Rado -

Some quick thoughts: If the difference in cash flow (the outcome from a
future decision based on the forecast) is proportional to the "error" (where
error is the difference between actual and forecast), use MAE. If the
difference in cash flow is proportional to the square of the error, use MSE.
In this sense, MAE is a V-shape, and MSE is a U-shape. Both are nonlinear;
MAE is piecewise-linear. In practice, you may obtain very similar results
for MAE, MSE, RMSE or MAPE. It may be more important to "look at the data"
first using a time-sequence plot to detect seasonality and outliers before
selecting a forecasting method.

- Mike
www.mikemiddleton.com

"RL" wrote in message
...
Hi!

I am modeling forecasting functions in MS Excel. In case of weighted
average, exponential smoothing and Holt Winter's methods one has to
optimise the coefficients alpha, beta or gamma using any error
calculations such as MAE, MSE, RMSE or MAPE.

Has anybody give a thought about mathematical background of such a
optimization with Solver such as:

- which is better to optimize MAE or MSE?
- in the first case we are talking about linear optimization in
the second non linear?
- Anybody draw error function for MAE or MSE and tried to
describe solver optimization function in Coordinate System for both
functions?

I am also open for cooperation with anyone who is involved in
business forecasting training!!!

Regards

Rado