Pustaka Ilmiah Universitas Padjadjaran

Abstrak

Comparison Of Differencing Parameter Estimation From Arfima Model By Spectral Regression Methods

Spectral regression method is one of the popular methods for estimating the difference parameter of ARFIMA(p,d,q) model. Spectral density function of ARFIMA(p,d,q) model was formed to construct linear regression function for estimating the difference parameter d by Ordinary Least Square (OLS). This method has attracted many researchers because it could cope the difficulty in derivation of the autocovariance function of ARFIMA(p,d,q) model. The estimation of d by using regression method could be done directly without knowing p and q parameter. This method was proposed by Geweke and Porter-Hudak (1983) and modified by Reisen (1994) with smoothing periodogram by parzen window. Then, Robinson (1995) added l trimming on this periodogram. Hurvich and Ray (1995) and Velasco(1995a) used modified periodogram by cosine – bell tapered function, Velasco (1999) changed independent variable of spectral regression 2sin(ωj/2) by j ( index of periodogram frequency). In this paper we will compare the estimation accuracy among five methods by using simulation study in two conditions, i.e clean data and data with outlier.

From simulation results, GPH method shows a good performance in estimating the differencing parameter of ARFIMA model both clean data and data with outlier. Above all, estimation of spectral regression methods are better for ARFIMA(1,d,0) data than for ARFIMA(0,d,1) data.