Bitcoin price prediction
Keynote presentation: HERE
Objective
The project's objective was to find a multiple regression model that could predict the price of the cryptocurrency bitcoin.
Dataset
The dataset consists of daily (without weekends & holidays) prices in U.S. dollars of bitcoin, stock prices of the leading global manufacturers of GPUs (NVIDIA & AMD), Big Tech companies (Alphabet, Amazon, Meta, Apple, Microsoft & Netflix), payment-processing corporations (Visa, MasterCard, Discovery & American Express), and Tesla stock prices.
Source & scope
The dataset was prepared with the use of Google Finance. It starts from 11/19/2015 till 12/30/2022.
Operations on the dataset
Skewed data were normalized by use of log transformation. The winsorization method was conducted in order to deal with possible outliers.
Correlation plot considering all features.
Models
The first model consists of the price of bitcoin against all other stock prices. The second one: bitcoin against four groups. Each of the groups covers the average stock price of the group participant. So, for example, the TECH group has the mean price of Big Tech companies: Alphabet, Apple, Meta, Amazon, Netflix & Microsoft.
Correlation plot regarding this model.
Method
Multiple linear regression has been used in this project. Choosing model was conducted using a stepwise algorithm implemented in R language.
Results
The best model is the one that was chosen by the stepwise algorithm that consists of:
btc ~ nvda + amzn + v + axp
in which nvda has the lowest p-value.
Final plot: residuals vs. fitted
Summing up, neither of the models provides satisfying results to be used in practice -- no significant p-value and meager R-squared value are one of the reasons for this statement.
Files
- bitcoin-price-prediction.R
- dataset.xlsx
bitcoin-price-prediction.R
This is the project's main file, and the whole R code is covered here. In order to be usable, the following packages should be loaded:
library(DescTools)
library(ggplot2)
library(dplyr)
library(Hmisc)
library(pastecs)
library(corrplot)
library(ppcor)
library(readxl)
library(forecast)
dataset.xlsx
This is the dataset of the project.
This work is part of my exam in Financial Data Science @ the University of Pavia, Italy.