An example illustrates the elements of a forecast model

This is an example of an FGS forecast.  It uses actual data from an industrial service parts business from 1982 to 1984.  This item has all of the elements of a forecast model.

In green you see the demand history in units per month on a 5-4-4 calendar for this example.  This 5-4-4 calendar is a typical fiscal calendar that has 5 weeks in the first fiscal month of the quarter, then 4 weeks in the next month, and 4 in the third.  Since a calendar like this can cause false seasonality, internally FGS converts demand history to a daily rate before performing its calculations.

The blue line is the trend line for the forecast.  It is projected into the future as the forecast (red). The elements of this forecast model are:

• Level: 712 units per month
• Trend: +15.9 units growth each month
• Seasonality:  1 cycle per year that peaks with a high in the spring and a low in the fall
• Selling days adjustment: the demand daily rate is weighted by the number of days in each forecast period

In the graph you can see these four model elements add up to the total forecast model:

• Orange Level
• + Blue Trend Line
• + Green Seasonal Cycle
• + Red Adjustment for the 5-4-4 Calendar

Residual forecast error

The forecast model is backcast (i.e., projected backwards) and is compared with the demand history.  The difference is called the residual error.  Initially, this is the error used when you calculate safety stock.  The screen to the right shows this item's error distribution . FGS compares the green histogram of the residual errors with the normal error distribution curve (blue).   Obviously these errors are normally distributed. 

SKUs with low, lumpy demand often fit the exponential distribution (red) requiring different safety factors for safety stock.  Getting the right error distribution is essential to predicting service and inventory levels.  Alternatively, this SKU could be put on a less-frequent calendar, for example quarterly (calendar analysis).