Bootstrap ponderado: uma avaliação numérica

by Inácio, Felipe Chaves

Cross sectional data usually display some form of heteroskedasticity. Since the ordinaryleast squares estimator of the linear regression parameters remains unbiased and consistentunder heteroskedasticity of unknown form, it is common practice to use it together with aconsistent estimator of its covariance matrix. The chief goal of this dissertation is twofold.First, we investigate the sensitivity of Cribari?Neto amp; Zarkos?s (2004) inversely adjustedweighted bootstrap estimator to different resampling schemes. Second, we investigate theimpact of non-normality on inference under heteroskedasticity, in particular on inferenceperformed using the aforementioned estimator