Betting the NFL, Part 3

[reno]

I believe that the NFL is an easy sport to beat. When I lived in Vegas and moved money, I got the NFL plays of Chuck Sharp, a legendary Las Vegas sportsbettor who has since retired, with riches galore, to a life of golf, pina coladas, and hot 18-year old chicks in Thailand. Now, whether Chuck originated these plays or was just a big-time money-mover himself is immaterial. What is material is the fact that these plays won.
I would say these plays hit over 60% in the years I followed them. I don't know how these plays were generated, but I do believe that I know the major key to beating the NFL: Make a better mathematical line than the linesmaker.

As I stated in an earlier post, forget angles and trends. The key to beating sports is in the math.

I am now working with two computer programmers. Our goal is to systematically break down one sport after another, and then duplicate what Billy Walters did. The first sport that we have concentrated on is baseball. Thanks mainly to my input, which is based on thousands of hours of research I did, the basis for the program was established. What the programmers did, which was impossible for me to do by hand, was super fine-tune the program. For example, we know exactly how many previous starts is most predictive of a pitcher's performance in his forthcoming start. I have concentrated on scalping in baseball this year because our program was not finished until very recently. Although the program-in-the-making won, I did not feel confident enough to bet serious money. Now, we have run our program through our 8-year database. Over the past 8 seasons, the program, which averages 500--600 plays per season, hit 52.5% at a +117 take. It won every single year. This program is for sides. We are now working on totals, and should have something very soon. I can't wait to see how we do next baseball season.

The next sport that we will do is hockey. I don't know the first thing about hockey, and really have no interest in the sport. But, the mathematical principles that I applied to our baseball program will be the basis of our hockey research. The reason we are doing hockey next is because we believe it the sport with the weakest lines and the greatest opportunity, after baseball, to make money. We won't be ready this hockey season, but we will the following one.

Now, back to the NFL. Since the NFL offers relatively few betting opportunities, our development of a computer program for that sport is on the backburner. Although, we won't have a definitive system for betting the NFL until next year, or even the year after, I have done tons of research on the NFL, and if I weren't planning on vacationing during much of football season, I might seriously bet the sport just on the basis of what I know.

What I have done is an 8-season study of the various statistical categories that Pro Football Weekly uses to rate teams. In addition, I also studied what I call "creative stats," such as yards-per-point.

I did a correlation study between wins and losses and these stats to determine which stats were most significant in team performance. I now know exactly how significant every stat is. For example, total offensive passing yards per game and average yard per rush for an offense are very insignificant.

The 11 most significant stats, in order, were:

1) Point differential
2) Yards-per-point differential
3) Offensive points
4) Offensive yards-per-point
5) Average-gain-per-pass play diferential
6) Points against
7) Turn-over edge diffferential
8) Total yard differential
9) Average gain per pass play
10)Defensive yards-per-point
11)1st Down Differential

Now, these stats mean very little unless you can translate them into a betting line that beats what Vegas puts out. You can throw away point differential (points for and points against) because this stat, in and of itself, is grossly innaccurate over a short- term moving average. And in the NFL, you need to use a short-term moving average, say 4 to 5 games, to get an accurate read on a team. The key stat, and the stat that I had predicted would be the key stat prior to beginning my research, was yards-per-point.

Many astute handicappers are well aware of the value of yards-per-point as the ultimate "creative" stat for handicapping the NFL. Yard-per-point is such a great stat because it incorporates and summarizes what football is all about: production (yardage) and points (scoring). Yard-per-point, in my mind is the stat that best allows a handicapper to get an accurate statistical read on a team over a short-term basis.

Just as Jim Jasper's now out-of-print book "Sportsbetting" got me started on my baseball research, another out-of-print book, "The Winner's Guide to Pro Football Betting," by Art Glant and Leigh Cohn, got me started on my football research. This book preaches the value of yard-per-point and has a formula for making a betting line, using a 4-game moving average, with this stat. Glantz and called this method the "Dudley Formula." According to him, (note: the book was writen in 1983) prior to 1983, the formula had hit over 70% winners against the spread over the previous 10 seasons, and never had a losing season.

Anyway, it would be a bit involved to explain Glantz's formula and how he applies it against the betting line. Furthermore, his formula, from my perspective, is incomplete.

To summarize: yards-per-point, in my opinion, is the stat that serious handicappers should focus on if they want to beat the NFL.

[Count Zero]

I think I disagree, but I will check my numbers before I risk looking stupid in the eyes of a fellow marin landsman. I know for sure I disagree with your first sentence. I too have examined what I think is every stat my self-written, interface-free database is capable of generating, relative to performance ATS. I found yards per rush, first downs, yards per pass, and a few other somewhat more problematic things to be predictive. I recall having high hopes for yards per point, but did not find it to be predictive in recent years. I know it is a highly-regarded measure; maybe I am using it improperly, though I recall having tried quite a few things. I will redo enough of this research to be able to make a more cogent statement as soon as possible.

