Genetic algorithm optimization technique for linear regression models with heteroscedastic errors

Abstract

Most widely used statistical technique for estimating cause-effect relationships is the Linear regression methodology. Ordinary least squares (OLS) method, which is valid under certain assumptions, is generally used to estimate the underlying parameters. If the errors are not homoscedastic, OLS estimates lead to incorrect inferences. In this article, use of the powerful stochastic optimization technique of Genetic algorithm (GA) is advocated for estimation of regression parameters and variance parameter simultaneously even when nothing is known about the form of heteroscedasticity. Parametric bootstrap methodology is employed to obtain standard errors of the estimates. The methodology is illustrated by applying it to a dataset.