Volatility Forecasting Of Bitcoin Prices: Time Series And Machine Learning Approach

Abstract

Digital currencies have attracted considerable attention worldwide over the last few years. Bitcoin has the greatest market capitalization among cryptocurrencies; investors, analysts, and financial experts are interested in predicting its price volatility and obtaining optimal returns. Assessing and forecasting the behavior of cryptocurrencies is a difficult endeavor because of the existence of severe events, information asymmetry, and nonlinear characteristics of time series data. Using multiple heterogeneous auto-regressive models and the deep learning approach LSTM, this study aims to explore the risk and return characteristics of Bitcoin. To forecast volatility,[1] closing price data for Bitcoin from March 2013 to January 2024 were utilized. LSTM and GARCH deep learning models were implemented using Phyton libraries. The performance of the Machine Learning and the statistical models were evaluated with performance matrices like root mean squared error (RMSE), and Mean squared error (MSE) using different plots, and loglikelihood methods. The findings of the study reported the superiority of the volatility forecast of the LSTM approach over traditional econometric models. The findings of the study are useful for investors, financial institutions, fund managers and policymakers to establish volatility strategies to adopt new business models.