Essays in econometrics: Selection of common regressors in large panels and finite sample analysis of block bootstrap in nonlinear models
Abstract
The first essay proposes a formal econometric method to consistently select the optimal subset of explanatory variables in large-scale panels. The selection process is implemented in two steps. First, an optimal order of the candidate variables is defined by their partial contribution in the total variance. The best subset of regressors is then determined by applying an information criterion to the set of ordered candidates. This two-step selection technique is sequentially consistent and has good sampling properties under weak dependence in the errors. The number of cross sections increases the likelihood of correct specification both asymptotically and in small samples. Moreover, this search procedure is much less computer-intensive and requires fewer computations than any exhaustive search approach. The second essay investigates the small sample properties of an approximate factor model solution in the context of large-scale panels. This essay uses the estimation method of Bai and Ng (1999) and assesses the sampling behavior of the estimated common components, common factors and factor loadings without restrictions on the panel dimensions. The Monte Carlo results show evidence for normality and suggest that the asymptotic theory of Bai (2001) applies in relatively small panels with weak serial- and cross-section correlation in the errors. The third essay reevaluates the inference in the generalized method of moment estimation of the nonlinear rational expectations model of Hansen and Singleton (1982) using the improved block bootstrap of Hall and Horowitz (1996). The study assesses the refinements achieved by block based resampling method as an alternative to the asymptotic theory in small samples and as an improvement upon the conventional bootstrap in the case of temporal dependence. The simulation study assesses the differences between the nominal and the true rejection probabilities of test statistics and coverage probabilities of symmetrical confidence intervals. The results confirm the well-known size distortion of the asymptotic approximation and show lower levels of errors in the rejection probabilities when using block bootstrap inference.
Recommended Citation
Rachida Ouysse,
"Essays in econometrics: Selection of common regressors in large panels and finite sample analysis of block bootstrap in nonlinear models"
(January 1, 2002).
Boston College Dissertations and Theses.
Paper AAI3066228.
http://escholarship.bc.edu/dissertations/AAI3066228
View More
