Tag Archives: Grantland

Buffett’s Billion Dollar Bracket Bet

Lightning photo courtesy of cnx.org.

The NCAA tournament has become a cultural phenomenon where everyone suddenly becomes a college basketball expert whether or not we’ve ever watched a game. This expertise has raised to a fever pitch this year as Quicken Loans has offered 1 billion dollars to anyone who provides a perfect bracket. But why?

This contest, underwritten by Warren Buffett’s Berkshire Hathaway, is often described as a one-in-9 quintillion change of winning, or 2^63rd power based on there being 63 games played by 64 teams in this single-elimination tournament. However, this model is obviously wrong since we know that some of these teams are better than others. Even the most casual NCAA bracket filler knows that a 1-seed (presumably one of the top 4 teams in the country) always beats a 16-seed (which is typically one of the worst 4 teams in the tournament). Similarly, a 2-seed almost always defeats a 15-seed with rare exceptions. After this point, the expectations start get a little trickier and the March Madness descends into full effect.

Even so, we know that top seeds tend to be safe well into the second week of the tournament. So, given that the NCAA games are not a pure coin toss, what are the real odds of filling out a bracket perfectly?

There are a few ways to go about this estimate. One is to go through historical data and look at how the NCAA bracket has carried out over time, then use the odds that each seed moves to the next round as the basis of a win expectancy at each round. This is the approach that the quant in me wants to take and it is definitely the most tempting way to go. However, a couple of basic problems keep me from taking that approach.

First, this approach lacks the specific context of how college basketball exists in 2014. Comparing NCAA tournament results from 50 years ago, when freshmen weren’t allowed to play college basketball and John Wooden’s UCLA Bruins owned the tournament to today’s world where “one-and-done” freshmen are often the best players on their team and talent is more distributed throughout the country seems to be an unfair comparison. Also, seeds are determined very differently, with power rankings, automatic berths, and decision makers changing on a year-to-year basis. The tournament today is very different today than it was even 10 years ago. In addition, we know that there are specific biases in seeding that seem to be errors, such as Louisville’s seeding as a 4-seed even as many experts believe that they are a top-4 team. These individualized biases make a deep longitudinal study an interesting historical exercise, but not necessarily the best predictive model.

Second, and more importantly, I’m on a plane right now and don’t have access to the numbers.

So, in the lack of true quantitative evidence, I created a simple qualitative model where I estimated the ability to choose the correct winner of each game. We can assume that almost everybody will bet on the 1 seeds to win the first round and be correct, so this initial assumption changes the odds from 1 in 9 quintillion to about 1 in 600 quadrillion. (Quicken Loans may as well just hand the money away right now…)

In my model, 1-16 game was a 99% chance of choosing the winner, a 4-13 game was treated as an 80% change of winning, while an 8-9 game or a semifinal or final game was treated as a 55% chance of picking the winner, since there’s almost always some level of information that shows that there is a favorite. Basically, as the talent differential gets smaller, the picks are increasingly due to chance. Put that together and the odds start changing significantly.

When I did this back-of-the-envelope calculation, I came up with much lower odds of 1 in 298 billion to get a perfect bracket. These odds are still astronomically high, but start to get closer to the real number. I’m sure that Warren Buffett went through a similar process, discounted this number significantly, then provided his insurance policy accordingly.

But this model has a flaw (now that I’m back on the ground in Orlando). It assumes that you will always choose the higher seed, whereas this isn’t always the case. Ed Feng of Grantland and Stanford identified a much better model that takes into account both the potential outcome that either the higher or lower seed would win. With his model, the odds are 1 in 4.5 billion. That’s still a really high number. In contrast, your odds of being killed by lightning are 1 in 280,000 (http://www.ehow.com/info_8607019_chances-being-hit-lightning.html). Yes, you’re 16,000 times more likely to be killed by lightning than to win.

So, how much would this policy for a billion dollars actually be? If you’ve got 15 million players for the billion dollar bracket, you’d expect a 1 in 300 chance to win. So, the back of an envelope approach says that you could expect to come out ahead by providing an insurance policy for a bit over $3 million.

You can start to play with the numbers a bit more to be more conservative, but it’s hard to price the maximum breakeven at much more than $5 million based on reasonable assumptions. A gut feeling says that Berkshire Hathaway probably felt comfortable with charging $5-$10 million as a policy for Quicken Loans. Based on Buffett’s philanthropic nature, let’s call it $5 million.

Now, for the next part. Does this make sense for Quicken Loans? According to Dan Gilbert, Quicken expects to get 15 million new leads from this process, meaning 15 million new potential customers for mortgages and other loans. Again, playing the back of the envelope card, assume that Quicken gets an average loan of $150,000 from each closed deal. Based on a 30 year loan, 6% interest and 3% inflation, Quicken gets about $98,000 in discounted interest. Add closing costs and let’s just say Quicken makes $100,000 per loan.