Should mention that I also profited from the Jasper book(s), helped get me started back in the Commodore 64 days. I use my programs to calculate several sets of power ratings and to research scheduling situations, etc. Have enjoyed your posts enormously, as you are strong where I am weak, but will be surprised if I am far off base on this one.

By the way, where's Part 2?

[Chilly]

Reno (and all),

I've lurked her for quite some time, and finally decided to register/post. Your approach to statisticall modeling sports is similar to what a friend and I try to accomplish. We have had decidely mixed results. The main problem in our view is the quality of data used to make predictions. I think the most important aspect of the data is the line/total that you use. We have a decade of data for most sports, but recent years we've collected the line data ourselves from the books that we use (and these change frequently). The older years were purchased from JR Miller, taken from USA today, CSI, and other sports sites with historical data. Needless to say, some are opening lines, some are vegas standardized closing lines, and others are off-shore opening, or after major line movements.

The point is that if you make predictions based on the lines in any way, you future performance will vary, almost assuredly in a negative way (additional error that is introduced will tend toward randomness, which is 50/50, and unprofitable). For example, In baseball, off-shore lines used to vary significantly from vegas lines, as favorites were really bet up off shore. This made some teams dogs where in vegas they were Pickems; some dogs went from +110 to +125, etc. If you assume the line represents probabilities in perception of winning, and try to incorporate that in categories of plays that you make, you've included plays that would not have otherwise met the criteria in the historical data. Those games would have non-plays in the prior years that your predictions made +100 units. When you play baseball for 1-10 sides a day, and add in 1 miscalculated game a day becuase of line data, you quickly find out that your margin of error is smaller than you thought.

Can I ask what line data you use? Is it consistent with the line data you use to make predictions?

Would like to talk about historical data some more...

Chilly

[gasman]

Points differential is significant in the sense that 100% of all NFL winners score more points than their opponents.

That is, however, what you're trying to predict - the "Y" variable" - so you can basically throw that in the trash. The statistical term for reversing the order of the variables within an equation is endogeneity.

Yards per pass play, on the other hand, as well as yards per point to a lesser extent ARE superior indicators of performance in the NFL and I would point any handicapper interested in bringing mathematical models into their repertoire to start with those two indicators first.

It's also important not to count variables twice. Teams with poor turnover differentials are likely to have poor yards per point numbers and vice versa.

[Chilly]

Also, what people seem to be talking about is explanation versus prediction. Sure, i can look at yards per point, and figure out that's WHY THEY WON, but it doesn't help you figure who will cover the spread. Just because variable significantly predicts performance does not mean it is of any use in predicting spread performance, and my beleif is that traditional variables are useless for prediction ats if operationalized the standard ways; you need to think counter-intuitively if you want to use offense/defense stats.

BTW: it doesn't matter if you use multiple indicators of theoretically important variables, like yards and turnovers: unless they are perfectly correlated, their utility will be measured empirically. Followed to its logical conclusion, all you need to knw about a bad team is that they are bad, because surely all the stats will bear this fact out...

mc

[chooch]

Ah,the search for the "Holy Grail", i.e. finding the specific numbers which correlate to victory ATS.

My trek started several years ago, took a part time gig play testing a NFL computer game.
The manufacturer wanted realistic statistical results and I'd submit a mountain of reports comparing NFL actual percentages vs. "game" output.

Constantly studying, evaluating and assimilating this data (kinda like the Borg),led me to some strong observations regarding, "what it takes to be successful".

In time, the little software co. decided the market for graphically pleasing, joystick type games was more lucrative than nerdy text based simulations.

Oh well, it was more a labor of love, my income is derived from hustlin' box cars around in the middle of the night.

I acquired point spread results (back to '76) and with my trusty TSN Pro Football Guides, built a database (over 3 years of data entry) which is unique and without equal.

Turned my theories and observations into "wagering formulas", proceeded to extensively test using '76-'92 results.

Sorry, several distractions here this a.m., doing a lot of starting and stopping with this post (losing my train of thought).

Besides my "cornerstone" formulas, in recent years I've added to my arsenal, using ideas gleaned from the pages of "Pro Football Reavealed- The 100 Yard War" (Stats Inc.).

My daughter, the tennis player, she's been watching the open and is all fired up to play...gotta go.

Are there any good long term trends, someone asked?

Just finished researchin' one of the best, league wide, trial rich, purely objective angles, I'VE ever seen.

Without regard to anything subjective (inj.'s,motivation,etc.),
1976-1998

226-142-16 61.4%

And consistent, record after any 0-1 wk (38 times),0-2 wk (9),0-2-1 (1), 0-1-1 (1).

Again,ATS week after above 49 wks, 49-16-3.

-chooch

[Count Zero]

You are going to mention this without telling us about it?

[ArticExpress]

Reno:

Guess I'll unlurk and post.

