A note on cointegrated relationships estimated with genetic algorithms

Abstract

Estimation techniques based on Genetic Algorithms (GA) have been studied in the presence of cointegrated variables. Several applications of GA to time-series have ignored the fact that the equation estimated by GA might be spurious. In this line, in this study it is shown that: (1) GA robustly detects this kind of relationship when the process contains a linear cointegrated relationship, (2) estimated models provide real fitness instead of spurious fitness. The well known cointegrated relation between income and consumption is estimated using GA.

Notes

Such a linear combination can be thought as transitory consumption, and it seems reasonable for this to be stationary.

The sample period has been selected in order to compare results with those supplied by Gujarati (2003) on the same series.

Robustness of the estimated model has been contrasted for different seeds. In all cases the fucntional form was

 + non-linear.