Purpose

    An autocorrelation is the correlation of a series with itself, shifted by a particular lag of k observations. The plot of autocorrelations for various lags is a crucial tool for determining an appropriate model for ARIMA analysis.
 

Preparations

    Run Statistics→Time Series/Forecasting→Autocorrelation and Partial Autocorrelation....

Results

    The computations of the autocorrelation coefficients r_k follow the standard formulas, as described in most time series references (e.g., Box & Jenkins, see Bibliography).
Standard error of rk. Under the assumption that the true moving average process in the series is of the order k-1, then the approximate standard error of rk is defined as: StdErr(rk) = Ö{(1/N) * [1+2*S(ri2 )]} (for i = 1 to k-1). Here, N is the number of observations in the series. However, under the assumption that the series is a white noise process, that is, that all autocorrelations are equal to zero, the standard error of rk is defined as: StdErr(rk) = Ö{(1/N) * [(N-k)/(N+2)]}. Select the White noise standard errors check box from Advanced Options to compute the standard errors in this manner.
Box-Ljung Q. At a given l ag k the Box-Ljung Q statistic is defined by: Qk = n*(n+2)*S[ri2 /(n-i)] (for i = 1 to k). When the number of observations is large, then the Q statistic has a Chi-Square distribution with k-p-q degrees of freedom, where p and q are the number of autoregressive and moving average parameters, respectively.

    A partial autocorrelation is the correlation of a series with itself, shifted by a particular lag of k observations, and controlling for the correlations for all shifts of 1 through k-1. The plot of partial autocorrelations for various lags is a crucial tool for determining an appropriate model for ARIMA analysis. The computations of the partial autocorrelation coefficients fk follow the standard formulas, as described in most time series references (e.g., Box & Jenkins, 1976).
Standard error of fk. Under the assumption that the true autoregressive process in the series is of order p £ k-1, then the approximate standard error of fk is defined as: StdErr(fk) = Ö(1/N). Here N is the number of observations in the series.