Elektronsko trgovanje na valutnem trgu s pomočjo Twitterja : magistrsko delo

Brvar, Anže (author), Oblak, Polona (mentor), Zupan, Blaž (comentor)

V magistrskem delu smo raziskovali uspešnost elektronskega trgovanja na valutnem trgu z metodami strojnega učenja.Primerjali smo uspešnost razvitih algoritmov, ki trgujejo s pomočjo objav (tvitov) na Twitterju, in takih, ki za učne podatke uporabijo pretekle vrednosti valutnih tečajev in tehničnih indikatorjev. Za transformacijo besedil v atributni zapis smo poleg znanih metod preizkusili tudi vektorje besed word2vec. Razvite metode transformacije besedil in njihove parametre smo najprej ovrednotili na sorodnem problemu zaznavanja sentimenta tvitov, nato pa jih preizkusili v trgovanju v simulacijskem okolju. Napovedi razvitih metod smo izboljšali z metodami za združevanje napovedi in tako dosegli do 250% vrednost začetnih sredstev pri simulaciji v obdobju zadnjih petih let. V delu poročamo o najprimernejši izbiri parametrov, ki imajo velik vpliv na uspešnost elektronskega trgovanja. Ugotovili smo, da je Twitter bolj primeren vir informacij za uspešno elektronsko trgovanje kot pretekle vrednosti valutnih tečajev.

• Language: Slovenian
• Keywords: valutno trgovanje, forex, twitter, strojno učenje, word2vec, napovedovanje, simulacija
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Publisher: [A. Brvar]
• Year: 2015
• Number of pages: 82 str.
• PID: 20.500.12556/RUL-73440
• COBISS.SI-ID: 1536667331
• Publication date in RUL: 13.11.2015

Secondary language

• Language: English
• Title: Algorithmic trading on Forex market with help of a Twitter

Abstract

In this thesis we study the performance of electronic trading algorithms with a help of machine learning methods. We compare the performance of developed trading algorithms that trade based on posts (tweets) on Twitter with those that trade based on historic foreign exchange values and technical indicators. Besides the well known methods for text transformation to attribute notation we also use word2vec word vectors. We evaluate all the developed text transformation methods and their parameters, first on simpler but related tweet sentiment detection problem and later with trading in simulation environment. We improve developed models' predictions with the prediction combining techniques and we achieve up to 250% of initial funds at simulation in the period of last five years. The results show that Twitter is a better source of trading information than foreign exchange rates and technical indicators.

• Keywords: foreign exchange, forex, twitter, machine learning, word2vec, prediction, simulation