To Aggregate or Not to Aggregate: Should decisions and models have the same frequency?

We examine the situation where hourly data are available to design advertising-response models, whereas managerial decision making can concern hourly, daily or weekly intervals. The key question is how models for hourly data compare to models based on weekly data with respect to forecasting accuracy and with respect to assessing advertising impact. Simulation experiments suggest that the strategy, which entails modeling the least aggregated data and forecasting more aggregate data, yields better forecasts, provided that one has a correct model specification for the higher frequency data. A detailed analysis of three actual data sets confirms this conclusion. A key feature of this confirmation is that aggregation affects data transformation to dampen the variance. The estimated advertising impact is sensitive to the appropriate transformation. Our conclusion is that disaggregated models are preferable also when decision have to be made at lower frequencies.