| March Madness Forum View the Forum Registry |
| |
| 0 |
Subject: Contest Scoring Simulator Posted by: Guru - [330592710] Fri, Mar 24, 16:33 I wish I'd have done this a couple weeks ago. It would have been a VERY helpful tool in making contest selections. Unfortunately, I didn't think of it until a few days ago. I built a contest scoring simulator. You can fill out a bracket, simply picking the winners in each game. Then, the simulator figures out the contest scores for each team, and sorts them from best to worst. I've done some limited testing, and it seems to work. It does reproduce the actual results to date if you plug in the results of the first 52 games. Let me know if you encounter any anomalous results. Oh well, it will be available in advance of next year's contest. I figured I'd better do it now before I forget. Pretty cool, huh? | ||||||
| 1 | Sludge ID: 59228918 Fri, Mar 24, 19:35 |
Guru, I wrote a program to simulate the tournament itself. I.e., based on some kind of rating system (I used Massey's computer ratings, but any could be used, really), you simulate the tournament play and pick the winners and losers randomly. Based on that, you compute what each team/basket scores. Simulate 1 million tournaments, and you can estimate the average score of each team/basket.
Using these averages, what I did was look at billions of combinations of teams and baskets and tried to find those combinations which gave high (or low - for my tanker team) scores. Basically, I ran out of time (and wife's patience), and had to go with what I had. The above procedure would find a nearly optimal combination under the assumption that your units are independent. Since there were seed and conference baskets, the units are not independent. I.e., it's not a good idea to short the #1 seed and then long all 4 #1 seeds individually. The procedure never gave me anything that absurd, but it did give some results with some overlapping. I wish I would have had more time to modify the procedure to account for/minimize this overlapping. Of course, I kept this under wraps. :) But now that the tourney is nearly done, it can be shared now. Anyone familiar with C? | |||||
| 2 | Sludge ID: 59228918 Fri, Mar 24, 19:57 |
NOTE: "Avg Winnings" is the average winnings if you long the team and DOES NOT include the cost of each team.
| |||||
| 3 | Guru ID: 330592710 Fri, Mar 24, 21:11 |
I also built a monte carlo tournament simulator. I used it extensively last year (in advance of the tourney) as a means to test out alternative scoring formats, and as a guide for pricing. Mine was just a simple Excel macro. One key assumption is assigning winning probabilities to the games. I used different assumptions - historical seed-based probabilities (smoothed), RPI and/or Sagarin ratings-based, and I even tried some dynamic assumptions where I changed a team's rating as it progressed through the tourney. (If a #10 seed wins in round 1, it probably is stronger than a normal #10 seed; if it wins again, it is probably even stronger; etc.) This year, I didn't even bother to dust off that model. I just priced by gut feel. For this type of contest, as long as the prices are reasonable, they tend not to determine the final standings. To win, you need to pick the correct teams, almost regardless of price. | |||||
| |||||||
Post a reply to this message: (But first, how about checking out this sponsor?) | |||||||