Bringing judgment into combination forecasts

This research investigates the benefits in forecast accuracy by combining judgmental forecasts with those generated by statistical models. Our study differs from prior research efforts in this area along two important dimensions. First, two different types of judgmental forecasts are evaluated for combination with statistical forecasts — one based on contextual knowledge and one based on technical knowledge. Contextual knowledge is information gained through experience on the job with the specific time series and products being forecasted. Technical knowledge is information gained from education on formal forecasting models and data analysis. Second, we investigate the conditions under which adding judgment to combination forecasts helps the most. Specifically, we test the improvement as a function of time series variability. Our results show that judgmental forecasts based on contextual knowledge, rather than technical knowledge, are the better input into combination forecasts. Bringing judgmental forecasts based on contextual knowledge into combination forecast improves forecast accuracy over the individual statistical and judgmental forecasts. However, the benefit attained from including contextual knowledge in the combination depends on the amount of inherent variability in the time series being forecast. More contextual knowledge is needed for combination forecasts if a time series has more data variability. If the amount of variability is low, less emphasis should be given to contextual knowledge when making combination forecasts. In general, our findings suggest a linear relationship between the amount of contextual knowledge needed and data variability.