A prediction method applicable to steered time scales

The approach to the statistical prediction of a time scale process reported in this paper follows a sequence of three operations: pre-processing, prediction and post-processing. The pre-processing operation consists into removing both the voluntary (steering corrections) and the involuntary (accidental rate steps) statistical discontinuities present in the original process. The output from the pre-processing is a well behaved process with stationary statistical properties to which a statistical prediction algorithm can be applied. The post-processing consists into adding back the removed rate steps to transform the prediction of the pre-processed time scale into a prediction of the original time scale. The paper discusses the algorithms used in the prediction of UTC(CH) and reports the performance obtained.