Estimating the effective tax functions with genetic algorithms

In this paper, genetic algorithms are employed to the estimation of the effective tax functions, which is derived from the principle of equal sacrifice and is a nonlinear function between individual effective income tax rates and economic income. The conventional nonlinear techniques to approach this problem, such as the Gauss-Newton method, suffer from the possibility of an exact or near singularity of the matrix to be inverted and the possibility of slow convergence. Therefore, under these situations, genetic algorithms may be considered as a complement for them. In this paper, based on the data from the ''Annual Survey on the Distribution of Per Capital Income'', the performance of genetic algorithms is compared with that of the Gauss-Newton method and the Marquardt method in estimating the effective tax functions from 1982 to 1994. Based on the criterion of mean square errors, genetic algorithms perform uniformly superior to these two methods in all these years. Following the earlier findings by Dorsey and Mayer (1995), this result further confirms the promising features of genetic algorithms in econometrics.