Will Ma­chines Make Bet­ter Traders?

Ma­chine learn­ing’s abil­ity to ‘learn’ with date, make pre­dic­tions, and con­tin­u­ously im­prove on the ac­com­plish­ment of its tasks may have earned its per­ma­nent spot on the trad­ing floors

Portfolio - - IN THIS ISSUE - BY LI HAO­HAN

Plat­inum An­a­lyt­ics is lead­ing the way in find­ing ap­pli­ca­tions for ma­chine learn­ing in forex trad­ing. The next-gen­er­a­tion fin­tech com­pany cus­tom­izes lo­cal­ized tech solutions for fi­nan­cial in­sti­tu­tions, and counts among its clients Ten­cent, Xin­hua, and DBS Bank. An of­fi­cial state­ment from the com­pany says it is “seek­ing to rev­o­lu­tion­ize the struc­ture and com­pet­i­tive dy­namic of the forex trad­ing sec­tor with its cut­ting-edge AI tech­nolo­gies, while boost­ing the ca­pa­bil­i­ties of traders in Sin­ga­pore to be among the top in the in­dus­try.” Plat­inum’s AI trad­ing tool uses sen­ti­ment and mar­ket data anal­y­sis to gen­er­ate trad­ing sig­nals, which re­sults in valu­able in­sight for the trad­ing com­mu­nity to bet­ter tap into mar­ket data, man­age risk and make money.

The al­go­rithm is self-learn­ing and be­comes more in­tel­li­gent over time. Tra­di­tion­ally used in the eq­ui­ties sec­tor, Plat­inum is the first to pi­o­neer the de­vel­op­ment of this ground­break­ing tool in forex trad­ing.

We met up with Mr. Qi Hong Bao, Chief Tech­nol­ogy Of­fi­cer at Plat­inum An­a­lyt­ics, to dis­cuss the com­pany’s solutions and ser­vices.

AI can­not re­place ex­pe­ri­ence and in­stinct. An ex­pe­ri­enced and savvy trader, as it stands to­day, can al­most al­ways do bet­ter than an AI.

HOW DOES MA­CHINE LEARN­ING WORK IN A FOREX TRAD­ING EN­VI­RON­MENT?

The tech­nol­ogy is split into three parts - the data ag­gre­ga­tor, the nat­u­ral lan­guage pro­ces­sor, and the AI core. The data ag­gre­ga­tor is re­spon­si­ble for col­lect­ing news, mar­ket price, so­cial me­dia, and trade data.

The fol­low­ing are three ex­am­ples of how the tech­nol­ogy works:

A. Pars­ing and clas­si­fy­ing news: HOW IT WORKS:

The data ag­gre­ga­tor col­lects ma­chine read­able news from Bloomberg/Thom­son Reuters. The Nat­u­ral Lan­guage Pro­ces­sor parses the news item and iden­ti­fies key sen­tence struc­ture and topics. The AI core uses statis­tics based ma­chine learn­ing al­go­rithms to clas­sify the news into sev­eral in­dices.

WHAT IT DE­LIV­ERS:

This al­lows the trader to see a com­plete land­scape of all news items af­fect­ing a sin­gle cur­rency or a cur­rency pair.

HOW IT IS SETUP:

The client sub­scribes to news items he is in­ter­ested in and sets up his dash­board with cur­ren­cies he wants to watch.

B. Cal­cu­lat­ing Mar­ket Im­pact HOW IT WORKS:

The data ag­gre­ga­tor col­lects ma­chine read­able news from Bloomberg/Thom­son Reuters and runs it through the clas­si­fier. The Nat­u­ral Lan­guage Pro­ces­sor parses the news item and iden­ti­fies key sen­tence struc­ture and topics, us­ing the clas­si­fied cur­rency as the ref­er­ence. The AI core uses al­go­rithms to link en­ti­ties and topics with a po­lar­ity dic­tio­nary to de­ter­mine sen­ti­ment. Via mar­ket data cor­re­la­tion anal­y­sis, the al­go­rithm then de­ter­mines whether this im­pact is bullish or bear­ish.

WHAT IT DE­LIV­ERS:

This al­lows the trader to gauge the im­pact of a piece of news item on the mar­ket.

HOW IT IS SETUP:

The client sub­scribes to news items he is in­ter­ested in and sets up his dash­board with cur­ren­cies he wants to watch. As news comes in, the plat­form will pro­vide im­pact anal­y­sis on each news item.

C. Gen­er­at­ing Trad­ing Sig­nals HOW IT WORKS:

The data ag­gre­ga­tor col­lects ma­chine read­able news from Bloomberg/Thom­son Reuters and runs it through the clas­si­fier and im­pact anal­y­sis meth­ods. The Nat­u­ral Lan­guage Pro­ces­sor parses the data and pro­vides a dis­sec­tion of the data item, in the case of un­struc­tured data. The plat­form then looks at the re­sults of the data clas­si­fi­ca­tion and mar­ket im­pact and alerts the trader of trade op­por­tu­ni­ties. Once an op­por­tu­nity has been iden­ti­fied, the trader has the choice of man­u­ally ex­e­cut­ing the or­der or flow­ing it down to the ex­e­cu­tion en­gine.

WHAT IT DE­LIV­ERS:

This al­lows the plat­form to ex­e­cute real time anal­y­sis on data to iden­tify trad­ing sig­nals in or­der to max­i­mize trader profit and min­i­mize loss.

