By James Bell
The Tracking Signal is a great high-level way to communicate and manage forecasting error bias. This can be really useful when using automatized or computerized forecasting systems.
One way to use a tracking signal is to set up thresholds of acceptable error using color as representations of the validity of forecasting. While an actual traffic light style metric with green, yellow, and red as visual sounds very fitting, keep in mind that these colors can be difficult if the reader is color blind. When we see our error increase over time, this tells us that the current forecasting model is less and less valid and no longer a great predictor of future actuals. Showing trends over time on the question of validity of forecasting may surface seasonality or possibly invalidate previously made assumptions and methodology.
This is the Cumulative Sum of Forecast Errors which shows the overall bias due to it’s use of non-absolute values. Because some errors are negative and some positive, the systemic over or under presents itself as CFE. This link will show you the calculation and explain the variables in calculating CFE.
Mean Absolute Deviation is beneficial when looking at an individual forecast. It’s the average of the absolute value of the errors. This link will show you the calculation and explain the variables in calculating MAD.
D.W. Trigg and A.G. Leach introduced a slightly different calculation that involves smoothing errors. While MAD is often transformed into a weighted average that does involve smoothing, Trigg’s method gives us yet another variation. From my brief research on this, it appears SAP help and many other organizations use this variation. To read more about this, I’ve included the citations for Trigg’s original work, and then the Trigg and Leach’s later work that you can look into for further research.
15 March 2020
15 December 2019
14 November 2019
07 November 2019
12 October 2019
26 September 2019
11 September 2019
Powered By Impressive Business WordPress Theme