Pooled Log Periodogram Regression (Q4544839)

Let \(X(t)\) be a stationary Gaussian long-memory process whose spectral density has the form \(f_x(\lambda)=|1-\exp(i\lambda)|^{-2d}f_u(\lambda)\), where \(d\in (-1/2, 1/2)\) and \(f_u(\lambda)\) is a symmetric, periodic positive smooth function. The goal of this paper is the analysis of estimators of the parameter \(d\) which governs the long-memory property of \(X(t)\). The authors propose the so-called ``pooled log periodogram regression'' estimator based on a set of \(mL\) periodogram ordinates with \(L\to \infty\) and compared it with the conventional log-periodogram estimator introduced by \textit{J. Geweke} and \textit{S. Porter-Hudak} [J. Time Ser. Anal. 4, 221-238 (1983; Zbl 0534.62062)] which is sometimes called GPH estimator. Consistency and asymptotic normality of the pooled regression estimator are proved. The pooled estimator is shown to have smaller asymptotic variance, but larger asymptotic bias, than the GPH estimator. Finite sample performance is studied by simulation. Simulation results are discussed and applications of both estimators are illustrated on real data of US inflation series and stock returns.