Abstract/Summary

This thesis investigates cryptocurrency markets. The first empirical chapter studies the Bitcoin volatility forecasting performance between popular traditional econometric models and machine learning techniques. We compare the 1-day to 2- month ahead forecasting performance of the Long Short-Term Memory (LSTM) and a hybrid Convolutional Neural Network-LSTM (CNN-LSTM) model to the traditional models. We find that neural networks outperform Generalised Autoregressive Conditional Heteroscedasticity (GARCH) models for all forecasting horizons. Furthermore, the LSTM model outperforms the Heterogeneous Autoregressive (HAR) model and by integrating the Markov Transition Field (MTF) into the CNN-LSTM model, we achieve superior short-term predicting results, particularly for the 7-day ahead forecasts. In the second empirical study, we document strong and significant evidence of cryptocurrency time series momentum (TSMOM) using volume-weighted market returns. The volume-weighted winner minus loser portfolios outperform other TSMOM strategies, generating 0.94% daily returns with an annualised Sharpe ratio of 2.17. Our findings cannot be fully explained by market, size, and momentum factors, are not subsumed by cross-sectional or MAX momentum returns, and are robust against wash trading concerns. Our results suggest that trading volume may provide signals to create profitable strategies in cryptocurrency markets. The final empirical chapter investigates the relationship between cryptocurrency names and trading volume, focusing on name fluency, structural complexity, and name-ticker similarity. We observe that less fluent and simpler structured names attract higher future trading activities, indicating investor preference for distinct coins. A negative correlation between name-ticker similarity and volume suggests that less obvious connections between names and tickers are favoured. Furthermore, name fluency shows an asymmetric effect, with low to moderate fluency modestly increasing abnormal trading activity, while highly fluent names see significantly larger declines due to investor aversion. Our findings reveal investor behaviour biases linked to linguistic factors, providing insights for cryptocurrency investors, issuers, and regulators.