MLB Data Analysis

[durito]

Hilarious thread. Would read again.

Makes you want to go to grad school, doesn't it.

[MonkeyF0cker]

Just because the guy can't talk the talk doesn't mean he isn't making money and won't continue to make money. He has said some strange things about "60% plays" and not treating betting like the stock market, but there are plenty of guys here who can talk EV and Kelly but are obviously squares or $5 losing bettors (Wrecktangle has got to be the #1 example). Many of you look up to Justin7 who had to hire a PhD mathematician to run a probit model for him -- something mesadog is handling all by himself. I think many of you guys are jumping on him for using the word "system", which for him is not a system as most use the term = betting based on a silly non-causal trend. His method is seemingly based on solid econometrics, but he may have a strange rule of thumb for when to pull the trigger. That's why I suggested he needed more experience and not more data.

There are a lot of sharps who don't calculate their EV but bet based on whether a line is off by more than x cents. They could calculate their EV if you hassled them but they choose not to. They just know it is +EV by some amount they are comfortable with and bet something they are comfortable with (so they don't need to know the EV to bet according to Kelly), or they don't need to know EV because they can't get down as much as they'd like to anyways or they don't want to attract attention. And when you're in a daily MLB grind for 6 months it can save time.

I agree with you on a lot of points, but this guy is also fishing for people to help him. If Justin can write a book on modeling, it's certainly within reason that this guy can throw out a few terms that he's read here; like "my probit model is hitting 60%." It looks more to me like he's trying lure in some poor suckers to help him create a model than actually having a viable model intact already. He stated that he's been a lurker in the forum for quite a while. If he's been a lurker, how can he not comprehend calculating his edge or anything that people are asking him? There are countless threads on those topics here.

I was actually thinking he sounded exactly like Wrecktangle. All he needs to do is misuse the term "Bayesian prior" a few times and they'd be twins.

[evo34]

Your "question" has a pretty obvious answer: if you don't have enough data to come up with a reasonable estimate of your edge, you don't have an edge. The OP doesn't have enough data to figure out if he has an edge (i.e., if his results are non-random); more importantly, he doesn't even understand the very concept of edge.

Uhoh. I sidestepped into uncomfortable territory. Therefore it must be unrelated.

My question is valid, in the real world of sports betting. A favorite sect mantra is to gloat that people can't calculate their edge precisely. So let's hear it from these geniuses. They either sidestep everything related to smaller sample sizes, or they make themselves look a lot better than they really are.

[Maverick22]

lol.... how long before one of you guys decides to grab the ruler?

[Philphan]

Mesadog, I am an applied math student looking to help. PM me

[mathdotcom]

I agree with you on a lot of points, but this guy is also fishing for people to help him. If Justin can write a book on modeling, it's certainly within reason that this guy can throw out a few terms that he's read here; like "my probit model is hitting 60%." It looks more to me like he's trying lure in some poor suckers to help him create a model than actually having a viable model intact already. He stated that he's been a lurker in the forum for quite a while. If he's been a lurker, how can he not comprehend calculating his edge or anything that people are asking him? There are countless threads on those topics here.

I was actually thinking he sounded exactly like Wrecktangle. All he needs to do is misuse the term "Bayesian prior" a few times and they'd be twins.

Interesting points sir

[Dark Horse]

---

[Dark Horse]

Your "question" has a pretty obvious answer: if you don't have enough data to come up with a reasonable estimate of your edge, you don't have an edge. The OP doesn't have enough data to figure out if he has an edge (i.e., if his results are non-random); more importantly, he doesn't even understand the very concept of edge.

There are, at least, two types of approaches to sports betting. One involves large sample sizes, the other does not.

1) Timeless models produce large samples and continue season after season. Preferably this type of work has a high degree of originality, because that puts it out of reach of competitors. When the bar is set very high, this may produces the stuff for the ages that adds to or redefines probability theory.

2) Smaller sample sizes, that are just as real, but won't last from season to season. These shorter term fluctuations are available to those 'surfers' who know how to read the waves. A true skill in itself. This field almost certainly will not include the rigid thinkers of the math sect, who typically confuse this profitable area with data mining. I don't know why these more easily recognizable fluctuations tend to last for a season only, but I know that they do (perhaps it has something to do with the 'group mind' of all the athletes in a league). They don't require huge amounts of research, but remain unidentified without a relatively high degree of awakeness. Bettors can profit from these shorter term fluctuations with full awareness of their temporary nature.

Needless to say, bet sizing for 1) is very different than for 2).

As to the math sect. It doesn't take a rocket scientist to figure out that their thought diet contains little or no original thinking. It if did, they would be far too engaged with their own constructive ideas to bother correcting every imprecise thought uttered here. And then gloat collectively over their perceived superiority. lol

[MonkeyF0cker]

As to the math sect. It doesn't take a rocket scientist to figure out that their thought diet contains little or no original thinking. It if did, they would be far too engaged with their own constructive ideas to bother correcting every imprecise thought uttered here. And then gloat collectively over their perceived superiority. lol

2x + 3 = 13

Solve for x.

