This dissertation consists of four essays that focus on the measurement and economic analysis of key risk factors behind macroeconomic and financial variables using state-space models. Chapters 2, 3, 4, and 5 develop and implement estimation approaches that can handle nonlinear linkages of economic forces and tackle issues when data are missing or contaminated by errors. Chapter 2 estimates an equilibrium term structure model that includes real and nominal uncertainty in particular that allows for changes in the responsiveness of the Federal Reserve to inflation fluctuations. These uncertainty, particularly those concerning monetary policy action are considered potential sources of risk variations that can explain several features in the U.S. government bond market including the upward sloping yield curve. Chapter 3, co-authored with Frank Schorfheide and Amir Yaron, develops a nonlinear state-space model to estimate predictable mean and volatility components in monthly consumption growth using a mixed-frequency data and accounting for serially-correlated measurement errors. We provide a methodological contribution that allow to maximize the span of the estimation sample to recover the predictable component and at the same time use high-frequency data to efficiently identify the volatility processes. The estimation provides strong evidence for predictable mean and volatility components in consumption growth. We show that the model can go a long way in explaining several well known asset pricing facts of the data. Chapter 4, co-authored with Boragan Aruoba, Francis Diebold, Jeremy Nalewaik, and Frank Schorfheide, considers the fundamental question of GDP estimation, focusing on the U.S., and provides estimates superior to the ubiquitous expenditure-side series by applying optimal signal-extraction techniques to the noisy expenditure-side and income-side GDP estimates. The quarter-by-quarter values of the new measure often differ noticeably from those of the traditional measures, and dynamic properties differ as well, indicating that the persistence of aggregate output dynamics is stronger than previously thought. Chapter 5, co-authored with Frank Schorfheide, develops the idea of using mixed-frequency data in state-space form. We show that adding monthly observations to a quarterly VAR, which then is estimated with Bayesian methods under a Minnesota-style prior, substantially improves its forecasting performance.