A0552
Title: Expression of circular time series models with complex-valued stochastic processes
Authors: Hiroaki Ogata - Tokyo Metropolitan University (Japan) [presenting]
Abstract: Circular data can be expressed by points on a unit circle of the complex plane. The proposal is to express circular time series models by complex-valued stochastic processes. Covariance and complementary covariance functions are first introduced. The Fourier transforms of them are called power and complementary power spectral density functions. By leveraging them, a periodic component of the circular time series data is extracted in the sense of the time axis. As examples of circular time series models, the circular mixture transition distribution model and the wrapped autoregressive model are seen. The former is explained to have the same autocovariance structure as that of the complex-valued autoregressive process, while the latter has non-vanishing autocovariance even when the lag h tends to be infinity.
