Refinamentos de testes na classe dos modelos não-lineares simétricos heteroscedásticos

Abstract

In this thesis, we deal with improvement for hypotheses tests in heteroscedastic symmetric nonlinear models. First, we derive Bartlett adjustments for likelihood ratio statistics and modified profile likelihood ratio statistics in order to improve likelihood ratio and modified profile likelihood ratio tests, respectively. Next, we calculate a type-Bartlett adjustment to improve the gradient test, a new hypotheses test proposed by Terrell (2002) which is asymptotically equivalent to likelihood ratio, Wald and score tests. We also treat about the local power of likelihood ratio, Wald, score and gradient tests in heteroscedastic symmetric nonlinear models. For each approach, we develop a Monte Carlo simulation study in order to evaluate the performance of the tests in finite-size samples.