Friday, October 1, 2010

Massey-Peabody Historical Performance

We talk about how our rankings present an unbiased evaluation of team performance.  Well, we need an unbiased evaluation of our unbiased evaluation, and the best way to do this is to look at their success in predicting outcomes relative to the point spread (PS)--the opinion of NFL bettors.  Our rankings do not account for injuries, or include matchup factors, motivation, etc. which should hurt their performance against the point spread (ATS).  Motivation becomes a large factor in the last two games of the season, since some teams have clinched playoff berths and rest starters, while other teams that are out of the running and may try out new players to evaluate them for next season.  Since our rankings do not leave room for such subjective judgments, we're not going to include the last two weeks of the season in our evaluation.  Additionally, since Massey-Peabody rankings do not use a "prior" going into the season--technically, our prior is to assume every team is average in every category entering the season--our rankings are not very informative in the first few weeks.  For this reason, we are not including Weeks 1 & 2 when looking at ATS performance.

The chart below shows how well the Massey-Peabody rankings do ATS based on the difference between the M-P line implied by the rankings and a 2.41 point home field advantage and the consensus closing point spread.  As the difference increases, the M-P Rankings do progressively better.

ATS Performance by M-P Difference (does not include 2010)
Diff W/L %
0-1 214-221 49.2%
1-2 200-205 49.4%
2-3 166-140 54.3%
3-4 125-97 56.3%
4-5 64-65 49.6%
5+ 96-76 55.8%



ATS Performance by Year (M-P Diff of 2 pts or more)
2001 52-43 54.7%
2002 57-30 65.5%
2003 42-45 48.3%
2004 53-38 58.2%
2005 49-47 51.0%
2006 51-43 54.3%
2007 46-44 51.1%
2008 52-44 54.2%
2009 49-44 52.7%
2010 15-9 62.5%

I want to look at this performance through an investment lens.  Most sports books charge a 10% commission on a bet--if your bet wins, you win $1.00; if it loses, you lose $1.10.  This means that that, for betting purposes, you need to win 11 times for every 10 losses to break even, a rate of 52.38%.  Return on money bet is calculated as (Win %)/1.1-(1-Win %).  During the 2001 season, betting based on Massey-Peabody in games with a difference of greater than 2 points would yield an average return of 4.4%.  This is informative, but does not allow for comparison to a traditional investment, where return is measured as a percentage of money invested.  To calculate return on investment (ROI) in sports betting, we'd need to establish what the initial investment is, and what % of bankroll is risked per bet.  

I'm going to do a little exercise to approximate ROI betting based on Massey-Peabody.  Although I would actually want to vary my bet size based on M-P/PS difference, let's assume, for bet staking purposes, that every bet has a 54.4% chance of winning (the average win % from 2001-2009).  I realize that we would not know this information at the time, but this will at least give an idea of what type of ROI to expect.  Using the Kelly Criterion (a formula that solves for the bet size that maximizes expected bankroll growth, based on odds offered, odds of winning, and bankroll), we would wager about 4% of bankroll on each bet.  Let's assume an initial bankroll of $1,000.  Basing all bets for a given season off of the season's initial bankroll, one would bet $40 per bet in 2001.  Using this rule, the chart below shows annual ROI.

Initial Bankroll Bet Stake Profit Annual ROI
2001  $1,000.00  $40.00  $188.00 18.8%
2002  $1,188.00  $47.52  $1,140.48 96.0%
2003  $2,328.48  $93.14  $(698.54) -30.0%
2004  $1,629.94  $65.20  $730.21 44.8%
2005  $2,360.15  $94.41  $(254.90) -10.8%
2006  $2,105.25  $84.21  $311.58 14.8%
2007  $2,416.83  $96.67  $(232.02) -9.6%
2008  $2,184.81  $87.39  $314.61 14.4%
2009  $2,499.43  $99.98  $59.99 2.4%
2010  $2,559.41  $102.38  $522.12 20.4%
This comes out to an average arithmetic return of 16.1% and geometric return of 7.1%.  This is a solid return, but nothing spectacular.  However, it is hurt substantially by the timing of the best season (2002 returned 96%) and by not varying bet size based on M-P/PS differential, since we know that a larger differential predicts a greater edge.  Massey-Peabody rankings do better as the season progresses; from Week 8 to Week 15, M-P Rankings win 57.0% with edges of 2 points or greater.  Even with the limited sample (only 465 games over the course of 9 years), the same betting strategy in the 8-week span yields  an annual geometric return of 14.5% and average arithmetic return of 20.0%.  Also, since Massey-Peabody rankings do not make any subjective adjustments, games where teams are missing key players due to injury (or suspension) produce an inaccurate line.  Filtering out these games would likely increase return substantially.

So far this season, Massey-Peabody rankings have performed admirably ATS.  While I would love to say this is due to the accuracy of the rankings, I know that it due primarily to positive variance.  It will be interesting to track returns as the season progresses.

-Rufus

No comments:

Post a Comment