Friday, February 15, 2008

The Quest to Backtest

The Herculean task of acquiring past data and parsing it into a useful format has begun. Sometime in the next 6 months you can expect to see results from the first few weeks of the 2007 season, and possibly results from the 2006 season. Other stuff will be attempted as well.

Tuesday, February 5, 2008

Superbowl XLII Bets

Here are hypothetical bets that would have been placed on the Super Bowl. I've used alternative spreads in addition to the normal spread to add some extra betting opportunities. Also the weird bet amounts come from an algorithm I developed which is used to determine bet size based on predicted %, payout, and bankroll. Sometimes it even has you on both sides of a bet, hedging yourself. Very interesting... here they are: (also bet amounts are in terms of percentages of your bankroll)

In order of best bet to good bet, starting with single bets.

Over/Under:
Under 53.5: 82%
Payout: 1.91
Bet Amount: 12.98
Win: 24.80

Spreads:
NYG +17.5: 77%
Payout: 1.36
Bet Amount: 10.75
Win: 14.62

NYG +15.5: 72%
Payout: 1.5
Bet Amount: 9.54
Win: 14.31

NE Win: 67%
Payout: 1.24
Bet Amount: 7.87
Win: 9.76

NYG +11.5: 63%
Payout: 1.91
Bet Amount: 7.81
Win: 14.91

NYG +7.5: 54%
Payout: 2.4
Bet Amount: 6.23
Win: 14.96

NYG +5.5: 44%
Payout: 2.85
Bet Amount: 4.36
Win: 12.42

NE -5.5: 56%
Payout: 1.36
Bet Amount: 4.23
Win: 5.75

Parlays:
Under 53.5, NYG +17.5: 63%
Payout: 2.6
Bet Amount: 8.76
Win: 22.75

Under 53.5, NYG +15.5: 59%
Payout: 2.87
Bet Amount: 8.01
Win: 22.93

Under 53.5, NYG +11.5: 52%
Payout: 3.65
Bet Amount: 6.87
Win: 25.07

Under 53.5, NE Win: 56%
Payout: 2.37
Bet Amount: 6.84
Win: 16.21

Under 53.5, NYG +7.5: 44%
Payout: 4.58
Bet Amount: 5.75
Win: 26.35

Summary:
After looking at the different bets, it becomes clear that the most likely outcome of the game was a close, low scoring affair. While NE was still predicted to win, their slim margin of victory opened up a myriad of betting opportunities.

The numerical results:
Had you started the day off with $100, by the end of the day you'd have $193.12 and a smile on your face.

Monday, February 4, 2008

2007 Results, Analysis, and Betting Applications

Well, since starting back in week 12, I'd say I've gotten some pretty good results. Now before everyone starts yelling at me, I'll preempt it by saying that yes, I know it was only a quarter of the season, and yes, I know that for the spread and o/u it was only a few games. At this point the results may be statistically inconclusive, but it's a start. I worked with what was available- the program was 'completed' in time for week 12. Up until the playoffs, changes were being made to achieve a greater degree of realism. Those changes in no way affected either team exclusively, they affected both teams in exactly the same way.

The results for the playoffs and Super Bowl:

Wins: 4-7 36.4%
Spread: 5-2 71.4%
O/U: 6-1 85.7%

The Law of Any Given Sunday proved too much for me. Though you can't argue with a spread/ o/u combined win percentage of 78.6%


The results for the 2007 regular season, weeks 12 through 16:

Straight-up Wins
% Range Games Wins Win %

50-59 34 21 61.8%

60-69 29 19 65.5%

70-79 14 11 78.6%

80-89 3 3 100.0%

90-100 0 0 0.0%

Total 80 54 67.5%


This is a graph of the above table. It shows the astonishing correlation between the predicted percentages and the real win percentages. The x-axis shows the different % Ranges- like in the tables. The y-axis is the win percentage. So the predicted range 50-59% has a real win percentage of 61.8%. The straight, dark line shows what would happen if all of the teams won exactly as they were supposed to, according to my predicted percentage. Lets say I pick 100 teams to win, and they all fall in 50-59. If my system was perfect, about 55% of those teams would actually win. Since the green line (what actually happened) is so similar to the black line, we're able to conservatively assume that the predicted percentage to win is the same as the real percentage of a straight up win.


Now that we know the probability of a team winning, all of a sudden the moneyline completely opens up. Now we can look at what Vegas is offering and immediately know if it's a statistically favorable or unfavorable bet.

