Autocorrelations and Autocovariances

The autocorrelation and autocovariance function is an important tool for describing the serial (or temporal) dependence structure of a univariate time series. It reflects how much correlation is present between lagged observations.

To estimate autocovariance or autocorrelation:

Choose Statistics __image\ebd_ebd84.gif Time Series __image\ebd_ebd85.gif Autocorrelations. The dialog shown below appears.

__image\autocorr.gif

The Autocorrelations and Autocovariances dialog has the following options:

Data

Data Set

Select or enter the name of a data set having time series as columns.

Variable

Select the column containing the time series to be analyzed or modeled.

Tip You can enter the name of a time series object directly in Time Series. For example, enter lynx to compute the correlogram for the lynx time series.

Options

Estimate Type

Select an estimate type from the dropdown list. The following options are available:

autocorrelation  to estimate the autocorrelation function (the default)

autocovariance  to estimate the autocovariance function

partial autocorrelation to estimate the partial autocorrelation.

Change Maximum Lag Default

Select to give a value for the maximum number of lags at which estimates are calculated. If this is not selected, the default is a number proportional to the logarithm of the length of the series.

Maximum Lag

Enter the desired maximum number of lags at which to estimate the autocovariance or autocorrelation function.

Results

Save As

Enter the name for the object in which to save the results of the analysis.

Plot Results

Select this to display a plot of the estimates of covariance or correlation against their corresponding lags. A 95% confidence interval around the zero line is included.

Related S-Plus language functions

acf, lag, acf.plot, ar, menuAcf