Comparação de testes de raiz unitária e cointegração em modelos de longa dependência

Abstract

In this work two tests based on the bootstrap technique are considered to test the presence of unit root and relations of cointegration in long memory processes. The proposed tests use the GPH estimator considered by Geweke and Porter-Hudak (1983) for estimating the memory parameter d of ARFIMA(p,d,q) models. Monte Carlo simulations have been used to compare the performances of the suggested tests with other widely used tests in the literature. For unit root, the ADF test (Dickey and Fuller, 1981) and the test based on the asymptotic distribution of GPH have been considered. In the cointegration evaluation, the Engle and Granger test (1987), the Johansen procedure (1990) and the test based on the asymptotic distribution of the GPH have been used. The results show that the bootstrap technique is a good alternative for the construction of hypothesis tests on the parameter of long memory models. In general, the suggested tests presented satisfactory results if compared to the usually applied tests in econometrical studies.