Improved Bitcoin Price Prediction based on COVID-19 data

TL;DR

This study investigates how COVID-19 pandemic data influences Bitcoin price prediction, demonstrating that incorporating pandemic-related features enhances the accuracy of machine learning models.

Contribution

It introduces a novel approach by integrating COVID-19 data into Bitcoin price prediction models, showing improved forecasting performance.

Findings

COVID-19 data improves Bitcoin price prediction accuracy

Feature importance analysis highlights pandemic-related factors

LSTM models perform better with COVID-19 features included

Abstract

Social turbulence can affect people financial decisions, causing changes in spending and saving. During a global turbulence as significant as the COVID-19 pandemic, such changes are inevitable. Here we examine how the effects of COVID-19 on various jurisdictions influenced the global price of Bitcoin. We hypothesize that lock downs and expectations of economic recession erode people trust in fiat (government-issued) currencies, thus elevating cryptocurrencies. Hence, we expect to identify a causal relation between the turbulence caused by the pandemic, demand for Bitcoin, and ultimately its price. To test the hypothesis, we merged datasets of Bitcoin prices and COVID-19 cases and deaths. We also engineered extra features and applied statistical and machine learning (ML) models. We applied a Random Forest model (RF) to identify and rank the feature importance, and ran a Long Short-Term Memory (LSTM) model on Bitcoin prices data set twice: with and without accounting for COVID-19 related features. We find that adding COVID-19 data into the LSTM model improved prediction of Bitcoin prices.