Read part one: HERE. TL;DR for part one: For the last several years, I've worked on building a bracket optimization model for March Madness designed to maximize my odds of winning pool(s). Part one gets into the nuts and bolts of how it exactly works, but essentially it consists of three main components:

• (1) an ensemble model of Kenpom, Torvik, and BPI rating systems to derive win probabilities for every possible matchup and examine percentage chances for every team to advance to each round of the tournament;
• (2) public selection data from ESPN used to simulate competing bracket entries across different size pools and identify misalignment in value; and
• (3) an input page whereby I create my own bracket and calculate the probability that such bracket wins a pool, measuring it relative to tens of thousands of simulations of tournament results and bracket entries I would compete against in pool sizes of 10, 25, 50, 80, and 100 entries. 

This is in pretty big contrast to Seth's tool, which gets into the nitty gritty for injuries, player stats, roster construction, team stylistic stats, recent performance, etc. Rather, my model is a little more centered on game theory (i.e. how do I stack the odds most in my favor to win a simple betting game like a March Madness bracket pool).

Happy March Madness. I'm eager to once again make my bracket optimization model available after receiving a lot of positive feedback last year. I've won money from the past three tournaments using this model for my pools, and every year I look forward to making updates to make it even more powerful. You can read my post from last year to get the crash course on how it works, but for this post I mostly want to highlight the yearly updates I've made. The major things I wanted to improve upon were:

• Have the model be more intuitive and the interface be better visually, specifically the Bracket Entry tab where the user enters their specific selections.
• Have the model run faster in order to accommodate a greater number of competing bracket entry simulations. Last year the model supported 100,000 competing bracket entry simulations, while this year's edition supports double the amount at 200,000. 
• Add more functionality to improve decision-making, specifically the new Pools tab.

The main tab you'll work in is Bracket Entry, where same as last year, you make your own selections through a series of dropdown lists. By doing so, you can observe how your odds of winning a pool change across different pool sizes, summarized in the table at the top of the page. In last year's edition of the model, this tab felt a little clunky so I made a number of edits to enhance user experience (including custom error messages when a selection needs to be updated).

​

In terms of new functionality, the major addition to this year's model is the Pools tab, which allows you to record specific brackets you've submitted in the Bracket Entry tab. Just click the Record this bracket entry in the "Pools" tab button as it's shown in the picture above. Following that, you can enter: (1) your wager amount, and (2) the pool size in order to derive specific win probabilities and expected payouts across multiple bracket submissions. If a singular bracket submission is like a "stock", the Pools tab acts like a "portfolio manager" by measuring expected returns and ensuring your odds are maximized and your Champion/Final 4/etc. are well diversified.

​

If you are interested in using this tool, please leave your email in the comment section (or send me one directly) and I will be happy to share via [email protected]. FYI - It is a large file, and you will need to be a little Excel savvy in order to use it. But I do include directions and a glossary to help you familiarize yourself with how it works.

Best luck with your pickings! Early indications suggest teams like Houston, UCLA, Tennessee, and UConn are some of the most favorable top-seeded picks to advance deeper into the tournament. While teams like Alabama (heavy public favorite) and Kansas (not rated as highly by Kenpom/Torvik/BPI relative to public's perception) look a little less attractive. I'll leave the rest of the picks to you!

- mgoDAB

Also, here's a data dump for fun.

​

E[PAP] is expected points against the public. I elaborate on this statistic in last year's post and explain its usage in my model. Essentially, it represents the expected incremental point value gained above the national average bracket entry by making a certain selection. I've found this to be the best statistic when identifying strategic picks designed to maximize my odds of winning pools.

​

​