The dispersing characteristics of static models of stochastic objects

The dispersing characteristics determining quality of static parametrical models of stochastic objects are constructed, and their estimations are found at the assumption, that the conditional variance of an output is constant. Parametrical models are nonlinear on the channel "input - output" but they are linear concerning optimized parameters. It is shown that the nonparametric estimation of regression is close to objectively existing ideal model and according to dispersing characteristics it is possible to keep up with change of quality of suboptimal parametrical models at selection of their structure. Quality of nonparametric and parametrical models compared on several numerical tests with various level of noise.