ABSTRACT
Unfortunately, despite the reassuring scientific nomenclature, time series analysis is still very much an art. Even in the relatively well understood linear world of ARIMA models and correlation-based measures, there is still considerable room for human interpretation, such as in model specification, in diagnostics evaluation, and (perhaps most importantly) in weighing the overall merit of a resulting model. This requirement for expert care is, if anything, amplified with the introduction of new nonlinear tools, especially in the area of model specification. None of the new tools are comprehensive enough to warn their users of all possible pitfalls, and thus they cannot be applied indiscriminately.
