| Περίληψη : | This thesis consists of essays on modelling and forecasting asset price volatility. The common motivation behind these essays is twofold: firstly, to exploit the rich informational content of volatility by integrating observable estimators of variance dynamics into parametric GARCH-type models and, secondly, to investigate the ability of these augmented GARCH representations to explain and forecast the observable variance dynamics at multiple horizons ahead. We start by considering a GARCH specification, which we augment to incorporate information from high-frequency data related to measurable characteristics of Realized Variance (RV). The choice of these exogenous features is well-motivated by the recent and ongoing stream of empirical studies on Heterogeneous Auto-Regressive (HAR) models. We find our ``augmented'' Realized-GARCH and Realized-EGARCH specifications to perform significantly better than other already-existing models in fitting the data (in-sample) and forecasting Realized Variance (out-of-sample). The enhanced performance of our models is primarily due to the inclusion of: (i) realized upside/downside semi-variances (indicating prevalent asymmetric effects in intra-day variance), (ii) heterogeneous terms of RV (exogenously approximating long-memory patterns in volatility), and (iii) realized jump or variance-of-variance indicators (capturing discontinuities in the RV process or attenuation biases in RV projections, respectively). Next, we introduce a risk-neutral variance proxy within a flexible parametric framework that is described by affine-GARCH dynamics and a variance-dependent pricing kernel.We find evidence of a sizeable priced volatility risk premium (of approximately -3%), that can be recovered in a robust and parsimonious way from the VIX dynamics through a joint estimation approach. We analyze the transmission mechanism of innovations from physical to risk-neutral dynamics, as well as the impact of volatility risk on the news impact curves and impulse response functions of risk-neutral variance. Our approach reveals that accounting for volatility risk in this GARCH-based framework is of utmost importance for establishing a consistent link between the physical and risk-neutral probability measures. Finally, we provide an extension to the EGARCH and Realized-EGARCH that allows capturing long-run and short-run dynamics of log-variance. We find that decomposing variance into long/short-run dynamics, significantly improves the ability of the model to jointly explain the observable dynamics of returns and RV. Our preliminary results indicate the presence of a long-run component that is highly persistent and not very responsive to past shocks, as well as a short-run component that is less persistent and transmits most of the impact of past shocks on variance. Interestingly, we find shocks to RV to have an equally strong impact on both long- and short-run components (more pronounced for the short-run component), which implies that shocks to returns impact mainly the short-run volatility, whereas shocks to volatility itself may have a more long-run effect. As we discuss, the model extensions that we present in this thesis have direct and important economic implications for asset-pricing, volatility forecasting and risk-management.
|
|---|
Copyright © 2013 Βιβλιοθήκη Ο.Π.Α. Με την επιφύλαξη παντός δικαιώματος.