Forecast Revisions as a Feature, Not an Embarrassment

ForecastingNowcastingRisk

Forecasts age quickly when tourism seasons shift or energy prices jump. In Thailand Metric Echelon forecasting workshops, we treat revisions as a dataset rather than an apology. Participants build a simple revision log: what release moved, how the model reacted, and what judgment overlay changed. This discipline matters for analysts who brief executives monthly. A fan chart that never moves erodes trust faster than a wide interval that updates honestly. We practice narrating why a band widened even when the point forecast barely budged. We also separate model error from data error. A late revision because the statistical office corrected a print differs from a model that misread seasonality. Mixing the two breeds cynicism. Cohorts leave with a template that tags each revision with a cause code. The last piece is humility about nowcasts. Bridging equations are useful, but they inherit assumptions from the soft indicators feeding them. We ask participants to write one paragraph on what would falsify the bridge in the next quarter. That paragraph rarely makes the slide deck, but it saves teams from overfitting narratives.