Modeling realized measures of volatility

Although volatility is essential for many applications in finance, it is generally an unobservable process with no universal definition. Through association with integrated variance, a multitude of non-parametric volatility estimators collectively referred to as realized measures was proposed. The topic of this cumulative dissertation is modeling time series of realized measures with a special focus on forecasting. Currently, most popular univariate models are basically restricted linear regressions with some economic argumentation, whereas multivariate methods usually face complex unresolved numerical and theoretical issues. Chapters 2 and 3 propose alternative approaches to modeling univariate realized measures from two different perspectives, distributional with copulas and non-parametric with B-splines, respectively. Chapter 4 introduces a novel factor-based approach to modeling multivariate realized measures. All proposed methods are discussed theoretically and compared to respective benchmarks within corresponding extensive empirical studies