As we know forecasting is a key element of demand planning process. There are various methods available for forecasting, starting from qualitative forecasting to quantitative forecasting. These two methods have their own use. In this article, we are going to talk about composite forecasting and its uses. Composite forecasting is a most talked about term and mainly because of the benefits that it has which a single forecasting model might not be able to provide. There have been numerous studies done across the world to identify the benefits of composite forecasting.
Composite forecasting focuses on using a combination of models (giving weights) to get the better result instead of relying on just one forecast. The main concept of composite forecasting is that you get better results by averaging the forecast numbers from various sources. A company might get a number of forecasts from different teams and there would be an algorithm in place to generate the forecast numbers statistically. Composite forecasting considers that there might be valuable information stored in some form, in various forecasts available and thus it is not a good idea to discard the rest and keep only one. Discarded forecast numbers usually always contain some independent useful information and thus it is not a good idea to discard them and keep only one forecast. A way to mitigate the risk of having a greater impact of one forecast and less impact of other forecasts is by weighting the forecasts. A weighted average of the forecasts may give smaller mean squared error than any one forecast or simply averaging the forecasts. This helps in eliminating biases in the forecasts available to the planning team. This also helps both the team and the company as the efforts of various people (sales, supply chain, finance, and management) is not lost and a synergy is built in the system.
“Evidence by a biologist named Levin (1966) suggested that, rather than building one master model of the real world, one should consider building several simpler models that, among them, would use all available data, and then average them. Bates and Granger (1969) followed Levin’s research by testing combined forecasts of the international air travel market. They started with the premise that improvements are greatest if the forecasts are based upon different information, rather than merely upon different assumptions about the relationships. They used five different methods to obtain one forecast. Nine different combined forecasts were developed from the five original sets of forecasts using various weighting schemes to combine pairs of forecasts. They combined methods like exponential smoothing and BoxJenkins, and they found that the error of these two individual forecasts were much higher than the error of combined forecasts of both methods. Overall, the errors of the combined forecasts were smaller than the error of either of the two components in all but one case, where the combined forecast and best component tied. These findings imply that gains can be achieved by using a set of simple extrapolation methods and combining the forecasts.”
Composite forecasts are particularly widely used in cases of new product launches where there are inputs from many people on the product’s behavior and no one is very confident. In such cases, composite forecasting would be advantageous as it will combine the forecast numbers from all the sources and give a better result. It makes the job of demand planners much easier as they can easily combine the forecasts and work more on analyzing the forecast and other scenarios. Companies can benefit from such method when faced with dilemmas of arriving at the best result.
It is however not advisable to use composite forecasting when the statistical models alone are doing well and there is no sudden fluctuation in the demand.