Forecasts made by econometricians are typically conditioned on actual values of explanatory variables, even when at the time of the forecast, such variables might not be available. As a first step, one might test the adequacy of econometric specification by comparing conditional post sample forecasts with those of a univariate ARIMA model. Second, when explanatory variables must themselves be forecast, those for which this can be done only badly, should be omitted from the final model. A better forecast will result. An example of screening out badly forecasted explanatory variables is presented.