2008 Season Results

CHECK OUT THE POST ABOVE, BLACK BOX NAME CHANGE CONTEST (win $20 $50, free to enter, just submit a new name!)

These results from last year, excluding weeks 1 and 2, which there wasn't enough data for. It's broken into 3 sections: Straight up Winner, Against the Spread, and Over/Under. I broke down the approximately 200 games into 5 intervals, each interval representing a confidence spread of 10%. When you see 96 games in the 50-59% interval, that means of the 200+- games NFLSim picked, 96 games fell in the confidence interval of 50-59%. Likewise, 70 games had winners predicted as 60-69% favorites (in the straight-up category). To the left of the 'Games' column is the 'Correct' column. That is simply the number of NFLSim's correct picks in that interval. Accuracy is the percentage of time that NFLSim correctly picked a game in that interval.

1. Straight up Winner
NFLSim was absolutely phenomenal when it came to picking the winner. At first glance, you see the season record at 139-70 (66.5%). That's nothing special until you take a deeper look at the statistics.

Straight-up	% Interval	Wins	Games	Win %
	50-59%	61	96	63.5%
	60-69	45	70	64.3
	70-79	27	36	75
	80-89	4	5	8
	90-100	2	2	1
	Total	139	209	66.5

First, lets take a look at the 'expected number of wins' for each interval. In the 50-59% interval, it can be expected that after a large number of games, 55% of those games should be picked correctly(the mid-point of the 50's). In the 60-69% interval, 65% should be picked correctly, and so on. The closer NFLSim's number of correct picks is to the expected number of correct picks, the better NFLSim approximates reality. This is very important, because if you know the exact probability of a team winning, you automatically know if Vegas' payout makes for a good bet or a bad bet. The closer that the predicted probability is to the actual probability, the smarter your bets are. The smarter your bets, the more $ you make! (This is the basis for the Bet Optimization Algorithm.... which is new and much improved this year.)

If we take a look at the first interval, 50-59%, 96 games took place. In theory, you should expect that 55% of those teams predicted to win should actually win. 55% of 96 games, 55% * 96, equals an expected number of wins of 53. NFLSim actually picked 61 winners, so it beat the expected number of wins by 8. This is great because it picked more winners than it was supposed to, but it was pretty far from the expected number of wins. From 60% on is where it gets very, very interesting. NFLSim reaches near theoretical perfection - the probability given by NFLSim is the actual probability that the team will win!!! If anyone has seen this feat achieved by anyone else please let me know.

Here is a table of expected numbers of wins:

	Expected Correct	Actual Correct
50-59%	53	61
60-69	46	45
70-79	27	27
80-89	4	4
90-100	2	2
Total	132	139

In the graph: Red is what should have happened, Blue is what did happen.



Now that I've gone through the whole explanation of the stats, I'll skip straight to the numbers.

2. ATS
ATS didn't do as well as straight-up. Looks like it was pretty close to a flip of a coin. The right team would win by the wrong score. I would hate to tamper with the program when the straight-up accuracy is so reliable after 209 games. I'll probably just stop posting ATS until I can fix it up.

ATS	% Interval	Wins	Games	Win %
	50-59	37	86	43
	60-69	46	78	59
	70-79	18	33	54.5
	80-89	0	3	0
	90-100	1	1	100
	Total	102	201	50.7

	Exp. Correct	Actual Correct
50-59%	47	37
60-69	51	46
70-79	25	18
80-89	3	0
90-100	1	1
Total	127	102

Interestingly, the slope of the trend line is almost what it should be. Had 1 of the 80% games won, it would be almost perfect (albeit far lower than it should be). That suggests that all I may need to do is to add/subtract from the calculated probability to arrive at the true probability... interesting. There is back testing to be done!

3. Over/Under
Finally, if you haven't gone to sleep yet, here are the O/U stats. O/U actually did much better than I expected. It performed much better than the ATS and its accuracy approached the expected accuracy.

Over/Under	% Interval	Wins	Games	O/U Win %
	50-59	77	145	53.1
	60-69	30	48	62.5
	70-79	3	6	50
	80-89	2	3	66.7
	90-100	2	2	100
	Total	114	204	55.9

	Exp. Correct	Actual Correct
50-59%	80	77
60-69	31	30
70-79	5	3
80-89	3	2
90-100	2	2
Total	121	114

O/U actually performed very well compared to what was expected. And hell, at 56% accuracy, you'll be making money anyway.












You've seen the performance, you've seen the stats, you know what to expect. What it comes down to:

Everyone's a winner with Black Box Sports. (especially if you win $20 $50 in the name change contest)

CHECK OUT THE POST BELOW FOR THE NAME CHANGE CONTEST (win $20 $50 for sending in your best name)

Jonathan
Black Box Sports