Generalized empirical likelihood for a continuum of moment conditions

This paper extends the generalized empirical likelihood method to the case in which the moment conditions are defined on a continuum (CGEL). We show, for the iid case, that CGEL is asymptotically equivalent at the first order to the generalized method of moments for a continuum (CGMM) developed by Carrasco and Florens (2000). Because the system of equations that we need to solve becomes singular when the number of moment conditions converges to infinity, we treat CGEL as a nonlinear ill-posed problem and obtain the solution using the regularized Gauss-Newton method. This numerical algorithm is a fast and relatively easy way to compute the regularized Tikhonov solution to nonlinear ill-posed problems in function spaces. In order to compare the properties of CGEL and CGMM, we then perform a numerical study in which we estimate the parameters of a stable distribution using moment conditions based on the characteristic function. The results show that CGEL outperforms CGMM in most cases according to the root mean squared error criterion.