ABSTRACT
In practice most real time series have properties that do change with time, albeit slowly in many cases, and we have already met some methods for dealing with non-stationarity. This chapter discusses the problem of non-stationarity more generally, because it is important to understand the different types of non-stationarity that may arise and methods for dealing with them. Cyclical changes in mean can also be dealt with by filtering, by differencing or by fitting a global model perhaps using a few sine and cosine terms of appropriate frequency. Complex demodulation is an alternative approach, which studies signals in a narrow frequency band to see how they change through time. A structural change may produce a short-term transient effect or a long-term change in the model structure, such as a change in mean. A technique, called spline regression, can be used to compute a smoothed approximation to a time series, even when it is unequally spaced.
