Ten Week Countdown 2009!

[TWRoO]

koffieboon said:

TWRoO said:
@Ken... it is very true.... under predicting is a lot better than over predicting, as proved by the PS3 last year.

If you predict 200k and the real figure turns out to be 100k, you are 100% out, but if the real figure ends up being 300k, you were still 100k out, but in percentage it is only 33%.

At least that is how it feels in the mind... in reality as the real figure is pretty much set (ie, we don't know it yet, and whatever we predict here will not effect what the real figure is unless one of us goes and buys a few thousand consoles) and if you predict 50k above or below it is will be the same percentage difference.... it's just that because we are predicting before the real figures (wel duh, no point predicting after) we then compare the real to our predicted in our mind, rather than comparing our predictions to the real.

So basically you just need to compare the predictions to the actual results and not the other way around. At least in my opinion being 50k over or under should be considered as an equally right (or wrong) prediction. Talking about scoring, how will these consults be scored and ranked?

About important dates: PSP Go releases in Japan on November 1, which means 1 day in week 1 I think. (I'll need to edit my own predictions because of it as well) Black Wii releases in the UK at November 6 and in the rest of Europe November 20. (which probably means a missive week 4 for Wii in Others seeing it is combined with the release of NSMB)

Yes you are correct, I compare the predictions to the results. So if you are 50k above, it's the same innaccuracy as 50k below, but it is still better to under-predict... last year there were 4-5 predictions that were over 100% wrong, which is only possible with over-predicting.

I suppose really there are two different methods, and neither is fair to both under and over-predicting.

The way I do it now: If A predicts 200k, and B predicts 50k, the sales turn out to be 100k.... this results in A getting 100 points, and B getting 50 points.... but you could argue they were just as innaccurate because A predicted double, and B predicted half (each bein 1/ the other, so is their opposite) After all, it is not too difficult to over-predict by double, but no-one is going to predict 0 sales (which would be the only way to get the same points as someone who predicted double if using the current method)

So the way I do it, where equal figure discrepancy = equal score... favours under-predicting.
The second way, where equal fractional discrepancy = equal score... favours over-predicting.

-----

Lets see what happens if we do both and divide by 2:

Scenario i) A predicts 50k, B predicts 150k.... real figure is 100k.

by method 1.... A and B both get 50 points.
by method 2.... A gets 100 points, B gets 33 points

the mean of the two methods gives.... A: 75 points and B: 42 points

Scenario ii) A predicts 50k, B predicts 200k.... real figure is 100k.

by method 1.... A gets 50 points, B gets 100 points
by method 2.... A and B both get 100 points

the mean of the two methods gives.... A: 75 points and B: 100 points

----

So when combining both methods, it shows more that B was predicting better in scenario i) (even though they re both 50k off.... you only need to X the real figure by 1.5 to get to Bs prediction, but you need to / by 2 to get to A)
Similary it doesn swing in the other direction in scenario ii)