An asymptotic minimax risk bound for estimation of a linear functional relationship (Q793479)

In this paper, the so called ''model with errors in both variables'': \(X_ i=\xi_ i+\zeta_{1i}\), \(Y_ i=B\xi_ i+\zeta_{2i}\), \(i\in {\mathbb{N}}\), is investigated, where \(\xi_ i\), \(i\in {\mathbb{N}}\), is an unknown sequence of nonrandom (p-q)-vectors and B is an unknown \(q\times(p-q)\) matrix, \(\zeta^ t_ i=(\zeta^ t_{1i},\zeta^ t_{2i})\), \(i\in {\mathbb{N}}\), are iid \(N_ p(0,\Sigma)\)-distributed errors. The main results of this paper are as follows. The first one is that the maximum likelihood estimator \(B_ n\) of B, based on the first n samples, is locally asymptotically minimax. The second one is that a lower bound for the error probability of any estimate of B is found. The third one is that the convergence rate of the distribution of the maximum likelihood estimate \(B_ n\) of B, to a limiting normal distribution is established.