f* = (bp - q)/b

where f* is percent you should wager, b is the odds, p is the probability that the wager is correct and q is the probability the wager is wrong or 1-p.

For most American
football and basketball bets for the spread or over/under the odds
are 100/110.

For example you want
to bet on college football game where you estimate that the
probability of beating the spread is 60% (or .6), based on the Kelly
criterion you should bet 16% of bankroll.

If the value of f*
is less than 0.0 you should not take that bet. For example, you
estimate you have a 52% percent change of getting the bet right the
value of f* is -0.008. For football and basketball you need to have a
probability of .525 or better in either direction to have a positive
Kelly bet.

Despite my PhD (or
because of it), I did most of my research for this topic on Wikipedia, but there is a good technical paper about by Jane Hung at Washington U.

There
are lots of articles for and against using the Kelly criterion I have
found it works quite well with my prediction algorithm where I
modeled 9 college football seasons using Kelly bets with a few
modifications (basically, I capped the percentage of bankroll at 30%)
and my model compounding the winnings,
in 8 seasons my model would have won money
with average increase over starting bankroll of 40,054.23% (sic).
This increase is really high a small sample due to an exceptional
year for 2013 model where the model had me winning 22,932.89% of my
payroll. If you remove that year there was still a 2140.175%
increase over the initial bankroll.

I will include my
raw Kelly criterion values on the page.

## No comments:

## Post a Comment