Second Half Middling

[count zero]

[buckeye]

I've dabbled with it a bit. It seems to have more potential in college hoops than most sports. I don't have numbers to crunch but a lot of teams that are up big at half seem to coast a bit in the second and the opposition makes it a close "cover" for the game!

[count zero]

Oops! I actually had more to say than that. Hope it works this time.



Following is a work in progress, all questions and no answers. I could use some guidance.

Was recently re-reading The Half Book (Granowski). The author presents a formula for projecting a second-half sides line from the whole-game line and the halftime margin. I calculated a few weeks worth of lines by this method and compared them to real lines from Olympic that I'd saved from last year; after making one small adjustment (the Granowski book was written in 1999), I got them to converge pretty well. I then decided to implement the formula in a program and write out the lines to my DB, then use those "lines" to research the performance of a variety of second-half betting concepts.

So my first question is: Do you think these calculated lines are better or worse for use in research than the actual lines would be? I can think of arguments for both sides.

OK. I started researching by looking at the frequency with which certain second-half margins landed. I didn't bother breaking down the games where the dog won the second half, since we can't hit a middle when that happens. The results from 1983 through last year using all games was:

2H A Push 412 9.501845
Fav By 1 107 2.4677122
Fav By 2 52 1.199262
Fav By 3 286 6.595941
Fav By 4 193 4.451107
Fav By 5 38 0.8763838
Fav By 6 138 3.1826568
Fav By 7 359 8.2795203
Fav By 8 61 1.4068266
Fav By 9 45 1.0378229
Fav By 10 206 4.7509225
Fav By 11 105 2.4215867
Fav By 12 33 0.7610701
Fav By 13 90 2.0756458
Fav By 14 192 4.4280443
Fav By 15 18 0.4151292
Fav By 16 25 0.5765683
Fav By 17 107 2.4677122
Fav By 18 32 0.7380074
Fav By 19 8 0.1845018
Fav By 20 26 0.599631
Fav By 21 61 1.4068266
Fav By 22+ 90 2.0756458

Fav Won 2H 2272 52.40 (57.9% of non-tied games)
Dog Won 2H 1652 38.10 (42.1% of non-tied games)
2H A Push 412 9.50

Total Gms 4336 100.00


So my second question is: Do you have any comment on the accuracy of this table? Is it so great and of such immeasurably profitable use that I should delete it immediately? Is it filled with errors? Anything I should have done different? My problem here is that I have nothing to check these numbers against, and programming errors are a worry.

The next thing I did was to get two separate tables for before and after the 2-point conversion came in in 1994:

1983-1993 1994-2001

Fav By 0 254 10.55% 158 8.20%
Fav By 1 56 2.33 51 2.65
Fav By 2 19 0.79 33 1.71
Fav By 3 156 6.48 130 6.74
Fav By 4 110 4.57 83 4.30
Fav By 5 12 0.50 26 1.35
Fav By 6 66 2.74 72 3.73
Fav By 7 203 8.43 156 8.09
Fav By 8 28 1.16 33 1.71
Fav By 9 17 0.71 28 1.45
Fav By 10 128 5.32 78 4.05
Fav By 11 48 1.99 57 2.96
Fav By 12 18 0.75 15 0.78
Fav By 13 51 2.18 39 2.02
Fav By 14 117 4.86 75 3.89
Fav By 15 8 0.33 10 0.52
Fav By 16 16 0.66 9 0.47
Fav By 17 69 2.87 38 1.97
Fav By 18 13 0.54 19 0.99
Fav By 19 3 0.12 5 0.26
Fav By 20 15 0.62 11 0.57
Fav By 21 48 1.99 13 0.67
Fav By 22 47 1.95 43 2.23
Dog Won 2H 906 37.62 746 38.69
2408 games 1928 games


Oddly, these results show the opposite trend from whole-game results -- the key numbers land less often with the 2-pointer available than without. 3 is about the same, but 0, 7 and 10 are all diminished.

So my third question is: WTF?

Next, I restricted the frequency distributions to games where the second half line was within a half point of the margin being tested. To give myself more time for relaxing later in front of the TV with a blunt or two, I only looked at the primary middling candidates 0, 3 and 7:

1983-2001 (All games)

When the Number of 2nd half This many For this
line is: occurences: lands on: times: percentage:

PK to .5 2001 0 207 10.34%
2.5 to 3.5 1193 3 102 8.55
6.5 to 7.5 313 7 41 13.10

1994-2001 (2-point conversion games only)

When the Number of 2nd half This many For this
line is: occurences: lands on: times: percentage:

PK to .5 879 0 76 8.65%
2.5 to 3.5 541 3 49 9.08
6.5 to 7.5 119 7 16 13.45


This param has little effect at 0, probably because PK is used for a wide variety of game situations, including, for example, blowouts. 7 and 3, on the other hand, are greatly potentiated and clearly good middling candidates.

My fourth question is: I'm guessing that 7 outperforms 3 and 0 (and that 3 and 7 do better with the 2-pointer here) because of small sample size. Agree? Disagree?

