A study of the leverage effect in Bitcoin Conditional Heteroskedasticity prediction

Abstract

In today's world, countries establish large companies specifically designed to effectively manage the risks associated with volatility exchange rates and leveraged assets. Bitcoin is widely recognized as the leading cryptocurrency, making it highly appealing to investors and traders. The recorded data reveals a growing number of transactions conducted using Bitcoin cryptocurrency. Consequently, accurately predicting the volatility of Bitcoin's conditional heteroskedasticity holds significant importance for the financial markets. Cryptocurrencies present a challenging situation due to their complex dynamics, extreme observations, asymmetry, and various other non-linear features. Modeling these cryptocurrencies accurately becomes difficult using statistical models, as they fail to consider the leverage effect and provide precise predictions. In this paper, we have employed a hybrid modeling approach by utilizing MLP neural network and GARCH statistical model. The experiments conducted on Bitcoin time series data have yielded promising results, showcasing a notable reduction in the prediction error of conditional heteroskedasticity volatility.