Using Hilton Contest as evidence.
I'm surprised he didn't do this study: past winners or top finishers in later years.
To me, the flaws of his analysis are obvious, but I'll let the forum go off on it.
-
Using Hilton Contest as evidence.
I'm surprised he didn't do this study: past winners or top finishers in later years.
To me, the flaws of his analysis are obvious, but I'll let the forum go off on it. -
Thanks for pointing to the article, it was quite interesting.
We can always quibble about any study. The technique used, the data selected, and the conclusions drawn. On balance, I thought he did a good job on this one and his conclusion was reasonable (and likely very true).Comment -
The NFL is not beatable.
Why you think the books let you bet so much on these games?Comment -
No idea why fairly smart people assume you can measure skill (or prove it doesn't exist) by looking at a tiny sample -- or, even given a proper sample, why they assume that bc most of population doesn't have a certain skill/tendency/bias, none of the population can have it. He could have said, "most people cannot beat the NFL." Instead, he told every reader that the reader himself cannot beat the NFL. Even if Pomeroy has never bet a game in his life, that blog post was embarrassing. Reminds me of the "Why you shouldn't bet on baseball" abortion.Comment -
KenPom Article NFL Betting
You have no skill at betting on NFL games
11.09.11
I’ll bring this back to college hoops eventually. But first, I’ve got data to show that betting on NFL point spreads is like playing the slots in that you have no control over whether you win or lose. I know, you think you are the exception. You study the trends and wait for Donny Gridiron’s 10,000 dime play and you win money all the time. And I surely can’t prove that you, specifically, don’t have magical predictive ability. But I’m nearly certain you don’t.
I have come to this conclusion using the results of the 515 participants in the Las Vegas Hilton’s Supercontest. The rules of the contest are fairly simple and provide a great experiment on whether there is skill in betting on NFL games. Each entrant pays $1500 to participate before the NFL season begins. Each week the contestants pick five games against the point spread. At the end of the season, the participant with the best record wins a pile of cash and everyone in the top 20 wins something.
There’s an idiosyncrasy in the contest worth mentioning as well – the Supercontest permits time travel. The participants use the point spreads posted as of Wednesday morning, but get until Saturday to make their picks. Thus, if information becomes available between Wednesday and Saturday, it will move the point spread in the casino but it won’t affect it in the contest. This provides the participants with an advantage that the regular gambler doesn’t have.
Because the entry fee is $1500, I think it’s a reasonable assumption that each participant thinks they are better than average at picking games. The great thing about the Supercontest is that the results are public. The Hilton releases all of the picks that each person makes and updates the standings each week. This provides us with information as to whether these participants really are better than average.
Keeping in mind that these folks think they know what they’re doing, they get to travel back in time, and they get to pick the five games they feel best about, the results are underwhelming. Through week nine, the group’s picks have been winners 51.0% of the time. For those not hip to sports betting customs, one needs to win 52.4% of the time to break even. Even using a time machine, people are not able to beat the casino. And if you remove the time travel option, these folks are probably doing no better than if they flipped a coin.
This is not proof that you, specifically, are unable to make money betting on NFL games. Nor is it proof the person behind the entry named “Iced Tea”, who is 32-13 on the season and shares the lead in the contest through week nine, has no ability either. However, there is compelling evidence that suggests this is also the case. What we need to know is if there are people in this contest that do have skill and if their ability is being masked by the poorer players in the group.
There are different ways to attack this, and the one I’m using is simple. Each week, let’s take the aggregate record of the players in the top 25 (including ties) of the standings and compare them to the players in the bottom 25 of the standings. If there’s any skill in betting against point spreads, the best players should rise to the top and the worst players to the bottom. And it follows that the best players should perform better than the worst players.
Furthermore, with each successive week, the worst players have much less incentive to perform well. The worst players have lost their chance to win big money, while the best players have large sums of cash on the line. The players at the top of the standings are not only better (theoretically) but they have incentive to do more homework to make the right picks.
Given that, the results are interesting. Let’s look at how the top 25 and bottom 25 in the standings have performed since week five. To clarify what these records represent, we are looking forward. So the week five results represent how those that were ranked in the top 25 heading into week five did on their picks that week.