Finally, before I lit up and retired to the couch to watch Fear Factor, I ran the above analysis again, except eliminating games where the 2nd half fav is losing at halftime by the same margin as the 2nd half line. In other words, if the Jets lead the Giants by 3 at the half and the second half line is Giants -2.5 at joint A and Giants -3.5 at joint B, that's still not a good middle, as the Giants will win the second half by 3 only when the game ends in a tie.

1983-2001

When the Number of 2nd half This many For this
line is: occurences: lands on: times: percentage:

PK to .5 1892 0 196 10.36
2.5 to 3.5 1192 3 102 8.56
6.5 to 7.5 305 7 41 13.44


1994-2001

When the Number of 2nd half This many For this
line is: occurences: lands on: times: percentage:

PK to .5 834 0 72 8.63
2.5 to 3.5 541 3 49 9.08
6.5 to 7.5 119 7 16 13.45


Thought this would produce a big improvement in hitting the middles, but no. So my last question is: why not?

Like I said, all this is first-draft stuff. I need a direction to follow, or somebody to slap me. Anybody?

[GottaWinToday]

It's going to take some getting used to this tech stuff. Let me try to ingest or digest this stuff and maybe, only maybe will I see the light or have an opinion......

I think this is going to be a great forum. I'm going to have to get my computer calc in gear though.

GottaWinToday

[count zero]

To address what buckeye mentioned in a hoops context, the Granowski book definitely recommends taking the 2H dog in a blowout. That's why the halftime margin is one of the params for figuring the 2H line. Apparently, the value of this play has been steadily declining as the books "get it." That's one of the reasons I think projected 2H lines might be more useful than the actual 2H lines -- numbers from several years ago were weaker.

Sorry about the formatting in my tables above. If you copy and paste into a WP or text editor using a fixed-width font, they should come out right. The headings for the last group of results are supposed to be:

When The Line Is:
Number Of Occurences:
Second Half Lands On:
This Many Times:
For This Percentage:

[j]

My bias is toward using actual lines, over those made from a formula but at the moment I can´t think of any defensible reason for that.

I´ve also heard line'makers speak of teams with habits, for example the Lakers last year had the profile of getting down and at the half, but during stretches of their season whatever that margin was was insignificant as the Lakers were probably going to make it up and everyone (public and linesmaker)knew that.

On those games, then, the line didn´t match up with what you would find it to be with an identical score for teams other than the Lakers.

I guess maybe I just reminded myself why I think actual data might be a bit more valuable than what could be generated from a fitted formula. By the way, I´m not convinced the books are getting sharper about making halve lines, it could be the players are and the books are responding.

[Mr White]

I have always wanted to dabble in halftime bets. I always suspected that they might be easier to beat a halftime line than a game line. Unfortuantaly I have never had access to a data set that contained halftime lines, scores and data. Does anyone know if there is a stats site out there that contains halftime data in an easily downloadable format?

[count zero]

Mr White, I never could find 2H lines in a useable format. Don Best is said to have archived 2H lines, but I couldn't find anything useful there, possibly because I'm not a subscriber. That's why I went to the projected lines. Of course, any database that gives scores by quarter, such as CSW, will have the needed halftime scores. As for halftime stats, good luck -- the "GameBook" section of the scoreboard at nfl.com has same, but only in a PDF format. You'd have to do an unbelievable amount of grunt work to get something usable.

I'd be glad to try to help with the lines if you can reconcile yourself to using projected lines (see defense of same below). They're in separate files by week, so it would be hard to e-mail them out, but I might be able to send along the code that generated them (in VisualBasic, the Excel query language, so if you're an Excel wizard you might be able to get Excel to generate the numbers). Or maybe I could make a change in the programming so it writes all the lines from 1983-2001 into one file. Or you could IM any interesting concepts you have and I could test them against the data I have.

j -- right, when I said the books were getting smarter, I meant that they were responding to the strategies of winning bettors. But this points up where projected lines might have an advantage over real lines. All those 2H lines from 1999 and earlier were hung without any general awareness that teams being blown out were great 2H bets. After that strategy became generally known, the way books figured 2H lines changed. So any results you find in researching with real lines from that period before 1999 are spurious -- the results they give won't reflect current reality. Projected lines will.

[KingOfTheSquares]

My take on half lines is this: You now have twice as many times to make a bet given that your book posts both 1st half and 2nd half lines before the game thus you should double your expected winnings a season.

Now this means that over a season, if you are picking at say: 55% and if you find an advantage in 3 games out of the 15/14 per week (call it 20%) in the NFL. You should have:

Sunday early... avg of_ 8_ games = 14.1 betting intervals
Snday late... avg of_ 4_ games = 10.0 betting intervals
Sunday nite... avg of_ 1_ games = 3.4 betting intervals
Monday nite.... avg of_ 1_ games = 3.4 betting intervals
Thursdays... tot of_ 3_ games = 8.3 betting intervals
saturdays... tot of_ 5_ games = 11.4 betting intervals

for a total of 50.7 betting intervals

To make it simple lets's just say that you bet the best game (if there are multiple games in the interval) then your net wins over the season is 2.78 units over the season (27.9 unit wins and 25.1 units losses at a 55% win rate)

This would double to +5.56 units (over 101.4 betting events) if you can bet each half that meets your 55% criteria

Now if you bet each game that meets the criteria, your unit wins becomes +4.1 for the 50 intervals or +8.2 for the half betting over the 101.4 games.

