Ocorrerá excepcionalmente no dia 30/05/2022 (segunda-feira) às 14h, presencialmente o Seminário de Pesquisa “EFFICIENT NONPARAMETRIC ESTIMATION OF DYNAMIC REGRESSION SYSTEMS” na área de Economia Aplicada, ministrado pelo professor Carlos Martins-Filho (Universidade do Colorado Boulder).
Resumo: This paper considers the estimation of a system of nonparametric dynamic regressions with error covariance that accommodates both contemporaneous and inter-temporal correlation. First, using a pre-whitening transformation that assumes knowledge of the error covariance, we propose a two-step local-linear system estimator that is asymptotically normal under certain mixing conditions. In addition, the system estimator is asymptotically efficient relative to the traditional local-linear estimator. Second, under the additional assumption that the error covariance depends on a finite dimensional parameter, we establish that our system estimator constructed under knowledge of the true parameter and under a consistent estimator are asymptotically equivalent. We show that several regression models in the existing literature are special cases of our structure. These include, but are not limited to, nonparametric dynamic seemingly unrelated regressions and certain dynamic panel data regressions.
Local: o seminário será presencial na sala 401A da FCE.