A comparison of linearization and quadratization domains (Q1265629)

A nonlinear regression model is approximated in the neighbourhood of a given point of the parametric space by the Taylor series either of the first order (linearization) or the second order (quadratization). The bias of a linear estimator of the vector parameter in the linear model must not be greater than a tolerable value for any vector parameter in the neighbourhood. This requirement determines the boundary of the neighbourhood, i.e.\ the linearization domain. In an analogous way the quadratization domain is defined (quadratic estimators are considered in this case). A comparison of the domains of these two kinds is given in the paper.