Bitcoin price forecasting using hidden : Markov models

Dissertant

Faik, Muhammad

Thesis advisor

Alawi, Abd al-Hamid Hamidi

University

Al Akhawayn University

Faculty

The School of Business Administration

University Country

Morocco

Degree

Master

Degree Date

2016

English Abstract

In this thesis we develop a prediction method based on continuous Hidden markov model with gaussian mixtures.

We apply this method to predict Bitcoin returns and build a systematic trading strategy based on this forecast.

Systematic trading rose exponentially on all asset classes in the last two decades, and prediction models helped hedge funds to build very pro table strategies on FX, Commodities, Equities, Interest rates.

The Bitcoin is a rising digital currency and its trading is gaining an increasing interest among institutionals, individual investors and technology enthusiasts.

Building a model on Bitcoin is an emerging eld, as Bitcoin started only in 2009 and started to be known around 2013.

The Hidden Markov model is a stochastic prediction method that has been used in voice recognition and is gaining interest in nancial forecasting.

We applied HMM model to Bitcoin forecasting in order to take advantage of its ability to nd non-linear paths in historical data.

Empirical results shows that the HMM algorithm can, for a given sets of parameters, give a very pro table strategy.

Using HMM with a sliding window and multivariate model enhanced our strategy.

Changing the other parameters have a limited impact on pro tability.

We tried di erent estimation methods such as Viterbi Algorithm and density calculation method.

We also varied the number of states and mixtures and tried both k-means and random initialization methods.

We used a simple and multivariate model in HMM and changed the triggering threshold of the strategy.

In addition, we switched to Multivariate HMMGM by adding more variables to data.

This change improved the sharp of the strategy.

As a conclusion, we succeeded to build a pro table strategy using HMM to predict Bitcoin, but its pro tability stays very sensitive to parameters.

Main Subjects

Financial and Accounting Sciences

No. of Pages

32

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-775262