We distinguish between three different ways of using real-time data to estimate forecasting equations and argue that the most frequently used approach should ge