Wk Top 25 Bottom 25 5 76-60 .559 74-70 .514 6 77-53 .592 90-65 .581 7 74-76 .493 82-68 .547 8 66-74 .471 60-65 .480 9 83-67 .553 84-91 .480 376-330 .533 390-359 .521 </PRE>(Keep in mind because I am including the top 25 and ties, there are varying number of people in each group for each week.)
There is a difference in the performance of the top 25 over the bottom 25 over the past five weeks, but it’s insignificant and less than half of a standard deviation. (If I was more devious I could have just removed week five and tilted the results towards the bottom 25, but I try not to be like that.) The folks that no longer have incentive have missed one more game out of 100 than the folks still fighting for a big prize.
To expand the results further, the bottom 25 has outperformed the rest of the field by a winning percentage of .521 to .506 during this time. There’s no significance to that either. It’s all noise. Given that 515 people flip a coin 45 times, you’d expect 3 of them to get tails at least 32 times. There are two that have won at least 32 games in the Supercontest if you include ties as half-wins. So assuming a random process provides a pretty good estimate there, also.
This is pretty strong evidence that betting on NFL is like playing the slots. I can’t prove that Iced Tea has no skill specifically, but in any group of 515 people, there are going to be some that appear to be really good (or bad) at guessing coin flips. That’s essentially what is happening here and why Vegas is what it is. The Hilton is brilliant to put on the Supercontest because it gives the illusion that certain people can win a significant majority of their picks.
I’m not suggesting people become rational and stop betting on NFL games. (Not that anyone would care if I did.) It’s the power of the masses that makes the NFL point spread the best forecast of the outcome. The paradox here is that it’s because so many people think they can win money that nobody can. Or at least nobody has the skill to do so. People make money on the slots all the time, but in the long run, it’s also a losing endeavor. The reality is that you can save your energy researching things like how the Chiefs have done against the spread in their last ten games against the NFC North and just flip a coin. The results of the Supercontest indicate that you’re as likely to get rich that way.
Some of you may see the connection to college hoops analytics, but I’ll expand on this soon.Comment -
The logic behind the analysis is sound but the sample size is too small.
With that said the market efficiency in the NFL is the highest amongst US sports and it should be no surprise that the lines are sharpest and limits the highest.Comment -
The higher the sample size the lower the % gets and most likely hover around 49-51%
If you put a huge group together every year the % will be the same, kind of crazy but makes sense.Comment -
I think the weakest part of the his argument is that entering a $1500 contest somehow presents evidence that you have a set of sharp ATS bettors that he can draw conclusions from... Ignoring that playing contests correctly will often lead to you have to do things you wouldn't be willing to make a bet if it was an isolated event why would someone who was able to conclusively beat NFL numbers even bother entering such an event?Comment -
I guess Fezzick was just a lucky son of a gun to win the Hilton,Two years in a row.Comment -
He also didn't mention that everyone in the contest must make 5 picks. What if you wouldn't put your money on but 2 of them that you felt sure of?! So picking more games gives it more of a chance to balance itself out.Comment -
Actually, I'm hoping KenPom's ideas gain a following.
...the more folks who think it can't be beaten, the better off we are.Comment -
all you got to do is pick the straight up winner. covers the spread 80 percent of the time or whatever it is.Comment -
Hilarious.
I love the initial assumption that the group is composed only of above-average handicappers because of the $1500 entry fee. Obvious correlation... Then the later assumption that a bettor who's ranked too far down to still compete for the prizes won't try as hard (but apparently will still bother to send in his picks every week).Comment -
His evaluation is flawed on so many levels I'm not going to waste my time.Comment -
@evo34
One more thing I found "interesting" in your rant.
You must be considering yourself at least "fairly smart" and I don't want to insult your intelligence or anything, but I do have to point out something rather elementary.
No one has to prove negative.
If I say that based on available data there is no Loch Ness monster, Big Foot, flying saucer
or people who can beat NFL by skill alone I don't really have to prove it.
The burden of prove is on another side. And it's pretty easy.
All another side has to do is to produce scientifically acceptable evidence. Thats it.
