ACCBoards

Apparently, no one can predict the NC double A tournament.

Quote:My bracket is a disaster: I seem to have an uncanny talent for picking all the wrong upsets. But perhaps I should find solace in the fact that the NCAA tournament is inherently unpredictable. That, at least, was the conclusion of a 2001 paper by the economists Edward Kaplan and Stanley Garstka. They mined every statistical tool they could think of in an attempt to crack the office pool. They searched for secret algorithms in past NCAA tournaments, and used Markov models to see if regular season performance affected post-season performance. They ran endless computer simulations, and plugged in a vast trove of player data.

The end result? They achieved "overall prediction accuracies of about 58 percent, but did not surpass the simple strategy of picking the seeds when the goal is to pick as many game winners as possible." In other words, all the fancy mathematical equations were utterly useless: the tournament remained a mystery, a stochastic process of indeterminate outcome.

There are some grad students at GT that used game theory to make much better predictions than 58%. I'll see if I can dig up the article. I haven't seen it circulate this year. So they may have moved on to other studies.

I downloaded the Kaplan and Garstka paper, but haven't had time to look at it.
Not sure I would understand it anyway.

Certain things you can gauge (and I forget every year - I should write this stuff down). For one, in a scoring format like Yahoo's, always pick all the 9 seeds in the first round. I don't like that choosing a 9 over an 8 gives the same score as picking a 13 over a 4, but that is how it is.

I was in a league once where your upset points were based on the seeding difference (picking an 9 over an 8 gets you 1 point, and picking a 13 over a 4 gives you 9 points). Unfortunately that system is easy to game. Pick all the lower seeds except for the 1's and 2's and you get so many upset points that it swamps all the games you missed.

Finally, never predict anyone (ahem) to go past the 2nd round unless they have a genuine big man on their team. 4 wings and 4 guards isn't going to get you very far regardless of their recruiting rankings...

I understand just enough of it to be confused.

That is exactly the reason the BCS formulas fail. A good predictor can be created, but once its understood coaches will start to manipulate it. (i.e. margin of victory)

accball1 Wrote:I was in a league once where your upset points were based on the seeding difference (picking an 9 over an 8 gets you 1 point, and picking a 13 over a 4 gives you 9 points). Unfortunately that system is easy to game. Pick all the lower seeds except for the 1's and 2's and you get so many upset points that it swamps all the games you missed.

^I like those the best, but wasn't aware of the "gaming" angle.. only office pool i ever won, so nobody else knew it either. If you double points per game per round and add the seed difference (not doubled), will that not "protect" the method? (e.g., 1,2,4,8,16,32 plus seed difference)

Interesting thought kidc, I have never really gone through the numbers to see what might make it work - the only thought I had was maybe limiting the number of upsets you can pick, like 1/4 of the games (first round you get 8, second round you get 4, etc) or something like that - but that is hardly a good solution.

I used to run a power ranking program that I could calibrate to account for end-of-the-season play going into the tournament. It worked pretty well on predicting the overall outcome (it was better than just picking based on the seeds) but it wasn't smart enough to pick the true upsets.