March Madness
It all began in 2017 in the basketball epicenter town of Chapel Hill. Spring break and the lack of an income had stranded me on UNC’s ghost town campus while the rest of my cohort left for the beaches. A plane ticket home costed around $400. If you adjust for inflation, for a working class adult, that’s $5000 - obviously a bit pricey. I offered to give up my seat on the overcrowded plane home in exchange for sizable airline credit, but the desk attendant said, “You can’t give up your seat if you don’t have one… How’d you get past security?”
With nothing to do and ample free time, I decided to gather data. When you’re a stats major, it’s what you do. You see some numbers on the side of the road and you stop to pick them up. Your shotgun passenger looks over and mouths, “No Jackson. It’s a large crowd. They look a tad smutty.” You brush aside their comments. This particular time, the numbers hailed from basketball land and even though peaches are best harvested in the summer, these basket stuffing numbers came from March. I blinked once and slept five nights then woke up with 12 harvests of data. With the data I built an algorithm to predict March Madness.
***
One of the things that disgusts me most in the world is chalk. Get that filth outta your bracket. If you don’t have at least one #12 seed beating a #5 seed, shame on you. If you have three or more #1 seeds in your final four, shame on you. That’s like predicting the Bucs, Bills, Packers, and Chiefs will all make the playoffs next NFl season; yeah probably right, but it’s not bold, it’s rather lazy.
In 12 days we enter the most magical time of the year: March Madness. And, I’m going to predict a perfect bracket. It isn’t possible (the same odds of being hit by lighting not once, not twice, but five million times) but just because it isn’t possible doesn’t mean I’m not going to do it (Hit me; I’m ready).
There exist two approaches to the big dance: 1) Attempting to win your office pool, and 2) Attempting to predict a perfect bracket. To the untrained eye, the approaches may look the same. But, one’s a chicken and the other’s a tyrannosaurus. The DNA’s the same, but the end result is dangerously different. If you attempt the chicken approach to win your office pool, you’ll have a lot of “chalk”, essentially a slew of unrisky picks. Whereas if you attempt the tyrannosaurus route, you’ll either be horribly right or horribly wrong. In the history of the tournament, no one has taken on the T-Rex and won.
Nevertheless, watch me whip. Here’s how I’m going to take down the beast:
***
1) Making sure my bracket has the optimal amount of madness. I’ve quantified madness. That’d be funny if I wasn’t serious. My algorithm defines madness as any negative difference in ESPN’s Basketball Power Index (BPI) metric. For example, if UNC (BPI = 10.7) beat CU (BPI = 13.3) then it would be 2.6 points of madness. I snagged the last 12 years of results and created madness distributions for every round of the tournament. It’d be embarrassing to admit that I used the empirical distribution, but only if you, the reader, knew that was the lazy man’s distribution. To emphasize, I’ve mapped an expected amount of madness (or upsets) for every round along with an acceptable range of variation. Other than 2018 when a #16 seed beat a #1 seed, I’m never surprised by the results. And, after re-calibrating my model/feelings, I switched to being unsurprised.
2) Generating multiple brackets. In 2019 (the last year where the tournament wasn’t cancelled) I filled out 50 brackets. My highest-achieving bracket finished 7th in a large Yahoo pool. Admittedly, filling out multiple brackets is dishonorable - but only so - if you chose the approach to win your office pool. Otherwise the extra amo helps in the hunt to kill the dino. My 953 KB Excel model randomly generates brackets. It confirms that each bracket has the optimal level of madness, plus a specified level of variation from 61 user-input switches. I hooked up VBA code to spit out thousands of scenarios. Backtested on 2019 and 2017 data, given 10,000 tries, my model can, at best, predict 75% of the bracket correctly. Still impressive (he blue-ribbon pats himself on the back), it’s not a near miss, not a failure gone wrong, not perfect. I want to be perfect.
***
Even when I fail, it’s still fun, still a win. My favorite part of the March Madness is the research/prediction portion. It’s something I actually have control over.
One of my favorite quotes:
“For someone to tell me I can’t win them all… that’s loser mentality. I play the game. And I play to win.” - superbowl aspirational quarterback Cam Newton
