On applying mixed RP/SP models to policy forecasts: some new evidence