Cryptocurrency’s realized volatility forecasting with GARCH and multivariate LSTM

This study investigates the predictive potential of Long Short – Term Memory (LSTM) models and Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models in the context of cryptocurrency volatility. This factor is essential to investment decisions and risk management in the crypto market. Through extensive experimentation and analysis, the paper presents unique insights into the performance of these models across four different cryptocurrencies. Due to LSTM’s robust adaptability and flexibility, three LSTM models perform best at forecasting volatility. However, the study recognized the potential for overfitting in LSTM models, underscoring the importance of model validation. Interestingly, one simple GARCH model outperformed others for Binance Coin (BNB), attributable possibly to its unique blockchain technology and supply dynamics. The study emphasized the need for model diversity and regular updates to retain model relevance due to the varying influence of diverse factors on cryptocurrency volatility. It is advisable for investors to treat with caution while leveraging these predictive models for investment decision making in the volatile crypto market.