|
1. What is MAPE?
2. Should we divide the Error by Actual or Forecast?
Why one over the other?
3. If we divide by Actuals, and when actuals are
zero, MAPE is undefined (excel gives #DIV/zero error). What to
do?
4. If Forecast error is greater than 100%, is accuracy
negative?
5. What is RMSE?
6. What is
weighted forecast error?
7. Why is Mean Percent Error not useful?
1. What is MAPE?
Mape is Mean Absolute Percent Error. This is defined as the Average
Absolute Error divided by the Average of the Actual Quantity.
A more elegant way to compute this would be to divide the Sum
of the Absolute Error by the Sum of the Actual Quantity.
MAPE = (Sum of Abs. Error) / (Sum of Actuals)
2. Should we divide the Error by Actual
or Forecast? Why one over the other?
Most commonly Error is divided by the Actual rather than forecast.
This avoids certain self-induced biases by the forecaster. If
we divide by forecast, then the forecaster has the incentive to
overforecast when in doubt. A higher forecast drives MAPE lower
and accuracy higher.
3. If we divide by Actuals, and when actuals
are zero, MAPE is undefined (excel gives #DIV/zero error). What
to do?
When actuals are zero, MAPE is infinite. By definition, forecast
error can be greater than 100%. However, accuracy cannot be below
zero.
Forecast Accuracy = max (1 - forecast error, 0)
If Actuals are 25 and forecast is 100, then error is 75 implying
a 300% error. But accuracy is always zero for cases where error
is higher than 100%. You can use the ISERROR function in excel
to overcome the DivisionByZero error. Although this is an excel
workaround, this actually preserves the impact of the actual being
zero in calculating the total forecast error. See the excel accuracy
template available in the downloads section!
4. If Forecast error is greater than 100%,
is accuracy negative?
By definition, Accuracy can never be negative. As a rule, forecast
accuracy is always between 0 and 100% with zero implying a very
bad forecast and 100% implying a perfect forecast.
5. What is RMSE?
RMSE stands for Root Mean Squared Error. This is an alternative
to measuring absolute errors. Here you compute the square of the
error and take a square root of the total.
6. What is weighted forecast error?
Weighted forecast error just turns the MAPE into a weighted
MAPE calculation. Each absolute error is weighted either
by price or some other factor of importance. The weighted
MAPE is the sum of all such weighted errors divided by the sum
of the Actual volume similarly weighted. We can calculate
a price weighted MAPE or a discrete-weighted MAPE that ranks
items based on importance.
7. Why is Mean Percent Error not useful?
Mean percent error uses a simple average of computed forecast
errors. The problem with this measure is that it may
weight low volume items disproportionately. You may have a
high forecast error on items that only are shipped a few units
each month. Typically items with such spotty volumes are
the hardest ones to forecast. Such items should be dealt
with using an alternative supply chain strategy instead of
focusing on their demand forecast accuracy. Hence Mean
Percent Error could create incorrect incentives.
If you have other questions, please
contact us.
|