Istraživanje utjecaja alternativnih podataka na predviđanje volatilnosti cijena valute Bitcoin

master's thesis

• Title (english): Exploring the Impact of Alternative Data on Bitcoin Price Volatility Prediction
• Author: Filip Radović
• Mentor: Klemo Vladimir (mentor)
• Committee member: Klemo Vladimir (predsjednik povjerenstva)
• Committee member: Marin Šilić (član povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-09-16, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu istražuje se i definira novi sustav predobrade podataka u cilju predviđanja cjenovnih serija. Taj je sustav zasnovan na događajima i funkcionira tako da uzorkuje događaje s tržišta i agregira ih prema unaprijed definiranim promjenama na tržištu. Također, u radu se procjenjuje utjecaj različitih značajki s posebnim fokusom na značajke dobivene s Twittera. Testira se više funkcija agregacije i ocjenjuje se njihova izvedba. Osim toga predstavljeno je i objašnjeno nekoliko metoda predobrade za vremenske serije. Rezultati rada pokazuju da sustav zasnovan na događajima ima potencijala i da način agregacije financijskih vremenskih serija može imati utjecaja na uspješnost predikcija modela. Osim toga pokazano je da značajke s Twittera nemaju utjecaj na tržište u predviđanju instrumenta BTC-PERPETUAL u visokofrekventnim uvjetima.

Abstract (english)

This thesis explores and defines a new system of preprocessing data for price series prediction tasks. The system is called an event-based system and functions by sampling events from the market and aggregating data according to predefined changes in the market. The impact of many features is explored with a focus on features that are extracted from Twitter. Multiple aggregation functions are tested, and their performance is evaluated. Finally, several preprocessing methods for preprocessing time series are introduced and explained. This thesis indicates that event-based systems have potential and should be explored in more detail. It is also shown that the type of event aggregation can impact model prediction accuracy. Finally, it is shown that features from Twitter are not good indicators of price change in a high-frequency setting.

• Keywords: Bitcoin; strojno učenje; predviđanje volatilnosti; predobrada vremenskih serija
• Keywords (english): Bitcoin; machine learning; volatility prediction; feature importance; time series preprocessing
• Language: english
• URN:NBN: urn:nbn:hr:168:573849
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 16:00:12