References

This study proposes a method of modeling long-memory phenomenon with time-varying long-memory characteristics. By adopting the ideas of Pintore and Holmes [1], a non-stationary covariance function is constructed by evolving the spectral representation over time in which the spectrum at each time follows a fractionally integrated process. The long-memory parameter may be time-varying but deterministic, and can be further specified via a parametric function or nonparametric splines. Maximum likelihood is used for parameter estimation which performs well as shown in the numerical examples. To illustrate the application, the methodology i