A New Method for Parameter Estimation of the GNL Model Using Real-Coded GA

Abstract

In this paper, a new parameter estimation method is proposed for the generalized nested logit (GNL) model using real-coded genetic algorithms (GA). We propose a method to recalculate and verify whether the offsprings violate constraints. In addition, we improve the selection and mutation operators in order to find the higher log likelihood. In the numerical experiments, the log likelihood of our method is compared to that obtained by the Quasi-Newton method and the normal real-coded GA, which use SPX and JGG, and not the mutation operator, with the actual point of sales data. Thus, we prove that our method finds a higher log likelihood than conventional methods.