HOW IT IS SETUP:

The trader chooses his sig­nal gen­er­a­tion pa­ram­e­ters and in­structs the al­go­rithm to pro­vide a range of en­trance and exit sig­nals.

Has the tech­nol­ogy been ap­plied to ex­ist­ing busi­nesses?

Yes, ma­jor banks are us­ing our tech­nol­ogy to au­to­mate trad­ing and an­a­lyze the forex mar­ket.

De­scribe the set up and its in­tro­duc­tion to the client’s ex­ist­ing sys­tem. How long is the pe­riod of sys­tem in­te­gra­tion? What will be re­quired from clients in­ter­ested in avail­ing of your prod­ucts and ser­vices?

We pro­vide the ser­vice as SaaS (soft­ware as a ser­vice). We host it on a cloud-based ser­vice plat­form where clients can log in through se­cure VPNs via an https por­tal to use the ser­vice. The in­te­gra­tion pe­riod varies, but is gen­er­ally mea­sured in weeks. Clients will have a pe­riod of re­quire­ment gath­er­ing with our prod­uct team to de­ter­mine range of data, cur­rency range, and al­go­rithm de­sign.

Who will gen­er­ate the data for anal­y­sis, and who will be in con­trol of said data? What se­cu­rity as­sur­ances are there for the data that are be­ing an­a­lyzed?

We take our data via three sources: ma­chine read­able API from syn­di­cated news sources such as Bloomberg and TR, crawled data from web­sites, and mar­ket data from ex­changes. Data en­cryp­tion, safety, and trans­mis­sion is han­dled by the plat­form. All data, both at rest and in trans­mis­sion, are en­crypted, and all traf­fic go­ing in and out are all through https and run ei­ther via se­cured VPN or leased line.

How will you mon­e­tize the busi­ness? What rev­enue streams are you look­ing at for Plat­inum?

We charge fees for both cus­tomiza­tion and also via SAAS fees.

What are your plans for the com­pany in the near and medium terms?

In the near term, we look to en­hance and im­prove our al­go­rithm ac­cu­racy and in­crease our AI an­a­lyt­ics ca­pa­bil­i­ties. In the medium term, we hope to ex­tend our tech­nol­ogy to other fi­nan­cial prod­ucts such as fixed in­come, eq­ui­ties, and fu­ture.

What role will traders as­sume as AI helps them make de­ci­sions?

Traders will still be reson­si­ble for for­mu­lat­ing the strate­gies, as well as pro­vid­ing high level de­ci­sion mak­ing in ex­e­cut­ing those strate­gies. The pur­pose of our AI plat­form is to free up the traders’ time and at­ten­tion so that they may spend more time and ef­fort shap­ing their strate­gies rather than rote ex­e­cu­tion.

In what as­pects is AI lim­ited in mak­ing de­ci­sions?

AI can­not re­place ex­pe­ri­ence and in­stinct. An ex­pe­ri­enced and savvy trader, as it stands to­day, can al­most al­ways do bet­ter than an AI. AI is good at tak­ing in a wide ar­ray of in­for­ma­tion quickly and at­tempt to piece to­gether pat­tens and rec­og­nize trends, but it lacks the fo­cused de­ci­sion mak­ing of a se­nior trader. The AI is good at solv­ing some prob­lems par­tic­u­larly well, and in this case, the prob­lem is in an­a­lyz­ing data that is both wide and fast, but it strug­gles when it comes to depth. The hu­man brain, on the other hand, is ca­pa­ble of mak­ing de­ci­sions based in in­for­ma­tion depths. In this man­ner, the AI and the hu­man brain com­ple­ment each other.

Who will be cul­pa­ble, or held li­able, when AI fails or makes a bad de­ci­sion?

So, first, I think cul­pa­bil­ity will pri­mar­ily need to be de­ter­mined by a case by case ba­sis. How­ever, just like hu­mans need to do their due dili­gence in proac­tively man­age their risk, so do the engi­neers and anal­y­sis re­spon­si­ble for cre­at­ing the AI. It’s up to the AI engi­neers to make sure the pro­gram is ro­bust and guarded against po­ten­tial risks and fil­ter out el­e­ments which may cause poor de­ci­sions mak­ing. The same ex­pec­ta­tions from traders should also be given to the AI. That said, in most cur­rent im­ple­men­ta­tions of AI, the trader still has the fi­nal say in the de­ci­sion makng process. How­ever, it’s equally im­por­tant for the AI to not mis­lead the hu­man trader by giv­ing poorly quan­ti­fied de­ci­sions with no con­fi­dence qual­i­fier. If a hu­man trader makes bad de­ci­sions which cause fi­nan­cial loss be­cause, say, they came to work drunk, then the cul­pa­bil­ity lies with the hu­man. If the AI al­go­rithm is im­ple­mented with un­fin­ished quan­l­ity as­sur­ance or in­suf­fi­cient back­test­ing, then for the same rea­son, the cul­pa­bil­ity is with the AI, even if the hu­man trader had the fi­nal say in mak­ing the de­ci­sion. Of­ten, in these cases, rul­ing out ob­vi­ous mal­ice or sab­o­tage, cul­pa­bil­ity will gen­er­ally lie with those who failed to do due dili­gence or ig­nored proper risk man­age­ment.

Newspapers in English

Newspapers from Singapore

© PressReader. All rights reserved.