Testing for periodicity in functional time series (Q1991685)

The authors study a classical functional model when a deterministic trend is contaminated by a functional random noise process. In the paper the deterministic trend is a periodic function. Such models are important for various time series applications where seasonal or cyclic components are present. Two cases of the random noise process are studied: Gaussian white noise and weakly dependent non-Gaussian processes. Three types of periodic components of increasing complexity are studied. The authors develop a general functional ANOVA test for dependent data and test statistics for each type of periodic components. Then the projection of functional data onto a low-dimensional basis system is studied. It is demonstrated how to apply multivariate procedures to these projections. Asymptotic properties of the tests are derived and their consistency is established. Numerical studies of local power of the tests are given. Applications of the theoretical results are demonstrated using pollution data.