© 2000 John Petroff
|
|
© 2000 John Petroff |
There are several other methods for isolating a seasonal or
cyclical pattern other than the moving average just outlined:
- single exponential smoothing
- quadratic exponential smoothing
- percent change
- moving percent change
- weighted average
The choice of the method is based on the mean square error, a visual inspection of the data and habits of the analyst. Calculation of mean square error would be the recommended test of goodness of fit, but that implies that the different techniques have to be tried one after the other, which is clearly a lot of work. An even more rigorous test of the procedures would be to compare forecast errors beyond sample data. This can obviously only be done when conducting such analysis over a long period of time, month after month, or even year after year. Clearly, the technique is not applicable to screening investment opportunities. It is most appropriate for tracking some important performance, and for making quick forecasts for a few months ahead, such as for sales revenue of a company or division. It is indeed primarily used by inside analysts.
See review question Q-5F6.1.
See research assignment R-5.9 and R-5.10.
| Previous: Decomposition |
|
Next: Linear programming |