I have been recently studying Production and Operations Analysis at Stanford University. Also, since forecasting was the first module implemented at my workplace. I would like to share some key insights gleaned from the class and the implementation of the system.
The description below will give you a 50000 ft view of the various business topics related to forecasts that you need to consider before implementing forecasting module (of any supply chain system).
(1) Forecasts are always going to be wrong (are you surpised ?)
(2) Forecasting aggregate units is generally easier than forecasting individual units, so spend your time wisely.
(3) Forecasts made further out into the future are less accurate. (common sense, duh)
(4) Forecasting technique should not be used to the exclusion of known (means human are still needed in addition to machines)
(5) A good forecast also gives some measure of error (see below for various metrics used)
(6) Better forecasts will result in lower inventory costs (for same service levels), so why wouldn't you want to improve forecasting accuracy.
Forecasts can be subjective or objective. Subjective forecasts are developed using customer surveys, sales force composites and Delphi methods etc.
Objective forecasting is generally done using Causal, Time series based methods. One needs to considers trends, seasonalities, cycles & randomness in the time-series forecasts.
Following metrics are generally used to evaluate forecasts:
(1) MAD: Mean absolute deviation
(2) MSE: Mean Squared Error
(3) MAPE: Mean Absolute Percent Error
There are two widely used methods for forecasting stationary series:
(1) Moving averages
(2) Exponential Smoothing
So, following questions naturally come to mind:
- At what level of the product hierarchy should you evaluate forecasts and for what horizon?
- What is the recommended statistical forecasting model?
- What is the big deal about S&OP process?
To be continued...
1 month ago
Posts