Show us some originality, DH. (The answer can't be 5 though, because that wouldn't be original.)

[evo34]

There are, at least, two types of approaches to sports betting. One involves large sample sizes, the other does not. 1) Timeless models produce large samples and continue season after season. Preferably this type of work has a high degree of originality, because that puts it out of reach of competitors. When the bar is set very high, this may produces the stuff for the ages that adds to or redefines probability theory. 2) Smaller sample sizes, that are just as real, but won't last from season to season. These shorter term fluctuations are available to those 'surfers' who know how to read the waves. A true skill in itself. This field almost certainly will not include the rigid thinkers of the math sect, who typically confuse this profitable area with data mining. I don't know why these more easily recognizable fluctuations tend to last for a season only, but I know that they do (perhaps it has something to do with the 'group mind' of all the athletes in a league). They don't require huge amounts of research, but remain unidentified without a relatively high degree of awakeness. Bettors can profit from these shorter term fluctuations with full awareness of their temporary nature. Needless to say, bet sizing for 1) is very different than for 2). As to the math sect. It doesn't take a rocket scientist to figure out that their thought diet contains little or no original thinking. It if did, they would be far too engaged with their own constructive ideas to bother correcting every imprecise thought uttered here. And then gloat collectively over their perceived superiority. lol

So, after the fact, the "shorter term" trends that happen to have worked out should have been recognized beforehand as legitimate? As long as you knew how to "read the waves?" Sounds about as legit as a horoscope.

[Dark Horse]

You're simply not practiced in this shorter term field. I could tell you that learning to surf (waves) takes time, but by your reasoning that would not make sense.

As indicated, field 1 of serious research eventually gets the scientific credit. But field 2 is real as well, even though it may be simplistic by comparison. If you don't wish to believe it that's fine with me. There's no scientific credit to be gained from these simple approaches; the reward here is cash. Wouldn't it be a mistake, when looking for the Grail, to overlook golden nuggets that are hidden in plain sight?

[Wrecktangle]

I was actually thinking he sounded exactly like Wrecktangle. All he needs to do is misuse the term "Bayesian prior" a few times and they'd be twins.

This from a person who thinks Kelly Criterion is for everyone, game theory is useless in sports, and Monte Carlo modeling is the pinnacle in sports modeling.

Here again we have a new guy who is ripped apart by the wolf-pack, simply for sport. This forum has turned into an unmoderated free-for-all where unwarranted attacks on anyone (especially new folks) is tolerated and thusly encouraged by the "moderator(s)."

[MonkeyF0cker]

This from a person who thinks Kelly Criterion is for everyone, game theory is useless in sports, and Monte Carlo modeling is the pinnacle in sports modeling.

Here again we have a new guy who is ripped apart by the wolf-pack, simply for sport. This forum has turned into an unmoderated free-for-all where unwarranted attacks on anyone (especially new folks) is tolerated and thusly encouraged by the "moderator(s)."

Never have I said any of those things. Certainly, you could provide some links to those remarks.

You're right though. The "attacks" were completely unwarranted. Someone who states that they have created a profitable probit model but is incapable of articulating their edge and answering any pertinent questions about it in their quest for help from other posters seems perfectly legitimate. I think you might just be bitter that nobody has answered your lobbying for help as well. Two peas in a pod.

[Wrecktangle]

Never have I said any of those things. Certainly, you could provide some links to those remarks.

You're right though. The "attacks" were completely unwarranted. Someone who states that they have created a profitable probit model but is incapable of articulating their edge and answering any pertinent questions about it in their quest for help from other posters seems perfectly legitimate. I think you might just be bitter that nobody has answered your lobbying for help as well. Two peas in a pod.

Oh, perhaps you are right, I get you and Mathdotcom confused at times as you sound so much alike. Sometimes I think you and he are the same person. But thinking it over, I do think I'm correct on Kelly. Not to mention your contention on Bayesian priors.

But if you are not working through a prior, likelihood, posterior, = new prior cycle; and then thinking that "half Kelly" is "optimal" in any way, you and the rest of the math-wolfpack are massively deluded. And if you cannot construct a proper prior, you cannot use Kelly in any useful way. Your contention that half Kelly is a "better" method simply points out that you simply don't know your priors and thus your actual win%s.

[bztips]

But if you are not working through a prior, likelihood, posterior, = new prior cycle; and then thinking that "half Kelly" is "optimal" in any, you and the rest of the math-wolfpack are massively deluded. And if you cannot construct a proper prior, you cannot use Kelly in any useful way. Your contention that half Kelly is a "better" method simply points out that you simply don't know your priors and thus your actual win%s.

