Three Essays on Inference for Poverty and Inequality Metrics

Book excerpt: "Income inequality and poverty have important economic, political, and social implications. Despite this, existing measures of inequality and poverty are plagued with problems. First, existing inequality measures are not defined for possibly negative variables.Extended inequality measures were recently proposed to overcome this difficulty but used essentially as descriptive tools without statistical inference.Second, when comparing poverty or inequality measures, standard procedures require that the two samples are independent or completely dependent (i.e., matched pairs).These assumptions are not satisfied by many household surveys, since they are rotated samples.Third, it is known in the literature that two popular inference procedures, namely asymptotic and bootstrap inference methods, can have poor performance in finite samples or in the presence of heavy tails.The thesis contributes to statistical inference on poverty and inequality measures. We first make the extended inequality measures formal by providing estimation and asymptotic inference methods including uniform inference on extended Lorenz curves.For comparing the extended (and traditional) inequality measures, we propose methods that are robust to independent, completely, and partially dependent samples.Finally, an exact nonparametric inference method is provided for comparing poverty measures.We provide simulation evidence to support the proposed inference procedures.Our methods are applied to several empirical applications including financial income inequality in Italy, cross-sectional dispersion of US stock returns and rural-urban poverty disparity in China"--