[usma1992]

college football. As some or many of you know, I have been working on generating several statistical models to predict the outcome of games. I have learned many lesson over the course of the last several years and I believe that I am finally dialed in. I am not asking for anything... but feedback on the picks that I send.

The last three weeks have been 62%, 68%, and 55%... I am hoping it continues... Here are the picks, critique them... I am interested in feedback. Good luck to all...

09:00 PM	LaLafayette	23	NorthTexas	30		4.5	56.0	U
8:00 PM	Houston	0	SMU	0		6	60.0	O
9:00 PM	Oregon	0	ArizonaState	0		9.5	68.5	N
8:00 PM	Connecticut	0	Syracuse	0		-4.5	43.5	N
12:00 PM	Nebraska	0	Northwestern	0		5	62.5	N
12:00 PM	BowlingGreen	0	Massachusetts	0		17.5	51.0	N
12:00 PM	NorthernIllinois	0	Akron	0		16	65.0	O
12:00 PM	Rutgers	0	Temple	0		5.5	41.5	U
12:00 PM	VirginiaTech	0	Clemson	0		-8.5	61.5	O
12:00 PM	Purdue	0	OhioState	0		-19.5	62.5	N
12:00 PM	LSU	0	TexasAM	0		3.5	52.5	N
12:20 PM	Auburn	0	Vanderbilt	0		-7	43.0	U
12:30 PM	WakeForest	0	Virginia	0		-3.5	53.0	N
1:00 PM	Army	0	EasternMichigan	0		2.5	60.5	N
1:00 PM	FloridaInt	0	Troy	0		-7	55.0	N
2:00 PM	SanJoseState	0	UTSA	0		13	52.5	U
3:00 PM	BostonCollege	0	GeorgiaTech	0		-13.5	61.5	N
3:00 PM	NewMexicoState	0	UtahState	0		-30	56.5	O
3:00 PM	Stanford	0	California	0		2.5	48.5	N
3:30 PM	BallState	0	CMU	0		3	65.0	O
3:30 PM	NCState	0	Maryland	0		3	45.0	N
3:30 PM	UNLV	0	BoiseState	0		-28	53.5	O
3:30 PM	BYU	0	NotreDame	0		-13.5	41.0	U
3:30 PM	Alabama	0	Tennessee	0		20.5	56.0	N
3:30 PM	MichiganState	0	Michigan	0		-10	43.0	N
3:30 PM	Indiana	0	Navy	0		-2.5	60.0	N
3:30 PM	TexasTech	0	TCU	0		2.5	55.0	O
3:30 PM	FloridaState	0	Miami	0		20	56.0	N
3:30 PM	WesternMichigan	0	KentState	0		-4	54.0	N
3:30 PM	USF	0	Louisville	0		-6.5	52.5	N
3:30 PM	Pitt	0	Buffalo	0		11	55.0	N
3:30 PM	Rice	0	Tulsa	0		-21	63.5	O
3:30 PM	FloridaAtlantic	0	SouthAlabama	0		-3.5	45.0	U
4:00 PM	LouisianaMonroe	0	WesternKentucky	0		-3	54.0	O
6:00 PM	Colorado	0	USC	0		40.5	57.5	N
6:00 PM	MTSU	0	MississippiState	0		-20	56.0	O
7:00 PM	Georgia	0	Kentucky	0		27.5	58.5	N
7:00 PM	UNC	0	Duke	0		10	63.5	N
7:00 PM	Cincinnati	0	Toledo	0		7.5	64.5	O
7:00 PM	NewMexico	0	AirForce	0		-10.5	57.0	N
7:00 PM	KansasState	0	WestVirginia	0		-2.5	71.5	O
7:00 PM	SouthCarolina	0	Florida	0		-3	42.0	N
7:00 PM	Marshall	0	SouthernMiss	0		-2.5	66.5	O
7:00 PM	Idaho	0	LouisianaTech	0		-30	74.0	O
7:00 PM	IowaState	0	OklahomaState	0		-14.5	60.5	N
7:00 PM	Kansas	0	Oklahoma	0		-35	57.5	O
7:00 PM	EastCarolina	0	UAB	0		3	54.0	U
8:00 PM	UCF	0	Memphis	0		23.5	50.0	N
8:00 PM	Baylor	0	Texas	0		-11	81.0	N
8:00 PM	PennState	0	Iowa	0		-3	42.5	U
8:00 PM	Tulane	0	UTEP	0		-15	50.0	U
10:00 PM	Washington	0	Arizona	0		-7.5	61.5	O
10:30 PM	Utah	0	OregonState	0		-10.5	47.0	N
10:30 PM	Wyoming	0	FresnoState	0		-14.5	64.5	O
10:30 PM	SanDiegoState	0	Nevada	0		-7.5	66.5	N

