Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Gold/Mining/Energy : Big Dog's Boom Boom Room -- Ignore unavailable to you. Want to Upgrade?

To: Dennis Roth who wrote (105278)	7/17/2008 5:59:10 PM
From: Paul Senior	Respond to of 206353

Dennis Roth. This is how maybe you do it: For each prediction point and corresponding actual result, you calculate the difference. Then you calculate the standard deviation of those differences. This gives the average difference and the spread of differences. There are data points going back to 2003 you say, so there might be enough points to get an okay estimate of the variation between predicted and actual. I'd use the normal two standard deviation calculation. This would tell me by the way I look at things if there's a "massive" difference between any predictive point and its actual result -- i.e. if the difference were greater than my arbitrarily chosen 2 standard deviation amount. For me, as well as the other poster who responded, I want to look at historical results (the distribution of the differences). I'd never want to make a judgment call without that based on just one point estimate, i.e. would never want to say 104 actual is huge compared to 85,without having some knowledge of how that difference has been in past. (I understand something like 104 could be very surprising and large by itself; I am responding to what I've inferred from the original post-- that the difference was surprising to a large extent because it differed so much from the predicted number.) Next: Doing this --- calculating the average difference and spread of differences (standard deviation) for the three different prediction models will result in one of the three having the smallest standard deviation. That would be the one where the predictions are closest to actual for each point. One then makes an adjustment or leap-of-faith to move forward to forecast (i.e. go from analyzing to predicting) that that model will in the short run be the one which will give the "best" predictions. All just my after-hours opinion.