All I have to do while making a negative claim (if I want to avoid to be marked as a proven idiot) is to make sure that my negative claims are not laughable. Bc if they are, i will be easily proven wrong. Right?Comment -
I'm tempted to agree with him. The issue with NFL versus ncaaf or ncaab is that you have so few games. It takes a couple of seasons before you can accurately quantify how well you're doing. If you choose to bet only locks/shoe-ins then the time you do spend finding those would be better spent capping ncaaf or a smaller market. If the goal is to maximize EV I don't see how NFL pays off versus the alternatives.
Would like to hear opinions from some NFL experts about this.Comment -
The Freakonomics guy did a very similar study, a high-buy-in picks contest, and as I recall his conclusion was not that players had no skill, but that the sportsbook does a better job of estimating the outcome of a game than any single player.Comment -
If you want to be scared of trying to out-analyze the NFL market, that's your choice -- a choice that has zero impact on the few people that do out-analyze the market. And, yes, when do observe something consistently over a sufficient sample and time period, you can start to predict that future results will be non-random -- i.e., that that information contained predictive value. It's this magical scientific principle that led to weather forecasts.Comment -
Your stubbornness is laughable.
Laughable negative arguments, like "Sun does not rise from the East" or "Law of gravity does not work in New York City" are predictably and easily destroyed. Thats why they are laughable.
In order for "No one can beat NFL by skill alone" argument to be laughable you should be able to correctly predict that you will win ATS every time you bet. Can you do that?
Hey, what the hell. Let me make it easy for you.
Can you correctly predict that you will win 5 Hilton contests in a raw?
Can you do that?
What about 4? 3? 2?
Can you correctly predict anything NFL related drunk monkey does not have a chance to do just by throwing darts at the sports tables?
I know it's insulting for our egos to admit that with all brain power and sophistication we posses we are still no match for randomness and randomness does rule the world (speculative areas of it for sure), but hey, what can you do?Comment -
derpComment -
You have a complete misunderstanding of the argument. There is no need for anyone to be able to "correctly predict" that they will win a contest in order to demonstrate that they have skill. Can Albert Pujols correctly predict that he will win a home-run hitting contest 5 times in a row??? No, but that obviously doesn't mean he doesn't have any skill when it comes to hitting home runs.
The title of Kenpom's article is "You have no skill at betting on NFL games" -- that is a ridiculous statement to make on its face. There is no way to PROVE or DISPROVE it based on any statistical test, no matter how large or clean the sample (and the sample he uses is small and dirty).
In reality, any OBSERVED outcomes reflect some combination of skill and luck; the question is how much skill and how much luck. Now it may well be that it's dominated by luck, but there is no way to "prove" that it's 100% luck and 0% skill, which is what Kenpom is claiming. The best he could possibly do is to claim that, based on his observed sample, the variation in skills at predicting NFL games is very small across bettors -- in other words, that most people are "about the same" in their prediction ability, in which case it's not possible to attribute any differential skill at all to any one bettor. But the fact that you can't attribute such skill doesn't mean that it doesn't exist -- it's just that you can't observe it because it's drowned out by the noise in the population. That is very different from claiming that no one has any skill at all in predicting NFL games.Comment -
Actually, I'm surprised to hear you say that. Seems a cool guy from the frequentist camp would find a way to "statistically prove" the winning population of Hilton pickers is different than the "average" 50% hamburger pickers, or not.
Here's how a dumb-shit number-fumbler like myself would think about it: The contest has been in place since 1987 (I was in that one), and given that you would have a binomial dist. of the average pickers, one could get the average results on the winner's tail of the dist. for all the contests over the years given that you know how many contest entrants showed up. If you then examine the tail of the actual results of the winners in the Hilton, say the top 20 guys, and if the tail is "thicker" i.e. more guys bin up in the tail, I'm sure a frequentist like you could search around in the box of tests you guys like to carry around and you could find some test that could show more guys showed up in the winner's tail than is normal (or binomial), or not to some statistical p value.
And then if you did that, you could publish it, and attempt to be cooler than KenPom.Comment -
I think you are confusing apple and oranges, not me.
Why are you pulling this argument into non speculative areas?
Concert pianist does not need luck to play Rhapsody in Blue and your family dentist is not dealing with uncertain outcomes when he drills your rout canal.
