Overview of Time Series Analysis

A time series is a sequence of observations at equally spaced points in time. In GenStat, a time series is stored in a in the order in which the observations occur. GenStat cannot deal explicitly with unequal spacing of time points; to analyse this type of data you need to perform some sort of adjustment or interpolation of the values to produce equally spaced points in time. Allowance is made for missing values in the data, but these should not represent more than a small proportion of the data.

The basic steps of model identification (or selection), model estimation (fitting the model to the observed data to obtain estimates of the parameters), and model checking, for univariate time series data, are provided within GenStat for Windows. This includes graphical display of data and calculation of sample statistics, simple ARIMA model fitting and forecasting from these models.

A wide range of facilities for fitting more complex Box-Jenkins models (including regression with correlated errors or transfer function models) is available in command mode. This includes several procedures for graphical exploratory analysis of time series data, including spectral analysis. Commands also enable filtering of time series and calculation of Fourier transforms for use in spectral analysis. See TSA for further details of the commands available.