Kernel-based off-policy estimation without overlap: Instance optimality beyond semiparametric efficiency

Open in new window