Abstract:
The paper investigates the robustness of the minimax property of the Hotelling test for distributions close to normal. It is proven that the $T^2$ test maximizes, among all tests, the level $\alpha$ of minimal power on the set of alternatives to within $O(\varepsilon)$.