High-frequency trading: what are the risks and the returns for financial markets?

Hi! PARIS is the new Cen­ter on Data Ana­lyt­ics and Arti­fi­cial Intel­li­gence for Sci­ence, Busi­ness and Soci­ety cre­at­ed by Insti­tut Poly­tech­nique de Paris (IP Paris) and HEC Paris and recent­ly joined by Inria (Cen­tre Inria de Saclay).

The term refers to the plac­ing of buy and sell orders on the finan­cial mar­kets over an extreme­ly short time frame, on the order of nanosec­onds, gen­er­al­ly in reac­tion to events. For exam­ple, when the Fed decides to cut inter­est rates, high-fre­quen­cy traders will place orders a nanosec­ond after the macro­eco­nom­ic announce­ment is made. 

Obvi­ous­ly, humans can’t react on this time scale. High-fre­quen­cy trad­ing is char­ac­terised by two phe­nom­e­na. First­ly, it is auto­mat­ed trad­ing: it requires the use of algo­rithms and com­put­ers to place orders. Sec­ond­ly, it must react extreme­ly quick­ly to mar­ket events, whether macro­eco­nom­ic announce­ments or changes with­in the mar­ket such as price movements.

If we look at the Paris Stock Exchange, for many years we had this image of traders shout­ing their orders around the trad­ing floor. This has not been the case since 1986, when the trad­ing floor was replaced by a com­put­er RAM. Since then, traders have been car­ry­ing out their trans­ac­tions all over the world via online ter­mi­nals. This move towards automa­tion began in the late 1960s and end­ed for equi­ty mar­kets in the ear­ly 1990s. This automa­tion led to so-called algo­rith­mic trad­ing and, over the last twen­ty years, to high-fre­quen­cy trading.

On the equi­ty mar­kets, it’s esti­mat­ed that around two-thirds of trans­ac­tions are car­ried out by high-fre­quen­cy traders. This may seem high, but it’s not all that sur­pris­ing. It’s a bit like super­mar­kets: while they account for a large pro­por­tion of con­sumer trans­ac­tions in France, they are mere­ly inter­me­di­aries. High-fre­quen­cy traders are not the final hold­ers of secu­ri­ties, but the prin­ci­pals. They gen­er­al­ly only hold the secu­ri­ties for a very short time.

It’s true that it has evolved con­sid­er­ably. The traders we often see in the movies, behind their com­put­ers or on the phone giv­ing their orders, have gen­er­al­ly been replaced by com­put­ers and algo­rithms. Today, most trad­ing algo­rithms are devel­oped by com­put­er sci­en­tists. They are recruit­ed by spe­cialised com­pa­nies, par­tic­u­lar­ly in high-fre­quen­cy trad­ing. These pro­files can pro­vide a def­i­nite com­pet­i­tive advan­tage, as they have exper­tise in writ­ing trad­ing-spe­cif­ic codes and have devel­oped exper­tise in opti­mis­ing mar­ket access speed.

While high-fre­quen­cy trad­ing requires an algo­rithm with the right strate­gies, it is true that the mar­kets are well aware of them. It’s just that the time scale for putting them into prac­tice is much short­er. Opti­mis­ing this speed of access to the mar­ket is there­fore cru­cial, because reac­tion time is essen­tial: the oper­a­tor with the short­est reac­tion time has an unde­ni­able com­pet­i­tive advan­tage. These tech­ni­cal skills are very spe­cif­ic and gen­er­al­ly con­cern peo­ple who have trained in com­put­er sci­ence, for exam­ple. If there is one area of the econ­o­my where we can talk about Big Data, it is the finan­cial mar­kets. Every day, they gen­er­ate an enor­mous amount of data in dig­i­tal form. Every trans­ac­tion, every order sub­mis­sion is a new piece of data. For high-fre­quen­cy trad­ing, it is there­fore nec­es­sary to sub­scribe to data feeds that are sold by the exchanges. This requires high­ly advanced tech­ni­cal exper­tise and tech­no­log­i­cal invest­ments that are cost­ly both in terms of human cap­i­tal and computers.

Investors like you and me don’t have enough cap­i­tal to access the oppor­tu­ni­ties these investors do when mak­ing these invest­ments. The prin­ci­ple of high-fre­quen­cy invest­ing is to obtain mod­est prof­its on each trans­ac­tion. These are very small prof­its, but because the trans­ac­tions are repeat­ed so many times, the prof­its are very large very quick­ly. If the invest­ment is large, risk-tak­ing is min­i­mal, because the strate­gies are well known and reac­tion time is opti­mised to make the most of oppor­tu­ni­ties. The unpre­dictable part is rel­a­tive­ly small. As a result, high-fre­quen­cy trad­ing firms have expand­ed rapid­ly over the last ten years. The prof­its that can be made are now much lower.

There are sev­er­al famous cas­es of insta­bil­i­ty caused both by algo­rithms and to the detri­ment of algo­rithms, includ­ing the 2010 flash crash in the Unit­ed States. The stock mar­ket crash last­ed just over 30 min­utes and was not direct­ly due to high-fre­quen­cy trad­ing, but it did cause these oper­a­tors to stop their algo­rithms, which prob­a­bly ampli­fied the crash. And then, as always with algo­rithms, there can be design flaws, bugs that were not antic­i­pat­ed. This was the case in 2012 for Knight Cap­i­tal, a major US trad­ing firm and one of the four major high-fre­quen­cy trad­ing oper­a­tors. One day, one of their algo­rithms devel­oped a bug and Knight Cap­i­tal quick­ly record­ed a loss of 440 mil­lion dol­lars and almost went bankrupt.

One area we still don’t talk about very much is the risk of intru­sion by a plat­form or algo­rith­mic oper­a­tor. Ill-inten­tioned hack­ing, for exam­ple by ter­ror­ists, could seri­ous­ly desta­bilise the market.

We can expect the trad­ing indus­try to use these tools one day. Some ini­tial thought has already been giv­en to this issue in a field oth­er than the finan­cial indus­try, name­ly online sales. Reg­u­la­tors, notably in the UK and the US, have shown that there is col­lu­sion between algo­rithms. Take Ama­zon’s pric­ing algo­rithms, for exam­ple: they are found­ed on arti­fi­cial intel­li­gence based on learn­ing first, a self-learn­ing process for repric­ing prod­ucts. But they don’t just raise or low­er prices depend­ing on the con­text, they find ways to charge non-com­pet­i­tive prices. If they are not cod­ed for this, they dis­cov­er ways of mak­ing prof­its by cre­at­ing implic­it collusion. 

Opti­mis­ing the speed of access to the mar­ket is cru­cial, because reac­tion time is everything.

This has not yet become an issue for the finan­cial mar­kets, but it soon will, because it is part of the same ques­tion. In 2017, Black­Rock, one of the largest Amer­i­can asset man­agers, began using arti­fi­cial intel­li­gence tech­niques to fore­cast returns and make port­fo­lio allo­ca­tions. The firm also relies on big data as part of its quan­ti­ta­tive and tra­di­tion­al invest­ment strate­gies. As for the invest­ment bank JP Mor­gan, it has already announced the devel­op­ment of a fore­cast­ing tool based on arti­fi­cial intel­li­gence to antic­i­pate the deci­sions of cen­tral banks.

Interview by Jean Zeid