Modeling and Forecasting of Multivariate Stock Market Volatility

This thesis contributes to recent developments in multivariate volatility modeling. The analysis focuses on stochastic volatility models and the direct modeling of realized (co)variances as precise measures of latent variances and covariances. Novel time-series models are proposed and analyzed in order to capture the complex serial and cross-sectional dynamics of daily and intra-daily asset return (co)variances and investigate the short-term information transmission on international financial markets as reflected by variance and covariance interdependencies. The proposed volatility models address both low-dimensional as well as high-dimensional volatility modeling. The models' in-sample properties are analyzed using model diagnostic tests while the out-of-sample forecasting performance is evaluated using comprehensive out-of-sample experiments including a range of prominent forecasting models from the relevant literature.