So, to break even on the insurance policy, all Quicken really needs to do is find 50 mortgages out of all of this. Can Quicken Loans convert 1 in every 300,000 qualified contacts into a mortgage? Probably so, based on their brand name and the assumption that their sales force knows how to qualify and close interested parties.

But in this context, all of the marketing around the billion dollar bracket suddenly makes more sense. Even if you include the marketing costs and all of the other effort that Quicken is putting into this, the end result is that they are getting millions of people’s verified contact information for what ends up being a small fraction of their potential value.

Now that you know the real numbers behind the Quicken bracket, the story changes considerably. The real story isn’t “Can you win a billion dollars based on a 1 in 9 quintillion chance of winning?” The real question is “Can Quicken Loans get, say, 250 new mortgages out of their marketing campaign to justify the marketing and insurance efforts they’ve put in place?”

And at the end of the day, everybody wins. Warren Buffett makes another 5 million dollars. Quicken Loans probably makes 50 million dollars. Yahoo gets its marketing money. And we all continue to get an online platform that helps us to continue our crack-like addiction with March Madness. Nobody loses. (Unless you’re a competing mortgage provider.)

What Big Data can learn from the NBA

94Fifty smart sensor basketball photo courtesy 94Fifty.

On Thursday, February 13, Grantland’s Zach Lowe wrote an article on the latest technological development in professional basketball: measuring biometric information in game settings. Four D-League (the NBA’s developmental league) teams will start using one ounce sensors fitted on player jerseys to start measuring metrics such as heart rate speed, and position. These sensors are currently available from one of three companies: STAT Sport, Zephyr, and Catapult. These sensors are not new to professional basketball, as nearly two-dozen NBA teams already use these devices. However, NBA teams currently only use these sensors in practice settings, rather than in game-time situations.

There are a couple of interesting Social Big Data lessons that professional basketball could potentially learn from this experiment that every Big Data expert should be interested in finding out.

First, consider one of the quotes from the Grantland article:

“As the research-and-development arm of the NBA, the NBA D-League is the perfect place to unveil innovative performance analytic devices in-game,” said NBA D-League president Dan Reed.

This concept of an R&D product where you collect more data in an experimental setting is one that many technology companies could start to use. For instance, does your core cash cow product have a corresponding R&D product that can be tinkered with without affecting your revenue? This is a good role for your freemium or single user product. (Heck, Facebook does this for their core platform, although DataHive does not recommend the level of iteration that Facebook provides unless you have a monopoly or duopoly of your core market.) This new use of heart rate and other physical information will provide insights on team tactics and performance if used correctly, thus leading to not just Big Data, but interactive and social Big Data where each player’s metrics are dependent on each other.

Second, and more interestingly from a tactical perspective, this measurement will allow basketball teams to more closely align physical effort with results. It is easy to simply believe that hustle and effort lead to better results, but these metrics may actually show that a lack of hustle could be one of several things. It could be a health issue or laziness or it could be good strategy in saving energy for key moments. Hustle and physical movement should not be measured in isolation, but in context of results. If a “clutch” player ends up moving less than an average player or saves exertion for peak moments, the economics of movement may actually state that excessive “hustle” is detrimental to performance. These sensors may also show that specific team tactics lead to greater efficiency, just as our analysis of shot taking shows how important it currently is to take shots from within 4 feet of the rim or on the sides of the three point line.

From a business perspective, most of us do not put out the physical effort of a professional athlete for a prolonged basis at work. But do we waste time and energy by going in the wrong direction? Are we getting stressed because our managers are not telling us the right information? There is a key challenge of understanding how to use this information productively rather than punitively. It can be easy to fall into the trap of simply stating that more time at work equates to greater productivity, but it may actually be that after a certain point, the error rate or lack of clear thinking outweighs the incremental productivity that would be expected. Follow the real business metrics rather than pure resource utilization.

However, as this occurs, one of the biggest challenges will be to translate sports analytics to business analytics. Keeping score is very easy in a rule-based sports environment, but more difficult in a business environment when it can often be difficult to define KPIs. Based on personal experience and interviews with multiple basketball analysts, DataHive has found that the academics and number crunchers conducting this analysis are largely unaware of the value that these findings could provide in the sports world. Although business analysts can quickly see how the heat and activity maps associated with basketball could translate into greater retail, field, and manufacturing success, one of the great challenges is that the sports analysts currently doing this work do not understand how their work could be translated to other fields. In our role of supporting Social Big Data for Human Insight, DataHive serves as a Sports Data Whisperer that wrangles the findings and techniques used in the sports world and brings them to the business world.

DataHive’s principals have long believed that the structured world of sports serves as a natural testing ground for the predictive, geolocated, and biometric data that is being introduced to the corporate world. Video feed metadata and sensor-based data are the Next Big Things in Big Data and it is only a matter of time before the corporate world follows suit. Regardless of your personal interest in sports, Big Data professionals should keep track of the surveillance and sensor data being used in the basketball world to see how this controlled setting provides potential insight for future enterprise technology efforts.