Forecast Metrics

You can use the User Defined Customer Forecast Profile and User Defined Site Forecast Profile tables to populate forecast metrics. The metrics are then converted to demand parameters. The fields that you populate are:

• Forecast Mean – Provide the forecast estimate or forecast mean.
• Forecast Error – Provide the standard deviation of forecast error.

The following information provides additional detail about how to use the forecast metrics.

Measuring forecast performance

There are several measures you can employ to determine how good the forecast data is:

• Mean Absolute Deviation (MAD): Measures average absolute deviation of forecast from actuals.

  

  • MAD measures absolute error; therefore, positive and negative errors do not cancel each other out.
  • You want MAD to be as small as possible.
  • There is no way to know if the MAD error is large or small in relation to the actual data.
• Mean Absolute Percentage Error (MAPE): Measures absolute error as a percentage of the forecast. For a 20% MAPE, enter a value of 0.2.

  

  • MAPE measures deviation as a percentage of the actual data.
  • You want MAPE to be as small as possible.
• Mean Squared Error (MSE): Measures variance of forecast error. It is the preferred forecast error measure.

  

  • MSE measures the squared forecast error - error variance.
  • MSE recognizes that large errors are disproportionately more “expensive” than small errors.
  • MSE is not as easily interpreted (that is, not as intuitive) as MAD and MAPE.
• Mean Forecast Error (MFE): This is also referred to as forecast bias.

  

  • MFE should be as close to zero as possible; that is, you want the minimum bias.
  • A large positive (or negative) MFE means that the forecast is undershooting (or overshooting) the actual observations.
  • MFE of zero does not imply that forecasts are perfect (error free); rather this indicates that the forecasts are “on target”.

The standard deviation of forecast error is defined as:

where:

: mean demand (or forecast mean or forecast estimate at the end of the forecast horizon)

: observed demand at time t

: forecast estimate at time t

: number of observations

Forecast error:

Standard deviation:

• Forecast error (or uncertainty) can be measured by MAD, particularly in time series analysis.
• An empirical rule for MAD: The standard deviation for a normally distributed random variable is theoretically equal to which is approximately 1.25 MAD.
• is true asymptotically; the difference between and approaches zero as the sample size goes to infinity.