Essays on the Informational Content of Implied Correlation and Observation-Driven Models

This thesis examines the application and comparison of a set of widely adopted parametric and non-parametric approaches in forecasting the realisation of risk measures, such as volatility. In Chapter 2, implied correlation is extracted from options and is utilised to predict realised correlation after decomposition and re-grouping. I also attempt to forecast market return using the re-organized signal. Starting from Chapter 3, I focus on parametric models, more specifically, the observation-driven models represented by GARCH and score-driven models (GAS). The observation-driven models are extensively constructed upon assumptions on innovation terms. Their predictive power concerning realised volatility is evaluated and compared. In addition, the performance of GAS and GARCH models is also compared to implied volatility comprehensively. In Chapter 4, I propose to construct GAS models with shifted Gamma (SG-GAS) and shifted negative Gamma (SNG-GAS) innovations for option pricing. Corresponding GARCH models and the Black-Scholes model are built as benchmarks for comparison. It appears that both GAS models outperform the other models. The superiority of SG-GAS and SNG-GAS is mainly driven by their ability to price out-of-the-money options accurately.