Broadband semiparametric estimation of the long-memory parameter by the likelihood-based FEXP approach

This paper proposes a semiparametric estimator of the long-memory parameter to fit a fractional exponential (FEXP) model by a likelihood-based approach. We establish that our proposed estimator is more efficient than the FEXP estimator proposed independently by Moulines and Soulier (1999) and Hurvuch and Brodsky (2001), and has the same asymptotic variance as the fractionally differenced autoregressive (FAR) estimator proposed by Bhansali et al. (2006) without pooling the periodogram. The Monte Carlo studies suggest that our estimator outperforms the FEXP estimator or is not inferior to the Gaussian semiparametric estimator (GSE) and will be also empirically effective in non-Gaussian processes.