Abstract:
This article proposes a number of frequency domain estimation procedures for
the ARFIMA model based on a spectral generalized method of moments (SGMM)
estimator. It is shown that the SGMM estimator includes as special cases; the
GPH regression estimator, a spectral instrumental variable estimator, and a spectral
nonlinear least squares estimator. A discussion of the large sample properties for
the SGMM estimator for both general and specific ARFIMA models is presented.
A Monte Carlo experiment shows that the small sample properties of the SGMM
type estimators are quite promising when compared to time and frequency domain
maximum Likelihood.
