On the validity of a nonlocal approach for MHD turbulence