*4.3. Regression Results*

After analyzing the TS applying Principal Component Analysis, we used the Principal Components Regression method (PCR) to adjust a model of linear regression by least squares using the four components generated on a previous subsection. For this test, we select the TS by period and divide them in a subset of training and another subset of test with 80% and 20%, respectively. The estimated error of prediction was calculated with the Mean Square Error (MSE), and the results are presented in Table 5, the best prediction values were obtained in the Quarterly and Weekly periods, and the high prediction error was obtained in the Yearly period; it is important to remember that the subset of Yearly period is composed only of 56 TS.

**Table 5.** Mean Square Error of prediction with a linear regression model.

