Bitcoin’s Realized Volatility Forecasting: A Comparison between GARCH and LSTM
Download as PDF
He Ma, Xinyi Lin
This study compares GARCH and LSTM models, assessing their pros and cons in economic and machine learning. It aims to enhance understanding of their relationship and applicability in various contexts. This article reviews the existing literature on Bitcoin and forecasting models, explains the dataset and methodology used for computing volatility, analyzes GARCH, GJR-GARCH, LSTM, and Bi-LSTM models, introduces three new input dimensions for the Bi-LSTM model, presents predicted outcomes, evaluates them using RMSPE and RMSE, and draws a robust conclusion with insightful findings and future research directions. The investigation reveals that the LSTM model has more outliers but effectively captures Bitcoin’s volatility trends, indicating its potential for forecasting complex patterns in the market.
Bitcoin’s volatility, GARCH, GJR-GARCH, Bi-LSTM, Multivariate Bi-LSTM