A design framework for prediction models based on the order book

Gantar, Klemen (author), Kononenko, Igor (mentor)

This master's thesis deals with the implementation and evaluation of the framework for order book based prediction model design. First, the field of financial markets is described along with the processes, that govern it. Here the focus is on the order book dynamics as it is the main topic of the thesis. Next, the field of machine learning and its relationship with the financial markets are described, followed by the findings of the related work which serve as a motivation for the implementation of the universal framework for order book data collection and analysis. The developed framework and its functionalities are then described. First, the data acquisition and data manipulation modules take care of the data which is then used to design and train prediction models, which are in turn applied to the market along with the trading strategy of choice. Finally, the framework and its functionalities are experimentally demonstrated. Real order book data is collected and used to produce a prediction model which is then evaluated and backtested in connection with a corresponding trading strategy. The thesis is concluded by the overall evaluation of the framework and its results.

• Language: English
• Keywords: machine learning, prediction model, deep learning, neural networks, limit order book, financial markets
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2019
• PID: 20.500.12556/RUL-111964
• COBISS.SI-ID: 1538417347
• Publication date in RUL: 17.10.2019

Secondary language

• Language: Slovenian
• Title: Ogrodje za grajenje napovednih modelov na podlagi knjige naročil

Abstract

Magistrsko delo obravnava razvoj in testiranje ogrodja za grajenje napovednih modelov na podlagi knjige naročil. Najprej je predstavljeno širše področje finančnih trgov in procesi, ki jih upravljajo. Pri tem je poudarek na dinamiki knjige naročil, ki je osnova za formacijo cen in zato osrednja tema magistrskega dela. V nadaljevanju je predstavljeno področje strojnega učenja v povezavi s finančnimi trgi in dognanja iz sorodnih del, ki predstavljajo motivacijo za izdelavo splošnega ogrodja za zbiranje in analizo podatkov knjige naročil. Nato je skozi module predstavljeno ogrodje, ki smo ga razvili v okviru magistrske naloge, in njegove funkcionalnosti. Te sestojijo iz zbiranja, obdelave, analize in uporabe podatkov knjige naročil za namen izdelave napovednega modela, ki je lahko v povezavi s trgovalno strategijo na koncu uporabljen na trgu. Zadnji del magistrskega dela demonstrira uporabnost ogrodja in njegovih funkcionalnosti z izdelavo in evalvacijo napovednega modela in preproste trgovalne strategije. Rezultati evalvacije so nato predstavljeni in ovrednoteni v zaključku dela.

• Keywords: strojno učenje, napovedni model, globoko učenje, nevronske mreže, knjiga naročil, finančni trgi