Parameter Estimation in Non-linear Regression Models

摘要

In many applications, the relationship between the dependent variable and an independent regressor is non-linear in parameters. In such situations, we do not get optimum estimates of parameters in closed form and various non-linear optimization algorithms are used to obtain the optimum estimates. These algorithms are iterative in nature and need good initial estimates of parameters as seed values for a faster and global convergence. This paper proposes various methods based on finite differences to estimate the parameters of non-linear models belonging to the asymptotic regression category. Some published data sets are used to illustrate the application of the proposed methods. It has been demonstrated that the proposed methods produce efficient initial estimates for optimization algorithms.