If you have $100 and you make 100 bets at 50% each, and each bets pays out 2:1, then after your 100 bets, you've won 50 of them. Since they pay out 2:1, you've made it back to $100, breaking even. Likewise, if you have $100 and you make 100 bets with a probability of 20%, to break even, each bet must pay 5:1. If each bet pays 4:1, you've won 20 bets, but at 4:1, you only make it back to $80, so you've lost money. At 6:1 you get up to $120. Break-even odds are neutral- in the long run you don't make money, and you don't lose money. We only make bets that are above break-even.

Let's look at the Super Bowl. The Giants were 11.5 dogs. Vegas was offering to pay out 4.25 to 1. The probability that the Giants would win- as calculated by me- was 33%. If you win 33% of your bets, each bet must pay out 3.03:1 (Break-even payout = 1/percentage in decimal form; 1/.33 = 3.03). So, since you need at least 3.03:1 odds to make an intelligent bet, you note that Vegas wants to give you 4.25. 4.25 is bigger than 3.03 so you bet on the Giants to win. And if you had bet on the Giants to win, you'd be one happy camper right now.

Note: If you bet in this manner using a reliable source of percentages, sensible money management, and patience, you'll be successful in the long run...with the moneyline! And it doesn't hurt when you're picking 68% of the games correctly... Of course this also works with the spread and o/u.

PLAYOFFS/SUPERBOWL w/ Spread and O/U

Playoff Rd 1 Win %
WAS
SEA 71%


JAX
PIT 66%


NYG
TB 68%


TEN
SD 59%

Wins: 2-2 50%


Spread and Over/Under were finally completed!
For spread and o/u, the percentage is the chance that the favorite team would cover the spread and that the total points would exceed the over. If the percentage for the spread is above 50%, if the favorite covers the spread, it's a win (green); if the favorite doesn't cover the spread, it's a loss (red). For a percentage below 50%, the favorite probably won't cover the spread. So if the favorite covers the spread, it's a loss; if the spread is not covered, it's a win.
To summarize: if the spread and over are above 50% (probably will happen) and the spread is covered and the points exceed the over, they are wins. If they are under 50% (cover and over probably won't happen) and the spread is not covered and the points fall below the over, they are wins. Visa versa for losses.

Example: SEA was projected to win 53%. They lost, so Win is red. The spread was predicted to be covered 27%. It was covered, against the probability, so it was a loss. The over was predicted 65%. Since the actual score went over the line, it was a win.

