Song, XiaojingBugis, Tala2023-12-272023-12-272023-09-22https://hdl.handle.net/20.500.14154/70457Financial researchers and traders seeking trustworthy forecasting tools have a formidable barrier in cryptocurrency markets, which are volatile and decentralised. This study explores cryptocurrency price volatility prediction and provides insights into digital asset-specific models and methods. The issue is that bitcoin markets defy predictability. These complex patterns are decoded using GARCH, recurrent neural networks, and hybrid models. The significant findings emphasise customised solutions. Bidirectional Long Short-Term Memory (BI-LSTM) models with 1D convolutional layers outperformed standard models in predicting Binance Coin (BNB) and Ripple (XRP) volatility due to their ability to capture complicated temporal connections. With asymmetric reactions, Ethereum's (ETH) volatility required unique approaches like the GRJ-GARCH model. This study concludes that the cryptocurrency ecosystem is complex and requires specialised solutions for each digital asset. Our findings further support the Efficient Market Hypothesis (EMH), which emphasises market efficiency in forecasting models. Future research and applications must incorporate robustness testing, regulatory compliance, and external factor integration as cryptocurrency marketplaces mature. Further research into hybrid models that combine GARCH and LSTM strengths is promising. This analysis helps us predict bitcoin volatility and shows how dynamic cryptocurrency markets are. Recognising their distinct traits and adapting forecasting models allows us to leverage the predictive potential needed in this continuously changing market.130enFinanceriskquantitative financerisk managementcrytpocurrencyvolatilityarchgarcharch/garchlstmdeep learning modelsThesis