Reno, I'm somewhat surprised to see this here, but I must comment on your 11 "most significant stats." I'll agree with your number one, but YPP is 6th down my list (with a corr. of approx. +.78 to +.85 depending upon the year). Your number 7: turn over edge dif, should be broken into ints and fumbles lost (ints is around -.45 to -.5, and fumbLst is -.30 to -.40) and rank around 16th to 25th depending on the year. Frankly, I'm not interested in revealing to the world at large what the rest are or how to use them, but you are right in that the transformation of the data into a point line is the key to use.

By the way, Glanz's YPP method basically quit working well after the '84 season, but variants of his method are still usable.

If you are interested in any more conversation in this area, I'll be happy to take it offline as am I really not interested in revealing any more info.

I can be reached at:
[email protected]

AE

[reno]

Count Zero, Part 2 is a few days back. It was meant to be entertaining rather then informative, and no one responded to the post.

Chilly, For our baseball research, we used the overnight Computer Sports World odds for the past two season, and for the 6 seasons prior to that we used the Stardust opening line.

I completely agree with gasman who realizes that yards-per-pass attempt and yards-per- point are excellent stats to utilize for handicapping purposes. The key is figuring out how to convert these stats into a number that beats the betting line. What one needs to do is create formulae that convert these numbers into lines and run them through a database on a cumulative basis (to avoid back-fitting) over multiple season and see how they do versus the betting line.

Yard-per-pass is an extremely important stat because pro football has evolved into an aerial circus. Teams that can make the big plays and stop them are teams that are going to tend to cover the spread more often than not. Yard-per-point is great stat because it at once accounts for points, yards, and turn-overs in a single stat.

The other stat that I would consider using is 1st downs. Ernie Kaufman, a well-know tout, used to peddle a handicapping system based on 1st downs. The problem with Ernie's system was that it equated 1 point to one 1st down. A virtual constant in the NFL ,from season to season, is that one 1st down equals 1.1 points. Consequently, this is a very convenient and easy stat to use for handicapping purposes, even though it is not as strong as yard-per-pass attempt and yards-per-point.

[reno]

The problems with Glantz's system include: 1) the system only uses offensive yards-per- point and defensive yards and ignores defensive yards-per-point and offensive yards. 2) It doesn't take into account recent strength of schedule. I agree, Glantz had a good idea, but it needs to be refined.

Artic Express, thanks for input. I'll file your e-mail address for future reference. Since we're not actively working on football now, it would be premature for me to contact you at this time. What's so great about all these ideas is that one actually test them
over multiple years. The art is coming up with the formulae that converts the key stats to lines, and the science is testing them against the spread.

[Count Zero]

Gee, this is great. Let me see if I've got it straight -- I take yards-per-point and use it in the way that everyone says is great but no one cares to specify. Then I add in the many, many other predictive stats that Arctic Express can't reveal. Finally, I modify the resulting line according to the fabulous angle developed by Chooch that he doesn't actually mention. That's my final line.

Wow, the NFL really *is* an easy sport to beat!

[WhiteShoes]

one key to making a line I think is doing more than mearly taking final scores and developing a power rating and thus, ignoring
the stats that lead to scores. This is why
many simply regression formulas don't facilitate accuracy because the reasoning behind the different connections (between stats) might not make any sense, or as someone pointed out, often double handicapp.

it's often hard to explain, you just have to try making a line and then examine why it doesn't work, fix it, and then try to improve it some more. cloudy enough for ya.

[wintermute]

A few points.

1) No sport is easy to beat. How can it be with the books taking 5 percent or more out of the pot. Over the long run no more than 1 or 2 percent of us are going to be winners.

2) That being said, if someone finds a winning system or edge, why are they going to blab all the gory details here? They lose the advantage they've found against the losing 98 percent. Just to know that a game is beatable is encouraging news.

3) Handicapping is overrated. If you poll the people who are actually winners you'll find that most of them are winners primarily because they shop around for lines, they don't bet favourites and they take advantage of scalping opportunities whenever they can.

[ajax50]

IM SUPRISED AT YOU GUYS.ALL THIS TALK ABOUT COEFFICIENT,REGRESSIONS,STANDARD DEVIATIONS IS FOR THE PHYSICSISTS.THIS IS FOOTBALL WERE TALKING ABOUT.FIND A GUY WHO PASSES OUT THE FOOTBALL SHEETS AT LOCAL OFFICES AND GET THE GAME RATIOS FROM HIM.THE MORE SKEWED TO ONE TEAM THE MORE I BET ON THE OTHER.IF 149 PEOPLE HAVE JAX AND 13 HAVE CAROLINA THIS WEEK THAN LOAD UP ON CAROLINA.

[gasman]

I think we've found a Contrarian. AJAX, I wouldn't disagree that no one seems to go broke fading the public but at its most fundamental level, contrarianism is still a one-dimensional form of handicapping.