A STOCHASTIC APPROXIMATION-ITERATIVE LEAST SQUARES ESTIMATION PROCEDURE

We consider the general nonlinear regression problem Y(x)=g(theta;x ) +lepsilon . A survey of some classical methods and stochastic approximation procedures for estimating theta is first given. We solve the nonlinear regression problem by considering the optimal stochastic approximation procedure by [3],[4]. This leads us to introduce a new procedure , called Stochastic Approximation Iterative Least Square Procedure SA -ILS procedure. The new procedure is applied to a number of nonlinear regression models. We report on the results of a simulation investigation which indicate that the new procedure is highly efficient with respect to the number of observations required to obtain the parameter estimates for given regression problem.