[Bsims]

Last year I fooled around a bit on an analysis I called WL Strings. Unfortunately I started after the football season started. I really think it would be effective only for the first few games of a season. And I think it's most likely to work for CFB since there isn't any pre-season. I've decided to share it in this forum as the project moves along. I'd welcome suggestions or questions.

The theory starts with a couple of assumptions. First, I believe the final lines are the results of bookies opening lines, sharps, and the public's wager patterns. In effect, the line reflects the overall perceived value of a team. This perceived value increases or decreases with the results of a team's games. These perceived changes are then reflected in the lines for the next games.

The question then becomes, do these changes over or under state reality? In other words, are lines moved too much, or too little? I am working on a program to address this question. I have a file with 10 seasons of CFB results. It has 6,976 games (I only use games between major colleges).

The program looks at each games final scores and the spread. It assigns a W, L, or P based on whether a team covered or not. Last year's version did not consider how much a team covered by. This year I plan on setting a range of how far from the spread will be considered a push and then generating the results.

That's enough for now, I've got to get back to the program.

[u21c3f6]

Bsims, This is very similar to what I do (successfully, i.e. profitably) in CFB as well as other sports. You should not have to go back more than 2 years to find the "sweet" spot where lines are either over or under rated. It is not about picking winners, it is about making selections that have a positive EV. Good luck.

Joe.

[Bsims]

Fred Brooks in his book "The Mythical Man-Month: Essays on Software Engineering" argues that the common practice when a project gets behind is to add staff. In fact, he argues you should remove some. Well this project is behind, but the staff is one, so removing me kills the project. So, we'll be late.

Thus far here are some results (range is the distance from the spread that defines a push).

Range WW Pct LW Pct
0 .494 .533
3 .475 .513
7 .483 .473
10 .505 .525
14 .532 .539

WW means the team that covered in their 1st game covers in the 2nd. LW indicates a team not covering in the first game covers in the 2nd. The data does tend to show an over reaction to the 1st game results. But overall it is inconclusive. I'll look at a little different approach.

[Bsims]

The initial question I had after last year was is there a push range that is better than a simple covered or not (push range=0). So I modified the program to iterate through differend push rates to see if there was an obvious difference. Here is the resulting table;

PushRange	Combination	Won	Lost	Push	W-Pct	L-Pct
0	W:L	1468	1582	51	0.481	0.519
1	W:L	1258	1338	187	0.485	0.515
2	W:L	1072	1168	310	0.479	0.521
3	W:L	911	986	412	0.48	0.52
4	W:L	744	791	453	0.485	0.515
5	W:L	610	656	477	0.482	0.518
6	W:L	491	528	467	0.482	0.518
7	W:L	421	440	483	0.489	0.511
8	W:L	354	359	472	0.496	0.504
9	W:L	279	277	450	0.502	0.498
10	W:L	209	195	385	0.517	0.483
11	W:L	148	140	364	0.514	0.486
12	W:L	118	105	344	0.529	0.471
13	W:L	98	91	322	0.519	0.481
14	W:L	74	69	289	0.517	0.483
15	W:L	51	54	225	0.486	0.514
16	W:L	36	44	188	0.45	0.55
17	W:L	28	21	139	0.571	0.429
18	W:L	16	19	129	0.457	0.543
19	W:L	11	15	110	0.423	0.577
20	W:L	10	11	92	0.476	0.524
21	W:L	5	8	75	0.385	0.615

What we see is that the team that did not cover in their first game has a slight advantage when playing against a team that did. But the advantage is pretty small. Moreover, there does not appear to be much change as we increase the number of games considered as pushes. All in all, not to productive.

[Bsims]

In spite of what I feel are marginal results, I dug out last year's program and modified it for this year's structures and data. I decided to use a push range of 0 (simple cover or not) just like last year. Finding games with a prior weeks loser versus a winner identified 16 games today. My analysis indicates that they should have a 5.3% chance of covering (to close to wager on). For what it's worth, here is the list;

WASHINGTON ST	-->	BOISE ST
CHARLOTTE	-->	KANSAS ST
CINCINNATI	-->	MICHIGAN
FLORIDA ATLANTIC	-->	WISCONSIN
UCLA	-->	UCLA-L
NO CAROLINA ST	-->	NO CAROLINA ST-L
LOUISIANA TECH	-->	LOUISIANA TECH-L
NEBRASKA	-->	OREGON
NEW MEXICO	-->	NEW MEXICO ST
NORTHWESTERN	-->	DUKE
PITTSBURGH	-->	PENN STATE
MISSOURI	-->	SO CAROLINA
TEXAS STATE	-->	COLORADO
TOLEDO	-->	NEVADA
FLORIDA ST	-->	LA MONROE
BOSTON COLLEGE	-->	WAKE FOREST

[Bsims]

Obviously I posted 3 games in error with the same team on both sides (copy and paste error). The others were 5-6-1, a small sample but in line with the conclusion that there isn't anything here worth pursuing.