Accuracy of Estimation of Parameters of Linear Regression on Errors in Variables

We consider ultimate accuracy of estimation of parameters of linear regression under the presence of bounded errors of measurements of input variable and regressors on usage of several main nonstochastic estimation methods. It was shown that for sufficiently low level of errors of regressors the minimax approach makes it possible to obtain exact values of estimated parameters. For high level of these errors only polyhedron method with explicit consideration of regressors errors makes it possible to obtain exact values of parameters on realization of usual requirements to sequences of errors and input data without errors. The obtained results are illustrated by means of the numerical example.