 440 million dollars this way), they generally control all instructions going to the exchange. The trades are additionally continually monitored by a staff that can immediately stop a transaction. The critics of algorithmic trading most often point to the role HFT companies played on May 6, 2010, when during a flash crash, American exchanges (and subsequently other markets) began to drop. To blame HFT, however, is deceptive. In fact the drop was caused by an incorrectly executed algorithm by the Kansas-based investment company Waddell & Reed that is by no means a ‘cheetah’ among traders. An American study from 2011 that focused on the flash crash in detail clearly concluded that was not within the capabilities of HFT traders to cause the crash, nor to stop it.* 

Despite the lack of proof concerning the detrimental impact of HFT, there are still opinions suggesting that trading should be artificially slowed, for example by the return of the ‘human era’ when only brokers of ‘flesh and bone’ were active at stock exchanges. Aside from the need to limit internet trading, because it would be hard to determine who clicked the mouse, the risk of errors wouldn’t disappear. In his book The Quants, American journalist Scott Patterson describes a situation within the Morgan bank, which in 1995 traded options to the S&P 500 share index. Back then, the problem was the system not being fully automated. Analysts, who calculated their best strategies on computers, had to call other employees to make trades as quickly as possible. Everything went just fine until one day an analyst misspoke on the phone and the bank sold options worth tens of millions of dollars rather than buying them…

Stephen Perkins, a long-term broker of PVM Oil Futures, made a similarly fatal mistake in July of 2009. He got so drunk that he bought crude oil on the market worth half-a-billion dollars. He was thus responsible at that moment for more than two thirds of the traded volume, corresponding to nearly one tenth of the world’s daily use of this strategic commodity. The price jumped up by 1.5 dollars and he caused his employer damage in the amount of ten million dollars. The British supervisory board then resolved the drunkard’s escapade with a fine totaling 72.000 Pounds and a trading ban for five years.

Of course, even computers don’t operate without slight mistakes and, on occasion, there are problems. In our society, machines gradually take over more and more tasks – for example a number of metros now operate without drivers of the homo sapiens species. There is no reason why financial markets shouldn’t go the same direction. There is no reason to consider the activities of high-frequency algorithmic traders mind-boggling, but there is also no reason to frown upon them. They don’t threaten our demise, they merely attempt to perfect trading strategies known for decades, if not centuries.

* KIRILENKO, Andrei, Mehrdad SAMADI, Albert KYLE and Tugkan TUZUN. The Flash Crash: The Impact of High Frequency Trading on an Electronic Market. 2011 [cit. 2013-01-22]. Available at: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1686004