# Better NCAA Brackets Through the Power Rank

Photograph by David E. Klutho/Sports Illustrated via Getty Images

The NCAA men’s tournament final in Atlanta on April 8, 2013

It’s amazing (and maybe a little alarming) how much brainpower is devoted to filling out NCAA men’s basketball tournament brackets. I recently looked at how data scientists are developing predictive analytics to choose winning college teams as March Madness gets under way.

Then there’s Ed Feng, creator of the Power Rank. His product is a fresh mathematical approach, ranking teams against one another and then converting those rankings into accurate win probabilities. Feng has proven quant skills—a Ph.D. in chemical engineering from Stanford University, with research in statistical physics. His approach features math that’s similar to the PageRank algorithm made famous by Google (GOOG).

Feng applies his ranking system to football, basketball, hockey, baseball, and soccer. For college basketball he ranks all 351 teams from top to bottom, and the calculations are based on teams’ results: who they beat, who they lost to, and by how much. Each team’s opponents are considered with the same calculation, creating an entire network of all teams playing against each other.

The numerical score for each team is its predicted margin of victory against an average team on a neutral floor. That makes each number simple to use—you just subtract, and the difference between the two teams is the expected margin of victory. For example, Feng’s top-ranked team is the Arizona Wildcats, with a 17.96 score. He has Oregon at No. 21, with a 12.4 score. If those two teams played each other, he’d give the edge to Arizona, with an expected margin of victory of 5.56 points.

This expected margin of victory can then be converted into a win probability. For example, a one-point spread equates to a 53 percent win probability. Bigger point spreads equate to higher win probabilities: A five-point spread is 65 percent, a 10-point spread 77 percent. So that 5.56-point difference in our Arizona-Oregon game would translate to a win probability of more than 65 percent for Arizona.

Based on preliminary brackets, Feng’s site is already giving probabilistic predictions for every tournament game. His simplest advice for winning your bracket: “Don’t get into big pools.” If you’re competing in a pool with fewer than 10 people, Feng advises that you go with the simple strategy of taking the higher-probability team every time. He says pools of 10 to 50 contestants still give you a decent shot of winning, but pools with more than 100 participants are “too random” to offer much hope.

In his predictions, Feng ignores regional seeds or poll rankings and relies on his Power Rank approach. He suggests your best bet for picking an eventual winner is to find a team that has a high likelihood of winning but that not everybody is picking: an “undervalued champion.” This year he thinks that team is Louisville, and he gives the Cardinals a 7 percent chance of winning it all. He points out that the team with the highest likelihood of winning tends to be overvalued by the public, hence his approach to go slightly contrarian.

Because an eventual champion needs to win six straight games, Feng is trying to find teams that show consistency. One of the factors he considers is the variance of efficiency: For example, Davidson College has the most volatile team in terms of consistency, so it was perhaps unsurprising that it lost in its conference tournament. Davidson is probably too risky to pick as a team poised for a deep run.

Another factor Feng looks at is three-point shooting. He thinks a team that’s been on a 10- to 12-game hot streak, shooting 3 percent to 4 percent above season average from three-point range, is due for some reversion to the mean. He advises betting against this type of team early, and thus fully expects Nebraska to lose in the first round.

For all that Feng does know, there’s still plenty he hasn’t been able to figure out yet. For instance, he wonders whether it’s possible to ever master “quantifying team chemistry.” He believes strongly that numbers alone won’t give you all the answers, because luck and the human element will still come through when making bets.

As far as Feng’s own personal brackets go, he says his friends are too smart to let him enter their pools.

LIMITED-TIME OFFER SUBSCRIBE NOW

### Companies Mentioned

• ##### GOOG
• \$533.0 USD
• 4.23
• 0.79%
Market data is delayed at least 15 minutes.

### Post a Job Lifestyle related Jobs

View all Lifestyle related jobs