And yes, Albert Pujols can correctly predict that he will win a home-run hitting contest 5 times in a row if the rest of the guys hitting balls will be posters on this forum.
So he can predict the result and easily be correct in his prediction by outperforming fellow human beings bc he has a superior set of skills and yet no one, according to Kenpom's article, can outperfom drunk monkey betting NFL games, bc no one, according to Kenpom's article, has set of skills superior to that of drunk monkey.
Again, please... Don't be ridiculous!The title of Kenpom's article is "You have no skill at betting on NFL games" -- that is a ridiculous statement to make on its face. There is no way to PROVE or DISPROVE it based on any statistical test, no matter how large or clean the sample (and the sample he uses is small and dirty).
Negative statements does not required to be proven and thus can not be ridiculous unless some one who claims them to be ridiculous can prove that they are.
And that brings us to this:
It's your question to answer, not his.In reality, any OBSERVED outcomes reflect some combination of skill and luck; the question is how much skill and how much luck.
And until you have a proven right answer, 50/50 chance stands correct.
He doe's not need to prove anything. If you don't agree prove him wrong.Now it may well be that it's dominated by luck, but there is no way to "prove" that it's 100% luck and 0% skill, which is what Kenpom is claiming.
That's how this type of argument resolved in science.
Yeh, yeh I know.The best he could possibly do is to claim that, based on his observed sample, the variation in skills at predicting NFL games is very small across bettors -- in other words, that most people are "about the same" in their prediction ability, in which case it's not possible to attribute any differential skill at all to any one bettor. But the fact that you can't attribute such skill doesn't mean that it doesn't exist -- it's just that you can't observe it because it's drowned out by the noise in the population. That is very different from claiming that no one has any skill at all in predicting NFL games.
The fact that no one ever seen a flying reindeer does not mean there is no one.
Correcto mundo!!!
Who knows, maybe it's out there. I just would not bet on it.Comment -
It won't give you a binomial distribution because of the structure of the handicapping contest. As the end approaches, those in the middle to lower half of the field, or people near certain payout levels will begin picking riskier bets hoping to make up ground, people "in the money" will pick safer bets, and the worst bettors will stop picking entirely. That will skew the results if you look at the contest afterwards. You'll see bunching near the top and at the big payout levels, and then a long tail of people who stopped playing because they were too far behind or were really risky at the end and lost.Comment -
Correction. From the actual paper that studied picks made during an offshore NFL Handicapping contest (which is a great read if you haven't read it) :
... there is little evidence that
there are individual bettors who are able to beat the bookmaker systematically.8
The distribution of outcomes across bettors is consistent with data randomlydemand equilibrate prices.
generated from independent tosses of a 50-50 coin. Moreover, how well a bettor
has done up to a certain point in time has no predictive value for future performance.
Finally, the evidence is mixed as to whether aggregating bettor preferences
has any predictive value in helping to beat the spread. In my sample, there
is some weak and ultimately statistically insignificant, evidence that bettors are
more likely to predict correctly when there is agreement among them as to which
team looks attractive. Altogether, the results are consistent with the conclusion
that the bookmakers are at least as good at predicting the outcomes of games than
are even the most skilled gamblers in the sample and the bookmakers exploit their
advantage by strategically setting prices to achieve profits that are likely higher
than would be possible if they simply acted as market makers letting supply and
From: http://pricetheory.uchicago.edu/levi...arkets2004.pdf
Comment -
Amen.Comment -
Well, I'm not really interested in the loser's tail, only the winner's. And I doubt that anyone is making safe bets as you can completely fall out of the money in one week, so everyone must continue to press.
The point is, if the actual winner's tail is better than the winning tail of a set of 50% picker binomials over the years, given the number of initial folks that played; that's good enough for me, but I could really give a sh*t. But regardless, if that is the case, it would go some way towards saying that some skill is actually involved.
BTW, I believe the serious players typically don't play in the contest anymore. I seems it's mostly touts, wanna-be touts trying to build a rep, and joe six-packs, sprinkled with a few serious players who live in town with time on their hands. So in the last several years, realizing how much luck is really involved, the better players have stayed away. Given that, the winner's tail may be thinner than one might think, and KenPom might be proven "right", but for a wrong reason.Comment
