Summary
The forecast error tab of Forecast Manager allows you to get some information on how accurate your operational and statistical forecasts are for a given hierarchy node selection.
It is located primarily in the bottom half of the Forecast Manager screen, in amongst the other settings tabs.
Forecast Error Sections
This tab is broken up into three sections, each reporting the same values. Here are the sections:
- Forecast Error - Current - This shows errors measured using all available current data, looking backwards over whatever is your error length horizon, typically 1 year. So, if today was June 1st, it is measuring error from forecast periods May of this year, to June of last year.
- Forecast Errors @ Lead Time Lag- This shows errors measured from the lag snapshot. With Forecast Error Lag in Global Demand Forecast Settings set to Auto (the default), that is the snapshot from one lead time ago, referred to as your "Lead Time Lag", based on the lead time or average lead time of products of the selected hierarchy node. With a fixed lag of 1 to 12, the heading reads Forecast Errors @ Lag N and the snapshot is the one from N forecast periods ago. This gives you the ability to compare how the error of your forecast is changing over time, using the same backwards-looking measure. So, if today is June 1st, and your lead time is 3 months, meaning one lead-time-ago is March 1st, it is calculating errors from February of this year, to March of last year.
- Forecast Errors @ Lead Time Lag vs Current Actuals - Since the backwards looking method can under-report future forecast error due to fitting, the "vs Current Actuals" variant calculates the error of your lead-time-ago snapshot using current data, which now will include data points not known at the time the snapshot was taken. These numbers will therefore typically be higher error than the backwards-looking values, since there is more unknown. So if today is June 1st, lead time is 3 months, so the lead-time-ago snapshot is from March 1st, then it is calculating error from May of this year, to June of a year ago, meaning 3 of the 12 actual sales months included in the error calculation (March, April, May) were not known at the time the snapshot was taken.
Forecast Error Values
Each grouping shows the same table of values. Each row has the same columns available.
Rows
- Benchmark Model - Shows the measures of the benchmark / naive forecast model. Labeled "Benchmark" in the two Lead Time Lag sections.
-
Stat. Model - Measures for your current statistical model. Labeled "Stat Model" in the two Lead Time Lag
sections - Operational Forecast - Measures for your operational forecast. The row is labeled with the name of your operational forecast series
Columns
- wMAPE - Shows the error percent of the row in question, measured over the error length, as wMAPE (weighted mean absolute percent error): the total units the forecast missed by, divided by the total units sold. This column was labeled % (MAE%) before; the calculation is the same. If you would like to learn more about these calculations please read the Forecast Error KPIs topic.
- Units - Shows the units corresponding with that error percentage
- $ - The enterprise value of the error. Values are shown in your system's default currency.
- Bias % - Shows the Forecast Bias Percent of this forecast series
- FVA - Shows the Forecast Value Added of this forecast series relative to its predecessor, such as the statistical FVA relative to the Benchmark forecast. FVA is shown for the Stat. Model and operational forecast rows in Forecast Error - Current, and for the Stat Model row only in Forecast Errors @ Lead Time Lag. The Vs Current Actuals section does not show FVA.
A few things worth knowing about these values:
- A blank cell is deliberate — if total actuals in the window come to zero, or the absolute errors sum to zero, the calculation is marked invalid rather than shown as 0% or a divide-by-zero.
- wMAPE is capped at 100 (10,000%). A cell reading exactly that has hit the cap; the true figure is higher.
Error Calculation Factors
This shows some of the factors and options going into the error values, configurable in the Global Demand Forecast Settings screen.
- Error Length - displays the period over which the average error is being calculated, typically 6 months to 1 year is best.
- LT Ago Lag Period - Shows the number of periods between the current period and the lag snapshot being used for this node. It reads Error Lag Periods when Forecast Error Lag is set to a fixed lag.
Troubleshooting
- Forecast Errors @ Lead Time Lag is actually a snapshot of what Forecast Error - Current looked like on the day the lag snapshot was taken.
- Keep in mind the Safety Stock Calculations by default use the Stat. Model Error
- At the beginning of an implementation project StockIQ may not have enough history to calculate all of the error metrics listed in the Forecast Error Tab. Calculations are limited to start of the Operational Forecast.
Tags/Metadata