[hutennis]

What is your average implied probability during those 3 weeks?

[hutennis]

Here is why I asked.

You have correctly predicted 34 out of 55 games during last 3 weeks.
That is 61.67% (average of 62,68 and 55)

If your IP of winning was 52.3%, which is more than reasonable assumption (it's like taking, on average -115 in -115/+105 line),
then the chance than you got to your results simply by being lucky is 10%.

When you see 10% chance of being lucky you just relax, don't get excited and don't pay much attention to the whole thing
b/c your results are simply not statistically significant. Most importantly, don't quit your day job.

Now, lets say you have correctly predicted results of 340 out 550 games with avg. IP of you winning = 52.3%

Now your chance to get there just by luck is pretty much 0.00 and you do have something to think about.
There are few more considerations, but you certainly right up there.

So, what is your avg. IP on those bets?

[sayhey69]

what kind of statistical model?

[hutennis]

what kind of statistical model?

I don't see how that would matter right now.
All that does matter is proper evaluation of current results.

[sayhey69]

well if the answer is univariate linear regression i think we can end any discussion at that

edit: also i was more interested if it really is a statistical model rather than "games where qb averages 300 yds and is played after 7pm go under" kind of bs

[hutennis]

Whatever model OP has used, he looked at his 3 weeks results and concluded

I believe that I am finally dialed in.

At this point, imo, the most important thing is to help him figure out where exactly has he dialed to.

[sayhey69]

lol i was gonna troll and make fun of that dialed in line but i decided to play it nice on this one. none of my college stat classes taught me how to get dialed in when creating models. i feel like i got ripped off

[jgilmartin]

My stats book addresses it in chapter 8.2, directly after chapter 8.1, "Getting in the Zone"


Just messing with you, OP

[bettilimbroke999]

Vegas heard about this thread and closed all their sportsbooks...they are very afraid

[Dark Horse]

Whatever model OP has used, he looked at his 3 weeks results and concluded

At this point, imo, the most important thing is to help him figure out where exactly has he dialed to.

Seems he dialed into the idea that it took him three years to have a three week run.

[thfootball]

What games are you betting? All of them?

[paranoyd androyd]

put headers on your chart next time to make things more clear

[usma1992]

For thank you for the IP comment. Though I am just giving you the results of the past three weeks. I have similar data from the previous year. Unfortunately, when I look back two years, I don't have enough data at the time the game was played. Which means, I could always use end of the year statistics but they are far from accurate and don't really apply to the point in time when Vegas created the Over/Under.

So the entire season from Week 5 on through Bowl Season... which approximates about 40*9 Weeks or so... totals 360 games.


As far as the "dialed in" comment. I have spent three years trying to figure out the relationship between off/def, how points are scored, how the game is extended... how matchups vs. lesser opponents drive the scoring one way or another.

I believe I finally have come up with a relationship that works and works consistently. I have posted the picks for this week and will continue to post. If you can't figure out the picks U stands for Under, O stands for Over, and N stands for Neutral bet. I think it is fairly self explanatory, but some might not understand.

What I find interesting about my model is that I have a window of 14 points +/-7 on the total calculation that I come up that I won't bet. This window seems very large. I am surprised that I'm betting as many games as I am.

