I looked at NFL win total bets twice: once in June when the lines first came out, and again in August after bettors had some time to chew on them and we got some movement. Now, since I was testing out some betting strategies, I didn't actually play these, but it is still valuable to go back and review my performance.
The idea was that I picked my 10 favorite plays in June, and then if the line moves in August were favorable, I grabbed them again. These are the plays I end up with.
Atlanta Falcons OVER 9 wins (+105) (June line) - win
Miami Dolphins OVER 7.5 wins (-110) (June line) - loss
Miami Dolphons OVER 6.5 wins (+140) (August line) - win
New Orleans OVER 10 wins (-125) (June line) - loss
New Orleans OVER 10 wins (+125) (August line) - loss
Kansas City UNDER 8 wins (-110) (June line) - win
Tampa Bay UNDER 6 wins (-120) (June line) - loss
Tampa Bay UNDER 6 wins (+105) (August line) - loss
San Fransisco OVER 10 wins (+105) (June line) - push
San Fransisco OVER 10 wins (+110) (August line) - push
Indianapolis UNDER 5.5 wins (-135) (June line) - loss
New York Giants UNDER 9.5 wins (-120) (June line) - win
Denver UNDER 9.5 wins (-120) (June line) - loss
Denver UNDER 9 wins (-110) (August line) - loss
St. Louis UNDER 6 wins (-120) (June line) - loss
This... was a disaster. These plays went 4-9-2 for a value of -4.81 betting units. On average, my preseason predictions were off by +/- 3.1 wins. Since the league wins 8 games on average almost by definition (STUPID TIES notwithstanding), +/- 3 is pretty goddamn terrible. Pointing out that my plays are disastrous might be fun for you, but the idea in making predictions is to not only dance around when they are successful, but acknowledge when they have failed so that future predictions may be improved. Let's consider what, to me, appears to be a rather apparent influence on the success or failure (mostly failure) of my predictions.
Of the 32 NFL teams, 6 began the season with new quarterbacks via either the draft or trade. (Not included on this list are teams that changed quarterback via competition, such as Tennessee or Arizona, because replacing a QB this way is unlikely in my opinion to drastically change the quality of play). 7 teams also had new coaches (I am including New Orleans on this list). There is some overlap here, so this covers a total of 11 NFL teams. Here's the list: DEN, NO, IND, JAX, TB, STL, WAS, MIA, OAK, CLE, and SEA.
Average preseason error for these 11 teams: 3.96 wins
Average preseason error for remaining 21 teams: 2.63 wins
This is too big a difference to ignore. Clearly, my system does a very poor job of considering the impact of changes at the quarterback and head coaching spots. What does my top 10 list of June look like if I exclude these 11 teams?
KC UNDER 8 wins (-110)
NYG UNDER 9.5 wins (-120)
SF OVER 10 wins (+105)
PHI UNDER 10 wins (+105)
NE UNDER 12 wins (-110)
ATL OVER 9 wins (+105)
CHI UNDER 8.5 wins (+105)
GB UNDER 12 wins (-110)
PIT OVER 10 wins (-125)
DET OVER 9.5 wins (+105)
In these 10 bets, I would have gone 5-3-2 for +1.75 BU. Stopping at 10 bets is sort of arbitrary. If I include bets for all 21 teams not included on the list above, and exclude those bets which I calculate have a negative value due to the vig, I would have added the following plays:
MIN OVER 6 wins (+100)
CIN OVER 7.5 wins (-130)
BAL OVER 10 wins (-110)
TEN OVER 7 wins (-130)
These plays bring my record to 8-4-2 for +2.52 BU. That's a pretty nice record, in my opinion. (Based on the August lines, I would have gone 7-5-3 for +2.12 BU - and even more dramatic is that on those 11 excluded teams I would have gone 2-2-7.)
Of course, it would be better if my system wasn't so deficient that it can't properly calculate the impact of a new coach or QB. Solving that dilemma will also improve my overall performance because these teams all play each other and interact with each others performance. But reviewing my predictions and identifying deficiencies is an important part of the process.