Cryptocurrency price returns volatility modeling and forecasting with GARCH models

Purpose The paper aims to identify suitable conditional variance models for the estimation and forecasting of cryptocurrency returns volatility. Design/methodology/approach The methodology comprises the use of GARCH-family models estimated by maximum likelihood considering different scedastic functions, number of parameters and error distributions. A cross-validation approach is conducted under different market dynamics to provide robust results. Findings Results indicated that the best GARCH methods for digital coins volatility modeling and forecasting are those associated with a small number of parameters, allowing for asymmetric volatility behavior and considering normal/student distributions. Research limitations/implications The findings indicated that volatility behaves differently for each evaluated cryptocurrency, and the selection of the best scedastic function depends on the corresponding digital coin more than the period under evaluation. Practical implications Investors should prefer parsimonious GARCH structures when modeling and forecasting cryptocurrency volatility, and must consider the current state of the market as the methods lose accuracy in high-volatile periods. Social implications The work provides a better understanding of the volatility dynamics of cryptocurrencies, providing evidence of more accurate tools for risk management in this volatile market. Further, better-informed investors on the risks associated with this market are less susceptible to high price variations. Originality/value The research presents an extensive experimental study to identify the optimal GARCH structure for modeling and forecasting return volatility in digital currencies, considering various market conditions and digital coins, which yields more robust results.