Improving forecast quality in practice
19 September 2014
By Robert Fildes
Some academics in forecasting seem to think that all that is needed to improve forecasting is a better algorithm, or perhaps, more ambitiously, better data. Presenting on this topic at the recent Forecasting Centre workshop provoked me to examine what research was out there that got beyond the conventional wisdom as to how to improve accuracy.
The first question to raise is whether accuracy is the primary key performance indicator for a forecaster. The answer is YES, it remains by far the most important objective (as various surveys show). Years ago, I’d interviewed most of the forecasters operating in the business units of a major multinational. They identified the following activities as their priorities:
|
Activity |
Respondents scoring important |
|
Developing consistent data |
83% |
|
Increased software support |
70% |
|
Improved techniques |
66% |
|
Improved data bases |
61% |
|
Improved communication with users |
35% |
The areas of possible improvement fall into four categories: Organization/Information Systems Resources, Forecasting Techniques, and their evaluation. To see how priorities have changed, we conducted a survey through the Forecasting Centre’s Network. Download and read our full report