A Proporção de pobres do Ceará está subestimada? uma aplicação da teoria do valor extremo

Abstract

The research of estimating the distribution of poverty in Brazil have been concentrated with the use of tools of statistical inference or the inefficient use of some ad hoc distributions, or through studies of convergence. This work contributes to a discussion of different methods of non-parametric statistical inference in order to estimate the evolution of the density of poor individuals by smoothing a stochastic Kernel density, based on data from the National Household Survey - PNAD (2001, 2003, 2005 and 2007), and it concluded that the average family income per capita has been increasing during the analysis. By taking the state of Ceará as representative for empirical analysis, it was contrasted two methodologies for estimating the proportion of poor in the period, which are: The traditional method (use of Discrete Uniform formulation) and an application based on the theory of extreme values (TVE). To provide more support to what is proposed, it made a scene contrasting the poverty rates from the Census-2000 with those obtained by IPEADATA and evaluated by the methodology employed here, the latter two based on PNAD-2001 dataset. By contrasting the results it is concluded that there is strong evidence of underestimation of the poverty rates as the traditional methodology is applied.