Consistent estimation for some nonlinear errors-in-variables models (Q918108)

When variables enter into an equation nonlinearly and these variables are latent in the sense that they are subject to measurement errors, then complicated problems of identification and estimation arise. A set of minimum distance estimators is proposed in such a framework and an alternative set of regularity conditions is considered which ensures the consistency of maximum likelihood estimators for general nonlinear models when the explanatory variables are unbounded. A two-step estimation procedure is also suggested. In the first step a consistent estimate of a subset of parameters is obtained. The second step estimates the set of the parameters under the condition that the first step estimates are in fact the known parameters. The conditions under which this two-step estimator is consistent are also established.