For instance, by the time I got to place the bet on the WVU game this week it moved from 71 to 72.5 and it became a Neutral bet because it was within the neutral window. My system predicted that the score would be 79 approximately... 7 less than that puts it at 72... 7 more puts it at 86. So at 71 the system said bet over at 72.5 the system said don't bet at all.

Bottom line: I am taking a 14 block of points out of the equation and I am still coming up with 25 bets out of 55 this week. I find that very surprising but encouraging.

[ScreaminPain]

A simple formula that can save a lot of time. The tricky part is accurately estimating the correct yardage. Linear regression helps.....


(rushing yds./gm+passing yds/gm+opp. rushing yds/gm+opp. passing yds/gm/2)/7.5=total pts. expected for game.

42-13 last month alone.

[dominate.]

OP, did you make it out to the 20 yr reunion at homecoming last weekend? If I had known about this, I would have loved to talk to you about it.

A simple formula that can save a lot of time. The tricky part is accurately estimating the correct yardage. Linear regression helps.....


(rushing yds./gm+passing yds/gm+opp. rushing yds/gm+opp. passing yds/gm/2)/7.5=total pts. expected for game.

42-13 last month alone.

That's pretty impressive. Do you bet all the NCAAF games?

[hutennis]

For thank you for the IP comment...

You are welcome, but...

Giving the fact the you did not supply your IP numbers, you don't seem to understand the critical importance of it.
You are too fixated on your model's parameters and win loss record, which is meaningless unless you compare it with your mathematical expectations.

Also, the format in which you present your picks is confusing and irrelevant.
If you want serious consideration of your picks you should:
first, do it before games start posting your selection only (for eg: Team, Over 53.5 or Team -14)
second, follow up after games are played adding to the first table 2 more columns: Win/Loss and IP of your selection winning.

This way the process can be kept clean from data fitting, wishful thinking etc.
Soon enough your gonna see the truth about your efforts.

If you don't want to do it like that - you are not serious and the whole thing is just a waste of time.

[usma1992]

I appreciate your feedback...

About the whether I am serious or not about what I am doing is a little bit much... I've taken a few years off of work to do it... I have not only tracked every game and percentage I have have tracked what type of bet is winning because my program attacks each game for three different angles. My percentages I feel are finally rolling in because of my diligence.

I will share my picks and the percentages after the weekend... for today...right now... I plan to watch the SMU game...

Dave

[usma1992]

Tot Gam Bet	408
Perct. Won	58.82%
Ttl Net Units	72
Over 52.38%	80.0%

Week1	0	52.5%	$1,002	40	4	9	44.4%	8	14	57.1%	3	7	42.9%	1	2	50.0%	1
Week2	0	57.8%	$1,106	45	6	12	50.0%	12	18	66.7%	3	4	75.0%	1	4	25.0%	1
Week3	0	54.2%	$1,143	48	8	16	50.0%	8	15	53.3%	2	4	50.0%	6	10	60.0%	1
Week4	0	37.0%	$1,000	27	3	8	37.5%	3	11	27.3%	0	1	0.0%	4	7	57.1%	0
Week5	0	59.3%	$1,131	27	6	10	60.0%	7	11	63.6%	2	2	100.0%	1	2	50.0%	1
Week6	0	55.2%	$1,192	29	6	8	75.0%	8	15	53.3%	0	1	0.0%	1	3	33.3%	1
Week7	0	71.4%	$1,625	28	7	10	70.0%	10	13	76.9%	1	2	50.0%	2	3	66.7%	1
Week8	0	61.9%	$1,920	21	5	7	71.4%	7	10	70.0%	0	1	0.0%	1	3	33.3%	1
Week9	0	59.4%	$2,177	32	5	7	71.4%	9	17	52.9%	1	1	100.0%	2	4	50.0%	1
Week10	0	50.0%	$2,078	30	6	10	60.0%	5	11	45.5%	1	3	33.3%	3	6	50.0%	0
Week11	0	56.7%	$2,248	30	4	6	66.7%	6	10	60.0%	2	6	33.3%	4	7	57.1%	1
Week12	0	76.2%	$3,270	21	5	6	83.3%	9	11	81.8%	0	0	0.0%	2	2	100.0%	1
Week13	0	72.4%	$4,520	29	4	7	57.1%	12	16	75.0%	2	2	100.0%	1	1	100.0%	1
Week14	0	50.0%	$4,458	10	3	5	60.0%	2	5	40.0%	0	0	0.0%	0	0	0.0%	0
Bowl	0	72.7%	$6,190	22	0	0	0.0%	10	13	76.9%	1	2	50.0%	5	7	71.4%	1

[rainmaker82]

How'd this turn out today?

