User Tools

Site Tools


Sidebar

Back to Main Website

Video Tutorials

User Guide

1. General Information

2. Starting Up

3. Connecting data

4. Demand and Sales Forecasting

5. Inventory Planning

6. Reference

evaluating-forecast-accuracy

4.10. Evaluating the Forecasts

It is always expected that actual values will differ from the forecast. Measuring this difference as a KPI and assessing it for the safety stock calculation is an important part of demand forecasting process.

There are many error measures used to evaluate the forecast. Among them:

  • Mean Squared Error (MSE);
  • Root Mean Squared Error (RMSE);
  • Mean Absolute Error (MAE);
  • Mean Absolute Percentage Error (MAPE).

Streamline evaluates forecast based on MAPE. It is obvious that in order to evaluate the forecast the actual data for the forecasted period must be imported.

Consider an example. Assume that today is the end of Jul 2017 and we build forecasts 6 months ahead (see figure below).

After the forecasts have been generated, Streamline automatically assigns them to the currently selected As of period. At the end of the next month – Aug 2017 – we generate the forecasts again. And so on. Let’s keep doing this until the end of Dec 2017. In the result, we have forecasts for 6 periods starting from Jul 2017 to Dec 2017.

Now, when we have actual data up to Dec 2017, let’s evaluate the forecasts generated as of Jul 2017. To do this, select this period in the As of control. Let’s look at the item Dark Beer in the East location (see figure below).

As you see, Streamline automatically loads and shows all available actual data regardless of the period selected in the As of control (see figure above) and evaluates the forecasts against this data.

Viewing Forecast Quality

The results of the forecast evaluation are represented by the Model MAPE and Forecast MAPE in the Model tab of the Panel (see figure below).

  • Model MAPE is MAPE calculated based on the model response for the data that was used to build the model. This response is indicated with the blue dots in the Plot (see figure above).
  • Forecast MAPE is MAPE of the forecast. It is calculated if there is actual data to evaluate the forecast against.

The Forecast MAPE shown in the figure above is the evaluation of five periods (Aug 2017 - Dec 2017) for which actual data exists.

To see the forecasts evaluation for all of the items in one report, go to the List view tab and select the Forecast error report (see figure below).

The Percentage error section of the report represents APE calculated for each item, location, and period. As we do not have actual data for Jan 2018, the corresponding column is empty in the report.

As you see, the table has gaps and empty rows in our example. A gap occurs when there are no actual sales in a period. Streamline can’t calculate the measure in this case. Empty rows arise if:

  • there is not enough of sales history to build a forecasting model;
  • the item is inactive; or
  • intermittent model is used.

The last column of the report represents MAPE calculated across all of the periods for every planning item.

The Overall WMAPE, shown in the List view toolbar, is a MAPE weighted across all the items in the project. If the products’ prices are given this is a price-weighted MAPE measure that represents overall revenue percentage error. If selling prices are not imported, WMAPE is calculated based on the volume of sales. You can read more about WMAPE on Wikipedia.

2018/04/02 14:55

Forecast Quality Measures

Streamline is able to evaluate forecast quality using two measures:

  • Error (MAPE) or
  • Accuracy (100% - Error).

The default measure is Error. You can switch between the measures by going to the menu File > Settings > Project tab > KPIs group and changing the Measure of accuracy parameter (see figure below).

The figure below shows the Forecast accuracy report when the measure is set to Accuracy in the Settings.

This option also affects the evaluation shown in the Model tab of the Item view (see figure below).


Next: Rolling Forecast

Download PDF

evaluating-forecast-accuracy.txt · Last modified: 2018/06/05 11:02 by admin