Playoff Rd 2 Win % Cover Spread % Over %
SEA 53% 27% 65%
GB






JAX


NE 63% 31% 53%




SD


IND 69% 46% 22%




NYG


DAL 72% 49% 45%

Wins: 1-3 25%
Spread: 3-1 75%
O/U: 3-1 75%


Playoff Rd 3 Win % Cover Spread % Over %
SD


NE 71% 33% 47%




NYG


GB 64% 41% 61%

Wins: 1-1 50%
Spread: 2-0 100%
O/U: 2-0 100%

Superbowl Win % Cover Spread % Over %
NYG


NE 67% 56% 20%

Wins: 0-1 0%
Spread: 0-1 0%
O/U: 1-0 100%

WEEK 16

Week 16 Win %
PIT 57%
STL


DAL 63%
CAR


NYG 51%
BUF


GB 66%
CHI


CLE
CIN 59%


KC
DET 59%


HOU
IND 73%


PHI
NO 62%


OAK
JAX 60%


ATL
ARI 51%


TB 58%
SF


NYJ
TEN 60%


MIA
NE 71%


BAL
SEA 70%


WAS
MIN 72%


DEN
SD 55%


Total % Range Games Wins Win %

50-59 7 6 85.7%

60-69 5 3 60.0%

70-79 4 3 75.0%

80-89 0 0 0.0%

90-100 0 0 0.0%

Total 16 12 75.0%

WEEK 15

Week 15 Win %
DEN 54%
HOU


CIN
SF 54%


BUF 53%
CLE


TEN
KC 64%


GB 56%
STL


BAL
MIA 54%


NYJ
NE 71%


ARI
NO 77%


JAX
PIT 72%


ATL
TB 70%


SEA 55%
CAR


IND 62%
OAK


PHI
DAL 69%


DET
SD 58%


WAS
NYG 66%


CHI
MIN 69%


Total % Range Games Wins Win %

50-59 7 4 57.1%

60-69 5 2 40.0%

70-79 4 3 75.0%

80-89 0 0 0.0%

90-100 0 0 0.0%

Total 16 9 56.3%

WEEK 14

Week 14 Win %
CHI
WAS 60%


MIA
BUF 64%


STL 51%
CIN


DAL 64%
DET


OAK
GB 57%


SD 54%
TEN


NYG
PHI 57%


CAR
JAX 60%


TB 58%
HOU


MIN 66%
SF


ARI
SEA 82%


KC
DEN 62%


PIT
NE 54%


CLE
NYJ 63%


IND 61%
BAL


NO
ATL 58%


Total % Range Games Wins Win %

50-59 7 3 42.9%

60-69 8 7 87.5%

70-79 0 0 0.0%

80-89 1 1 100.0%

90-100 0 0 0.0%

Total 16 11 68.8%

WEEK 13

Week 13 Win %
GB
DAL 63%


HOU
TEN 59%


JAX
IND 79%


SD
KC 54%


ATL
STL 66%


NYJ
MIA 63%


DET
MIN 55%


SEA
PHI 60%


BUF
WAS 63%


SF
CAR 61%


DEN 51%
OAK


CLE
ARI 61%


NYG 51%
CHI


TB
NO 53%


CIN
PIT 80%


NE 65%
BAL

Total % Range Games Wins Win %

50-59 6 3 50.0%

60-69 8 5 62.5%

70-79 1 1 100.0%

80-89 1 1 100.0%

90-100 0 0 0.0%

Total 16 10 62.5%

2007 Picks and Bets

Before I start listing everything, I'll post this website: http://sports.espn.go.com/nfl/features/talent. Granted they're NFL experts, but what do they know anyways? The column you should check out is Accuscore, the industrially-sized handicapping website (also the only other play-by-play simulator). They finished the season 163-89, or 64.7%. But wait- what really matters is the spread, right? Well, sort of, but we'll get into that later. Accuscore claims an (un)impressive 54.7% spread prediction on the season, with spread and over/under combining for 60%. But isn't break-even about 53%? Hmm...

By the way, from weeks 12 through 16 (for week 17, it's not worth the work if most teams don't even show up), I was 67.5%. Accuscore was 63.7%.

Get ready for lots of numbers, as there are many weeks to catch up on:

Winning predictions in green, losers in red, ties in black.

Week 12 Win %
GB 52%
DET


NYJ
DAL 72%


IND 72%
ATL


TEN 58%
CIN


HOU 53%
CLE


OAK
KC 60%


SEA 60%
STL


MIN 56%
NYG


WAS
TB 75%


NO 58%
CAR


BUF
JAX 57%


SF 52%
ARI


DEN 70%
CHI


BAL
SD 65%


PHI
NE 78%


MIA
PIT 81%

The table: If % Range is between 60-69, it includes the 3 teams where the win probability fell in that range, i.e. SD, SEA, and KC. Of those 3 teams, 2 of them won. Therefore the win % for teams in the 60-69% range is 66.7%.

Total % Range
Games Wins Win %

50-59 7 5 71.4%

60-69 3 2 66.7%

70-79 5 4 80.0%

80-89 1 1 100.0%

90-100 0 0 0.0%

Total 16 12 75.0%

My First Post, an Introduction

If you've read the 'About Me' column then you already know about me and my program. Over the past 6 months, I've worked everyday to perfect this simulator. By week 12 I was able to start picking games straight up. I managed to get the spread/over/under feature working just in time for the playoffs.

It hadn't occurred to me to create a blog to document my progress until...this afternoon. I stumbled onto a blog here during my search for solid statistics showing the win percentages of 'experts', handicappers, and websites- some of which, as you know, charge hundreds of dollars for one week of picks. I was curious to see how the success of 'expert handicappers with 20 years of experience' compared to the success of 'some kid with with a computer'. I was ecstatic to see that not one entity came close to matching me.

Unfortunately, my lack of a blog has prevented me from sharing my prognostications with the public until now. Therefore, I shall present to you all of my results, along with all 'theoretical' bets* that may have been placed on those picks. This will demonstrate both the accuracy and efficacy of my program.

*The amount of each bet may seem weird- the size of each bet was determined using a formula which takes into account the probability of a win, the payout, and the bankroll, in order to maximize profit and minimize losses...duh.