[usma1992]

my criteria. I made minimal adjustments to the program in order to get my percentage to 52%(after the fact) but it wasn't a great week.

I will post the picks again this week. I am not at all concerned about having a losing week or even a 40% week... If I am averaging 58%... I will take it all day long.

My minor adjustments... tried to account for games that were complete Blow Outs... the scoring comes to an abrupt halt in the second half. Several games played that scenario this weekend... I have to figure out how to model that.

Enjoy
Dave

P.S. I will post my picks probably tomorrow.

[sbearc820]

ScreamingPain,

How do you determine a play? Does the line need to be off a specific amount? I am very interested in trying to understand this formula. Does gm+passing yards mean average per game?

[trytrytry]

You are welcome, but...

Giving the fact the you did not supply your IP numbers, you don't seem to understand the critical importance of it.
You are too fixated on your model's parameters and win loss record, which is meaningless unless you compare it with your mathematical expectations.

Also, the format in which you present your picks is confusing and irrelevant.
If you want serious consideration of your picks you should:
first, do it before games start posting your selection only (for eg: Team, Over 53.5 or Team -14)
second, follow up after games are played adding to the first table 2 more columns: Win/Loss and IP of your selection winning.

This way the process can be kept clean from data fitting, wishful thinking etc.
Soon enough your gonna see the truth about your efforts.

If you don't want to do it like that - you are not serious and the whole thing is just a waste of time.

note to self..Go back and read all previous 550 posts by hutennis...

carry on...

[sapidoc]

What I find interesting about my model is that I have a window of 14 points +/-7 on the total calculation that I come up that I won't bet. This window seems very large. I am surprised that I'm betting as many games as I am.

For instance, by the time I got to place the bet on the WVU game this week it moved from 71 to 72.5 and it became a Neutral bet because it was within the neutral window. My system predicted that the score would be 79 approximately... 7 less than that puts it at 72... 7 more puts it at 86. So at 71 the system said bet over at 72.5 the system said don't bet at all.

14 points +/-7 .... are you adjusting this depending on the total? Or just filtering when off by more than 7 in either direction for all totals?

Is predicting a total of 35 on a 28.5 line different than predicting a total of 79 on a 72.5 line?

[09bear]

A simple formula that can save a lot of time. The tricky part is accurately estimating the correct yardage. Linear regression helps.....


(rushing yds./gm+passing yds/gm+opp. rushing yds/gm+opp. passing yds/gm/2)/7.5=total pts. expected for game.

42-13 last month alone.

What application does this formula have if you use the same numbers for defense?

(rushing yds allowed/gm +passing yds allowed/gm+opp. rushing yds allowed/gm +passing yds allowed/gm /2)/7.5=total points expected

I would think the avg of these two formulas would provide a more accurate estimate. Using the offensive numbers only doesnt account for defense.

[TR35]

Very interesting. I would like to see one example of this from start to finish with real numbers inserted. Ie Florida/Georgia game. Is that possible?

[TR35]

Quick thought: would it be more accurate using a teams last 3 games as opposed to whole season? Florida is much better offensively now than it was at the beginning and they are in conference play and not piling up yards against the sisters of the poor.

[chunk]

Short answer is no. Not that recent form shouldn't be taken into account.

[usma1992]

Hopefully, the positive news continues...

[bane]

Quick thought: would it be more accurate using a teams last 3 games as opposed to whole season? Florida is much better offensively now than it was at the beginning and they are in conference play and not piling up yards against the sisters of the poor.

Doesn't have to be either/or. Just use a weighted average, favoring more recent games.

[TR35]

What was this week's %?