Three essays in econometrics: Estimation with persistent regressors, MCMC inference about factors, and the FAVAR

The dissertation concerns three aspects of time series analysis. The first Chapter considers estimation of a slope coefficient on highly persistent and predetermined regressors. It provides an optimal estimator in the class of procedures that are median unbiased irrespectively of the degree of persistence, the result which holds for a wide class of monotone loss functions. The estimator dominates previously available alternatives in terms of expected square losses across the domain of near nonstationarity. Empirical application documents encouraging performance of the proposed estimator for forecasting asset returns.; The second Chapter investigates Markov Chain Monte Carlo (MCMC) inference about factors in large dynamic factor models. It demonstrates how the MCMC methodology can be applied to approximate the joint likelihood of factors and parameters in large models, of dimension preventing reliable maximum likelihood estimation. The Chapter further discusses the flexibility of the method in incorporating identifying assumptions as well as potential costs of the additional parametric structure it imposes. It concludes with the empirical exercise aimed to asses the informational content of the MCMC factors.; Finally, the last Chapter, which is coauthored with Ben S. Bernanke and Jean Boivin, examines the issue of sparse information sets typically encountered in vector autoregressive (VAR) analyses of the effects of monetary policy. It demonstrates how a structural model can be set within a factor framework, resulting in a factor-augmented vector autoregressive (FAVAR) model, and estimated by either the standard principal components approach, or the parametric procedure described in Chapter 2. An empirical implementation of the method suggests that the information captured by the FAVAR helps to properly identify the monetary transmission mechanism.