Predicting Bitcoin’s volatility : master's thesis

Pristovnik, Jan (author), Todorovski, Ljupčo (mentor), Basrak, Bojan (comentor)

The master's thesis addresses analyzing and modeling the volatility of Bitcoin, the cryptocurrency with the largest marketcap. Volatility is a statistical measure of the dispersion of returns. We approximated it with realized volatility calculated on intra-daily log returns. We defined two baseline models based on a constant value and martingale property and tried to outperform them with both econometric and machine learning models. We used three error functions relative to our baseline models: MAE, MAPE, and RMSE. The best-performing econometric model is the HAR model. The best-performing machine learning model, which also outperforms the HAR model, is the LSTM recurrent neural network.

• Language: English
• Keywords: volatility, forecasting, time series, cryptocurrency, time-series analysis, machine learning, recurrent neural networks, LSTM
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FMF - Faculty of Mathematics and Physics
• Year: 2022
• PID: 20.500.12556/RUL-143592
• UDC: 519.2
• COBISS.SI-ID: 135369475
• Publication date in RUL: 29.12.2022

Secondary language

• Language: Slovenian
• Title: Napovedovanje volatilnosti Bitcoina

Abstract

Magistrsko delo obravnava analizo in modeliranje volatilnosti Bitcoina, kriptovalute z največjo tržno kapitalizacijo. Volatilnost je statistična mera razpršenosti donosov. Aproksimirali smo jo z realizirano historično volatilnostjo, na podlagi visoko frekvenčnih logaritemskih donosov. Definirali smo dva osnovna modela, bazirana na konstantni vrednosti in martingalski lastnosti, ter ju poskušali preseči z ekonometričnimi modeli in modeli strojnega učenja. Uporabili smo tri različne funkcije napak, relativno na naše osnovne modele: MAE, MAPE in RMSE. Najuspešnejši ekonometrični model je model HAR, najuspešnejši model strojnega učenja je rekurenčna nevronska mreža tipa LSTM. Slednja je boljša tudi od modela HAR.

• Keywords: volatilnost, napovedovanje, časovne vrste, kriptovaluta, analiza časovnih vrst, strojno učenje, rekurenčne nevronske mreže, LSTM