[a4u2fear]

Starting a new thread with hopefully more insight. In a prior thread, I used regression to create a formula/model from 2000-2011 and tested the formula on 2012 and 2013.

When my model said the total was 2 points higher than it should've been (vs closing line)
The model was 61W - 39L - 1T for 61% (playing the under)
In these games, closing line was closer the actual total 37 times and the opening line was closer 50 times (14 ties)

When my model said the total was 2 points higher than it should've been (vs opening line)
The model was 55W - 46L - 1T for 54.5% (playing the under)
In these games, closing line was closer the actual total 38 times and the opening line was closer 50 times (14 ties)

If I expand the 54.5% vs opening line, where the model said the total was 4 points higher:
28W - 20L - 1T for 58.3%

Now, before you say 49 plays are not enough, that is 49 plays over two years, from weeks 9 to 17. So I played 49 / (9*16) = 34% of the total allowable games.

I've read that I should be testing my model ALWAYS against the opening line because A) beating the closing line is always a good thing and B) the market is efficient. My argument here is, I have a local bookie and can only bet the lines 1 hour prior to game time. In this last hour, the totals are moving much less than the spreads, as only high winds/really bad weather strolling in, or big player inactives occur.

Thoughts are appreciated.

[a4u2fear]

Should've stated that the model vs closing line had 101 plays, the model vs opening line had 102 plays. Of these, both models vs the line shared 81 plays.

[nikossf]

very interesting... would love to see model in use. start posting plays.

[a4u2fear]

Whoops, messed up total plays, should've been 49/ (2years*9weeks*16gamesperweek)=17%

[JR007]

good luck

[HeeeHAWWWW]

Random points at 100mph:

1) Use rmse to compare closer/opener totals vs results.

2) Rather than banding by points, do it by % difference from dejuiced implied probability. It's only going to be a minor difference, but it's an easy change to improve accuracy.

3) Split your bets into bands of estimated edge vs IP, and check correlation against that band's RoI.

[gui_m_p]

Forget that you can only bet at a local book 1 hour prior to game time. Now suppose that you'd have bet all picks at pinnacle opening line. See what lines you would get and compare it with closing ones.

[a4u2fear]

Just a brief summary, for the performance vs closing line:
when line is off by 1 pt or more, under is 72-58, 55.4% (+8.2 units vs juice)
when line is off by 2 pts or more, under is 61-39, 61.8% (+20.2 units vs juice)
when line is off by 3 pts or more, under is 42-26, 61.8% (+13.4 units vs juice)
when line is off by 4 pts or more, under is 32-18, 64.0% (+12.2 units vs juice)

Hence why I chose the 2 pt guideline

[a4u2fear]

HeeeHawwww

2) Rather than banding by points, do it by % difference from dejuiced implied probability. It's only going to be a minor difference, but it's an easy change to improve accuracy.

Can you provide an example?

3) Split your bets into bands of estimated edge vs IP, and check correlation against that band's RoI.

Can you explain a little further? Does my summary above provide a little better analysis now that I show the pct hit vs juice?

[a4u2fear]

So, a wise man once said (Justin7), if you yourself are a "sharp" and books know it, when you bet, you will often beat the closing line.

Now obviously I'm not saying I'm a sharp, but if indeed my system is effective and I ever get to some level where I could move the line (unlikely), even though I'm betting the "closing line", I will often beat the closing line because it will move after my bet is placed.

Ok more stats:
System was 30-21 in 2012, and 30-18-1 in 2013, near identical performance.

The system is 23W-8L in weeks 15,16,17, unbelievable 74%

When the total is less than 50, system is 64%, when it is 50 or above system is 58%

[chrismunney]

nice

[Jason Jones]

Interesting