non-gravity on 08 November 2012
In easy terms your score is an approximation of the percentage you are off.
In more advanced terms your score is [((Real-Predicted)/Real+(Real-Predicted)/Predicted))/2]*100 then round and take the absolute value. The maximum score has been set at 100.
This table shows some of the scores for under or overpredicting
| -60% | -40% | -20% | -10% | -5% | 0% | 5% | 10% | 25% | 50% | 100% | 150% |
| 100 | 53 | 23 | 11 | 5 | 0 | 5 | 10 | 23 | 42 | 75 | 100 |
Being 5k off a 75k week gets you the same score as being 100k off a 1500k week
As to why this method is used, this may be insightful.
The green curve represents (Real-Predicted)/Predicted). This one would encourage overpredicting.
The red curve represents (Real-Predicted)/Real). This one would encourage underpredicting.
The blue curve is the average of the two which we use. This one tries to find a balance.
