Temporal Data Mining

Prediction    ◾    141

2. To nd the most important lags. ese are lagged versions of the

time series with the highest autocorrelation.

ese, as will be explained next, can be used to compute the autoregres-

sion equation of the time series. Table 4.9 shows the autocorrelation for the

rst three lags of the S&P 500 data shown in Table 4.8. As we see from the

table, the rst lag has the highest value.

4.3.7 Autoregression

In autoregression, the main idea is to perform time series forecasting via

a regression equation where the independent variables are lagged values of

the dependent variable. A general form of the autoregression equation is

Yt ...