A lot of people have been stopping me in the halls asking, Is there any way we could somehow use a Markov-Chain model to help with March Madness tournament picks?
Well, folks, it’s your lucky day.
Paul Kvam and Joel Sokol out of Georgia Tech published a piece in Naval Research Logistics a few years back explaining a simple three-parameter model. (In the event that the Mudd doesn’t have an old copy laying around, you can read it here).
For those of you without time to crunch the numbers yourself, you might check out the handy LRMC Information Page. Last year I went with the “Pure” LRMC, but this year I’m feeling a little Bayesian.
The authors claim — and there appears to be something to their claim — that their model does a better job than the experts and competing methods (e.g., seeds, RPI, rankings). And here’s a shocker — they have Kansas over Duke in the final (like it took a genius to come up with that). On the other hand, they have BYU as their #4 team, so I will bet the farm on that one watch that game with interest.
