A major shortcoming of the U.S. leading index is that it does not use the most recent information for stock prices and yield spreads. The index methodology ignores these data in favor of a time-consistent set of components (i.e., all of the components must refer to the previous month). An alternative is to bring the series with publication lags up-to-date with forecasts and create an index with a complete set of most recent components. This study uses tests of ex-ante predictive ability of the U.S. leading index to evaluate the gains to this new 'hot box' procedure of statistical imputation. We find that, across a variety of simple forecasting models, the new approach offers substantial improvements.