Description

This study investigates the effectiveness of Expected Shortfall (ES) and Expected Upside (EU) estimation models in the context of cryptocurrency markets, with a focus on Bitcoin due to its high volatility. We employ advanced dynamic semiparametric models proposed by Patton et al. (2019) —including the One Factor, Two Factor, and Hybrid models—as well as the Single-Index Expectile Model, and benchmark them against traditional approaches such as rolling window and GARCH models. Model performance for ES and EU forecasting is evaluated using the Model Confidence Set procedure, based on two loss functions: the FZ loss (Fissler and Ziegel, 2016) and the AL log score (Taylor, 2019). The forecasting results are then applied in a portfolio analysis, where we construct cryptocurrency portfolios by minimizing ES and maximizing EU based on the top-performing models, and assess their return and risk-adjusted performance. This research contributes to risk management in high-volatility markets by offering insights into robust ES and EU estimation methods tailored to crypto assets. The paper represents ongoing research and contains only preliminary results.