Crypto-finance: machine learning forecasts, causal network dynamics, and informed option trading

Abstract

Crypto-finance, an emerging field at the intersection of financial technology and blockchain applications, is transforming traditional financial markets by introducing cryptocurrencies. This thesis presents a comprehensive exploration of crypto-finance, leveraging machine learning (ML) techniques to forecast cryptocurrency volatility, assess the structure and dynamics of the cryptocurrency market through causal network analysis, and examine the predictive power of cryptocurrency options on market movements. Consequently, this thesis makes three significant contributions to the crypto-finance landscape.

The first contribution involves the use of ML techniques, including Random Forest and Long Short-Term Memory (LSTM) models, to forecast cryptocurrency volatility, incorporating both market-specific (internal) and broader technology and financial and policy uncertainty (external) determinants. It achieves superior forecasting accuracy through Genetic Algorithms and Artificial Bee Colony optimisation for LSTM hyper-parameter tuning. This study also compares multi-step forecasts and evaluates the performance of a universal model versus cryptocurrencyspecific models. Furthermore, it employs SHapley Additive exPlanations to provide a deeper understanding of the cryptocurrency market and its interrelations with other financial markets.

The second contribution addresses the gap in understanding spillover effects and information diffusion mechanisms within cryptocurrency and stablecoin markets.

By leveraging network theory, this study provides complementary information on the dynamic nature of these markets. It highlights stablecoins’ critical role in transmitting and absorbing market shifts. These findings emphasise the significant transmission impact of the mid-size crypto-assets, underscoring their impact not relative to their market size. It provides insights into systemic risk indicators, which benefit investors and policymakers.

The third contribution examines the predictive power of informed crypto options trading, focusing on Bitcoin. Analysing the Bitcoin options trading data from Deribit—a leading platform in the cryptocurrency derivatives market—the findings reveal that ML models, particularly Random Forest, provide superior performance over the traditional linear method. Moreover, we identify nonlinearities and interactions between factors in Bitcoin options trading and Bitcoin daily returns. This finding underscores the importance of ML and interpretable ML methods in leveraging informed trading signals in the volatile cryptocurrency market.

Overall, this thesis adopts ML techniques, interpretable ML methods, econometric methods, non-parametric statistics methods and network theory to enhance the understanding of crypto-finance. It advances academic discourse and offers practical strategies for managing cryptocurrency investments and regulatory approaches. These contributions promote the crypto market stability and investor confidence through advanced analytics.