The Predictive Power of Nonsense

The 2012 US Presidential race comes to a close today (thankfully), and there is no shortage of wacky indicators predicting the winner:

The 2012 US Presidential race comes to a close today (thankfully), and there is no shortage of wacky indicators predicting the winner:

• Iowa Electronic Markets
• FiveThirtyEight
• PollyVote
• University of Colorado

In primitive times a diviner could foretell the future by poisoning a chicken — whether it lived or died provided the answer. Today we have Halloween masks, coffee cups, cookie sales, and winners of sporting events all being credited with forecasting ability. But if you search hard enough (and it won’t even be that hard), you can always find perfect associations in history that have no predictive power.

For example, in time series forecasting, we can always find a model that fits the history perfectly (with a polynomial of sufficiently high order). But will this model be able to forecast the future perfectly? Or will it even be an appropriate model, generating forecasts that are at least in the ballpark and not too awful?

Here is my prediction: Out of all the organizations and people making predictions, somebody will get it just right, and be lauded as a genius for the next four years.

This evening, as we sit on our sofas eating popcorn and watching the election returns, please let us reflect on the classic bad forecasting practices that have been on display. And let us be thankful we didn’t have to kill any chickens in the process.