What is a composite forecast?
A composite forecast exists because no single forecasting method captures every demand signal, so blending several offsets each model’s blind spots. Planners feed outputs from different demand forecasting methods into one number that guides purchasing and stocking decisions. Across 30 empirical comparisons, Wharton research found that equally weighted combined forecasts reduced ex ante errors by an average of about 12.5%, ranging from 3 to 24 percent.
Fishbowl’s demand forecasting guide, which covers the individual methods a composite forecast blends. Source.
Key characteristics
- Blends multiple methods: it combines time-series, causal, judgmental, and statistical outputs instead of relying on one.
- Combines by averaging or weighting: the forecasts merge through a simple average or through weights set by each method’s past accuracy.
- Aims for higher accuracy: the goal is to cancel out any single model’s errors. In the M4 Forecasting Competition, across 100,000 time series, the top-performing methods were all combinations of mostly statistical methods, more accurate than pure statistical or pure machine learning methods.
- Works best with diverse inputs: methods that draw on different data add more value when combined, such as machine learning forecasting paired with human judgment. One study found that out of the top 17 most accurate methods, 12 were combinations of forecasting models.
- Feeds planning decisions: the blended output informs reorder points and replenishment through advanced analytics for reordering.
Example
A craft cold-brew coffee roaster blends a statistical time-series forecast, a machine learning demand model, and sales-team input into one composite number. The roaster then uses that number to set purchase orders (POs) for green coffee beans and packaging. Fishbowl guidance on how to combine multiple forecasting inputs follows the same approach: pair historical sales with real-time trends, seasonality, and supplier variability.
Related terms
FAQ
Is a composite forecast the same as an average of forecasts? A simple average is one type of composite forecast, but others weight each method by its past accuracy.
When should a business use a composite forecast? Use one when demand is volatile or no single method proves reliable, since blending methods produces a steadier, more accurate prediction.
Does combining forecasts always improve accuracy? Not always, though research shows it consistently reduces error on average when the combined methods are diverse and each adds real information.