This thesis exploits the information contained in high-frequency data to test and model the distributions of returns of financial assets and their volatility. In Chapter 1 we study the asymptotics of some common tests for normality when applied to returns standardized by noise measures of volatility based on the use of high-frequency data. Chapter 2 proposes dynamic models for conditional quantiles of daily returns and realized volatility exploiting the information contained in various components of historical volatility as well as option-implied volatility. Chapter 3 provides a comprehensive simulation-based comparison of alternative tests for jumps in asset prices in order to get a better understanding of the performance of the tests under different, empirically relevant, scenarios. Chapter 4 extends the testing procedures studies in Chapter 1 to the multivariate context and provides new empirical evidence about the validity of the mixture of normals hypothesis in foreign exchange markets. Chapter 5 studies the dynamics of the tail risk in the hedge fund industry. Finally, Chapter 6 introduces a new method for estimating large covariance matrices.