Embracing Imperfection

• Published date: 01 May 2016
• Date: 01 May 2016
• DOI: http://doi.org/10.1002/jcaf.22162

67

© 2016 Wiley Periodicals, Inc.

Published online in Wiley Online Library (wileyonlinelibrary.com).

DOI 10.1002/jcaf.22162

C

o

m

m

e

n

t

a

r

y

Embracing Imperfection

Tim Chartier

Every March, a large

amount of national atten-

tion in the United States

turns to NCAA Division I

men’s basketball. It hits its

frenzy on a Sunday, in fact,

Selection Sunday when the

initial pairings of teams are

announced for the March

Madness tournament. This

single elimination tournament

takes 68 teams from approxi-

mately 350 schools that play

men’s basketball at the high-

est level. After three weeks

of play, a national champion

is crowned. By the Thursday

after Selection Sunday, fans

have made their predictions

for who will win every game in

the tournament, from the first

round to the game that decides

the national champion. These

predictions create a bracket

and must be complete by the

tipoff of the first game on that

Thursday.

Friends, colleagues, and

even strangers complete brack-

ets and in doing so compete for

pride, office pool winnings, or

even thousands of dollars. In

2014, Warren Buffett insured

a billion-dollar prize for any-

one who could create a perfect

bracket for the tournament. By

the end of the second round (or

48 games), none of the over

8 million brackets had perfectly

predicted the 2014 tournament.

Using analytics to create

successful brackets is an active

area of my research. If we con-

sider only one year of data, my

work sifts through about 5,000

games involving 350 teams.

Analytics enables us to find

patterns and trends that might

otherwise be overlooked or

seem insignificant.

With the billion- dollar

prize looming in 2014, I

received considerable media

attention. In the ESPN online

bracket pool, my methods out-

performed over 97% of over 4

million brackets in 2009 and

over 99.9% of over 5 million in

2010. When the press and pub-

lic learned of our success, they

wanted to know more. Among

the various media interviews,

I appeared in the New York

Times and USA Today and on

the CBS Evening News.

There is an irony in this

attention. The phone kept ring-

ing in the hopes of creating

that perfect bracket. Yet on my

end, I’d quickly note that even

with my work a perfect bracket

is unlikely—very unlikely.

If a fair coin toss chooses

the winner of every game,

then your chances of a perfect

bracket are 1 in 9 quintillion.

That’s 9 with 18 zeros after it.

To get a sense of that number,

if you could complete 3 billion

brackets per second and never

repeat what you have already

created, it would take almost

100 years to create 9 quintillion

brackets.

With such horrible odds,

why use analytics? First, my

methods do not flip a coin.

Historically, people have been

about 70% correct in predicting

outcomes in the tournament.

So, on average, you have a 70%

chance of predicting the out-

come of a game. In this case,

the odds reduce to about 1 in

5.7 billion as opposed to 1 in

9 quintillion.

Now comes a key part

of using analytics. Suppose

my methods give you a 1%

improvement in accuracy. Your

odds just reduced to 1 in about

2.3 billon. If the accuracy

increases to 72%, your odds

drop to under 1 in 1 billion. It’s

still unlikely that my methods

will yield a perfect bracket,

but the odds alone show the

notable improvement.

Data analysis will rarely

perfectly describe a phe-

nomenon. Customers will

make an uncharacteristic

purchase, a stock will fluctu-

ate for an apparently random