Optimization of algorithms for estimating parameters of stochastic systems in conditions of indeterminacy

The paper is concerned with parameter estimation for a vector-valued regression model with correlated measurement errors. A general estimation method leads, even with incomplete knowledge of the correlation structure of measurement errors, to on adaptive estimate of unknown parameters which is asymptotically at least as accurate as the best linear unbiased estimate. The statistical properties of the method are studied, and the results of numerical experiments are given.