Genetic algorithm and simultaneous parameter estimation of the nested logit model

Abstract

The parameter estimation of the nested logit model is conducted either simultaneously or sequentially. This is well known fact that the sequential method yields less efficient estimates than the simultaneous one, although its estimates are consistent and asymptotically efficient. Due to the computational burden, however, the sequential estimation is more often employed. Recently, the genetic algorithm has received a great deal of attention for its efficient solution for the nonconvex multidimensional problem. Generally, the parameter calibration of the nested logit model is nonconvex and hence, its solution may not be a global one. A hybrid estimation algorithm combining GA and the gradient method for the simultaneous nested logit model estimation has been suggested. The hybrid algorithm is implemented in a code, named G-Logit. An experimental test results, although limited, showed that the hybrid algorithm effectively search the solution domain of the nested logit model calibration and produced better estimates than the solely used gradient method.