Bitcoin - Multiple Linear Regression

Table of Contents

• Introduction
• Motivation
• Dependent Variable
• Independent Variables
• Data Exploration
• Data Preparation
• Predictive Models
• Results

Introduction

Supervised machine learning to create a model that predicts the Bitcoin price.

Please download the PDF in this repo for more detail. Below only gives a brief overview.

Motivation

This was a paper I wrote during my Masters degree which achieved 90% the highest on the course. I was highly motivated in this project due to my interest in statistics, maths, and using machine learning to make predictions.

Data Sources

• Coindesk
• Thomson Reuters
• Tor Metrics
• Google Trends
• Federal Reserve

Dependent Variable

• Bitcoin/USD Spot Rate

Independent Variable (Features)

• Coindesk
  • Daily Bitcoin transactions
  • Difficulty mining Bitcoin
  • Total number of Bitcoins in circulation
• Thomson Reuters
  • Gold spot price
  • Oil spot price
  • Euro/USD spot price
  • GBP/USD spot price
• Federal Reserve
  • Effective federal funds rate
• Google
  • Google searches for Bitcoin
• Tor Metrics
  • Total number of Tor clients

Data Exploration

Correlation Between Variables

Summary Statistics

Standard Deviations and IQR

Time Series - Variable Percentage Change

Time Series - Bitcoin against Google Searches and Tor Clients

Data Preparation

Data cleaning

• Took first difference to ensure variables are on the same scale
• Filled in missing values for Google Trends by duplicating weekly values to whole week
• Date range adjusted to ensure all variables are within scope

Histogram - Showcasing effects of first difference transformation

Predictive Models

Results