Essays on iterative and two-step estimators with applications to financial econometrics.

摘要

This dissertation consists of essays on iterative and two-step estimators, with particular emphasis on the application of these estimators within financial econometrics. In the first essay, I develop a new iterative estimator for bundled parameter models, which contain both finite-dimensional parameters, often called parameters of interest, and infinite-dimensional parameters, often called nuisance parameters, particularly in a likelihood context. Applications to semiparametric GARCH-in-mean models and a semiparametric extension of Mertons' credit risk model highlight the usefulness of this new procedure. In the second essay, I propose a new semiparametric multivariate GARCH-in-mean model to analyze risk return dynamics across cross-sections of asset returns. The iterative estimation procedures discussed in the first essay are employed to obtain robust estimates of the risk return tradeoff. This essay demonstrates that, at least across the four different portfolios discussed in the empirical example, the relationship between risk and return is linear. The empirical results obtained in this essay differ substantially from existing semiparametric studies of the risk return tradeoff, which have generally uncovered a nonlinear relationship between risk and return. In the final essay, which is joint work with my advisor Eric Renault, we develop a new two-step extremum estimation procedure and compare this new procedure with existing iterative alternatives. In the confines of Gaussian copula models, we demonstrate that this new two-step procedure obtains much more precise parameter estimates, according to various loss measures, at nearly the same computational cost as existing iterative estimators commonly used in applications.