Thanks to Teamrankings for the data for this work. I hope that my models someday forecast as well as yours!
Yesterday, I tried some approaches to predict the NCAA tournament. My favorite of these is where I use team efficiency stats (shot selection, shooting percentage, turnovers per possession, and offensive rebounding percentage) to simulate whole games possession by possession. This approach predicts win-loss correctly in about 70% of tournament games over the past five years, so I decided to predict the whole tournament with this method for this year.
Below are the win probabilities that I generated for each region (the percentages on each line give the probability of winning the previous game).
Kentucky is the odds on favorite to win it all, and I predict that they will get out of their region. Their stiffest challenge could be Indiana, where I only predict a 53.3% chance of victory. Since you want to pick upsets with a chance of carrying through in big NCAA pools, I also show the odds for Indiana to advance if they do beat Kentucky (57% against Baylor). None of the other close underdogs would advance in the next round by my probabilities. The upset picks in this bracket are 12 VCU over 5 Wichita St. (strong probability of 61%) and 3 Baylor over 2 Duke (54.8%).
I didn’t make any adjustment in this region for Syracuse missing their center Fab Melo. Without him, it’s likely that Ohio State’s odds of winning in the elite eight are over 50% (they are quite close at 49.4% already). Syracuse may struggle even earlier with Kansas State, though the real upset predictions in this region are 11 Texas over 6 Cincinnati (53.3%) and over 3 Florida State (50.1%) and 5 Vanderbilt over 4 Wisconsin (very strong at 76%). None of the close underdogs would advance if I pushed them through one round, but I’ll check Ohio State’s odds in the final four below.
The most interesting prediction for me in this bracket is Florida State, who have only a 55.5% chance of beating St. Bonaventure. Many are projecting them as a strong three seed after wins over Duke and North Carolina in the ACC tournament. It’s possible that my estimates don’t give their defense enough credit. I estimate shooting percentage as a function of each team’s regular season shooting percentage and shooting percentage allowed, but Florida State might drag other teams’ percentage down more than usual. If they do play well, this region looks like a real challenge for Syracuse (strong two and three seed).
North Carolina looks to be the weakest of the three one seeds so far. They should get through Alabama/Creighton, but don’t have killer odds against my Michigan Wolverines (58.3%). Kansas has them pegged by a hair (51.3% odds), and even St. Mary’s, a plausible elite eight team, could expect a close game against them (around 45% odds against North Carolina). There are lots of upsets predicted in this corner: 9 Alabama over 8 Creighton (63.2%), 12 South Florida over 5 Temple (51.8%), 11 N.C. State over 6 San Diego State (66.5%), and 2 Kansas over 1 North Carolina (51.3%). If I push St. Mary’s through Kansas, they have pretty good odds against Georgetown as well (55.3%).
From my probabilities, it’s clear right away who the one seed should be in this region. Michigan State was the last one seed awarded, and maybe it should have gone to Missouri. Their win probability over Michigan State is 68.5%! Michigan State might struggle with St. Louis in the second round (59% odds), but benefits from weak four and five seeds (Louisville and New Mexico). I didn’t show it in the picture, but if I push St. Louis by Michigan State, then they have a 50% win probability against New Mexico. Missouri looks like the sure bet here — their toughest game might be Florida in the second round (62.3% odds).
Lots of upset picks again: 9 St. Louis over 8 Memphis (54.3%), 5 New Mexico over 4 Louisville (53.8%), 11 Colorado State over 6 Murray State (55.8%), and of course 2 Missouri over 1 Michigan State (68.5%).
The Final Four:
Even at this stage, Missouri spanks the competition. I estimate that they will beat Kentucky 57.3% of the time in the final four and beat Kansas in the final 60.3% of the time! Indiana would get beat even worse if they advanced. Although Kansas has 53.5% odds against Syracuse, Ohio State could expect to beat Kansas (54.9% odds), if it can get past Syracuse. Ohio State fairs no better in the championship game, however. Of the teams that I tested, only North Carolina comes close against Missouri (48% odds), though North Carolina doesn’t fair well against Syracuse or Ohio State in the final four.
The nuances of these late tournament match ups highlight the value of using multiple efficiency stats as inputs. Since each is a function of inputs from the two teams, different combinations can lead to scenarios where there is no clear “best” team in a set of three or four (i.e., no one beats everyone). North Carolina is a great example. They are very close with Missouri, but don’t handle Ohio State or Syracuse as well, even though Missouri does very well against these two teams. The tournament structure matters in these cases.
When you’re trying to make smart upset picks to win your pool, it’s important to think ahead as well. St. Mary’s is a great pick, for example, because they have strong odds against three likely opponents (Kansas, Georgetown, and North Carolina). Another strategy is to choose upsets like 13 Montana over 4 Wisconsin. The odds are low, but you need to hit some crazy picks to win most pools, and Wisconsin has very low odds of advancing past Vanderbilt. If you’re wrong, you don’ t lose much. If you’re right, you distinguish yourself from the pack.
One caveat on these results: Big 12 teams seem to do very well compared to their seeding. My best guess to explain this is that the Big 12 is having a bit of a down year, but I’m not acknowledging that in my setup. I adjust for strength of schedule by including conference membership in the efficiency stat estimation functions, but I lump all the big conferences together and all the mid-sized conferences together. This probably overrates the Big 12 this year. However, the same should be true of the Pac 12, which had an even worse year, yet Colorado doesn’t see much of a bump (I would have predicted Cal to upset Temple, had they beaten South Florida in the play in game, but advance no farther). We’ll see how the predictions go!