[Total Square]

nice to see the Granowski book get some play.
jack was a friend, and sadly he passed away earlier this year at a much too young age.

when the rec.gambling.sports newsgroup was started back in 95-96, he was the only sports book manager to come into the forum (before it became totally worthless) and shared his thoughts with the readers on a regular basis. at the time, he was the manager of the book at the fiesta.

at one point earlier in his career, he worked at the barbary coast. for the sole reason the 'he was from minnesota', he was put in charge of putting up lines on the nhl games. i don't know if there was much action on nhl back then, but it sure grew under his leadership. i heard that the barbary used to get their heads handed to them with his lines, but the players would also bet other sports that made the endeavor worthwhile for the house.

he was a real good, and smart, guy.

[pokerjoe]

First, thanks for posting all this.

About "projected" lines value: I think it's a better method. The sample of actual lines appears to be small; the definition of what the line is, vague anyway. (quick sign that someone's not all that sharp: when asked what's the line on a game, they only give you one number. Sharps give 2 or 3, right?)

Also, ranges make more sense (and expand the sample). "Team's favored by 2-4" makes for a larger sample, without losing accuracy. The difference in likelihood of a team favored by 2 winning by 3 and one favored by 3 landing on 3, is trivial, in terms of real world DB sizes. Especially in your projected lines method. I'd like to see the "# of times teams favored by 2-4 landed on 3", etc. I think it would have more validity.

For example, I once ran off a poker simulation in which it took more than a million hands for 7's to outperform the smaller pairs. If I'd used the results from a mere 100,000 hands, I'd conclude that 7's were weaker than 6's, which is nonsensical. I prefer to rely on theory for conclusions. Ultimately, honestly, I believe programming a computer to run off game scores (with 2pt conversion strategy factored in, as well as team superiorities) would be far, far better than actual game results.

"Oddly, these results show the opposite trend from whole-game results -- the key numbers land less often with the 2-pointer available than without. 3 is about the same, but 0, 7 and 10 are all diminished."

Let's remember that the changes wrought by the 2-pt conversion stem from game final scores. Teams try to get ahead or behind by 3, 7 and 10, for the GAME, not the 2nd half. So 2nd half scores should be LESS likely to land on the 7, because teams are less likely to take the extra point attempt in the second half than the first, because game situations warrant it. Scoring in 1st halfs is more likely to come only in 3's and 7's (teams don't worry so early about margin ramifications), whereas in the 2nd half, 5's and 8's will be more common as teams fail or succeed at 2pts, and 7's correspondingly less so.

Key then: the football primes (3, 7, 10) are MORE likely to land in the first half than the second; the non primes (especially 5 and 8) are more likely to land in the second half.

And your point about team's favored by the # they trail by, is very well made. Is there anyway a game with a team favored by 3, trailing by 3, can land on the #? Doesn't the 2ndH line include OT?

Well, let me think about the 3's: is there logically a reason the frequency of 3 pt. margins should change by half? First 1/2 should be pure (as I said, there's little 2 pt. scoring strategy). But 2nd Hs? What we're talking about, in all this, really, is the effect of final game margins strategy on 2nd half actual margins. And I don't think there'd be a difference in 3 pt. 2nd H margins, because the instances when teams forego trying FG's (because the trail by too many) should, far as I guess, be negated by the times they forego TD attempts to settle for FG's. And although 2pters increase the 3 frequency for the game (because strategy demands the attempt to get ahead or behind by 3 instead of 2 or 4), do they for the 2nd half? Maybe not.

[KingOfTheSquares]

Pokerjoe! Welcome back! Somebody said you had left the planet!

Anyway, about your point on building a simulation to project the occurance of 3s...well, I tried it and you have a little problem.

You can't model what's going on inside each coach's head. What I'm talking about is the same problem we see in the Spurrier thread: namely the arch conservative tactical moves that the NFL head coaches have been making since the game has been played. Therefore, you cannot run the sim optimally because the result will NOT come close to matching past reality simply because the coaches today and yesterday make non-optimal choices.

This is not same as figuring the best way to play a hand of poker. As you point out, given you have something close to a good random number generator (and we can get into a whole discussion on that as you have to go from 100,000 hands to 1,000,000 hands to see a difference. BTW, have you read Averill Law's book? He has a technique to sort out variation in Monte Carlo simulations).

This is about how to figure the odds of 3, 7 & the entire freq dist., occuring in the future given the changing nature of NFL coaching. And for that, unfortunately, past history is the best and only model...and yes, it will change as the coaching "fashions" change. And I contend there ain't a damn thing the best statisticians and sim builders "on the planet" can do about it.