Inductive modeling of time series: a detrending approach

The problem addressed involves the mathematical behavior of time series. One is often confronted with a data set that may possess a trend, even though the trend may not be detected by the naked eye. Spectral analysis (either Fourier amplitude spectrum analysis or maximum entropy spectral analysis) may aid in detecting cyclic trends in data, but trends are not necessarily cyclic. A code was developed that strips a time series of its trend. This trend may be a polynomial, an exponential, an autoregressive term, a sinusoid, a combination of these, or some other mathematical expression. As an illustration of the method described above, a time series that possesses very obvious trends is studied. 3 figures (RWR)