Sigh... Wreck, I know you just love to hammer the Bayesian angle, but really that's a philosophical dispute. Believe it or not, there's a real simple non-Bayesian justification for using partial Kelly: some people are just more risk averse than would be implied by straight log utility, which is one of the assumptions of the full Kelly formula.

Long story short is that partial Kelly can be optimal in a frequentist setting without having to resort to any of your Bayesian prior stuff.

[Inspirited]

2) Smaller sample sizes, that are just as real, but won't last from season to season. These shorter term fluctuations are available to those 'surfers' who know how to read the waves. A true skill in itself. This field almost certainly will not include the rigid thinkers of the math sect, who typically confuse this profitable area with data mining. I don't know why these more easily recognizable fluctuations tend to last for a season only, but I know that they do (perhaps it has something to do with the 'group mind' of all the athletes in a league). They don't require huge amounts of research, but remain unidentified without a relatively high degree of awakeness. Bettors can profit from these shorter term fluctuations with full awareness of their temporary nature.

What are these wave that you speak about? Can you give any examples. It sounds like timing a market or maybe more so playing differently in poker when you think you're on a hot streak or on a cold streak. In poker that's considered illegitimate superstition.

What in the world does riding the waves mean?

I've thought about trying to take a small sample of basketball games during the early days of a month to see where the O/U are falling and then go with it for the month. You do see months that swing profitably in one direction or another, but it's an arbitrary period of time and actually trying to generalize from the beginning of the month seems silly. Does your theory have to do with anything like that?

[Wrecktangle]

Sigh... Wreck, I know you just love to hammer the Bayesian angle, but really that's a philosophical dispute. Believe it or not, there's a real simple non-Bayesian justification for using partial Kelly: some people are just more risk averse than would be implied by straight log utility, which is one of the assumptions of the full Kelly formula.

Long story short is that partial Kelly can be optimal in a frequentist setting without having to resort to any of your Bayesian prior stuff.

As a Frequentist, I'm sure you believe in Intelligent Design and a flat world. "Classical" statisticians are being left in the dust by engineers and applied math guys who are interested in practical solutions to difficult statistical problems. The Frequentist approach simply has no consistent method to pull in new information and improve their predictions with their mis-mash of tests and "experiments."

And as I said, Kelly "proved" there can only be one optimal way and if you buy that then half Kelly cannot be. You simply don't know what your true win% is, and then you delude yourself into "fixing" it with limp measures like half Kelly, 1/4 Kelly, etc.

BTW, it must be tough finding a Doctor that will bleed you these days.

[bztips]

And as I said, Kelly "proved" there can only be one optimal way and if you buy that then half Kelly cannot be.

Nothing more need be said. Simply put, you do not understand the Kelly proof or its assumptions... which was already made clear from your statement awhile back that "Kelly is probably not optimal" -- clueless.

[Dark Horse]

What in the world does riding the waves mean?

It's a good question. I don't presume to know the theory behind the phenomenon. The underlying idea is that each season is different and that specific contours or outlines of the season will become clearer as the season progresses. (This also means that it would be a mistake to rely too heavily on previous seasons). I find this one of the more interesting themes in sports betting, but until now have never really thought about the why (that may change). In general the 'waves' appear to be more reliable towards the later part of the seasons.

[MonkeyF0cker]

It's a good question. I don't presume to know the theory behind the phenomenon. The underlying idea is that each season is different and that specific contours or outlines of the season will become clearer as the season progresses. (This also means that it would be a mistake to rely too heavily on previous seasons). I find this one of the more interesting themes in sports betting, but until now have never really thought about the why (that may change). In general the 'waves' appear to be more reliable towards the later part of the seasons.

Gambler's fallacy mumbo jumbo FTW.

[donjuan]

Pretty much every Dark Horse post in here is Gambler's Fallacy dressed up with obfuscating language.

[Dark Horse]

Pretty much every Dark Horse post in here is Gambler's Fallacy dressed up with obfuscating language.

Yes, that must be it.

Now go back to price hunting.

[mathdotcom]

Dark Horse: there's never complete certainty about anything. We are all just trying to make as best a guess as we can, and the guy with the best guess takes home the money. Please no more posts about waves or contours.

Wrecktangle: get a job.

[chunk]

It's a good question. I don't presume to know the theory behind the phenomenon. The underlying idea is that each season is different and that specific contours or outlines of the season will become clearer as the season progresses. (This also means that it would be a mistake to rely too heavily on previous seasons). I find this one of the more interesting themes in sports betting, but until now have never really thought about the why (that may change). In general the 'waves' appear to be more reliable towards the later part of the seasons.

I had seen some fairly cogent replies until this. Maybe you were on some primo stuff. I gotta say....this wave stuff scares me.

[Dark Horse]

Dark Horse: there's never complete certainty about anything. We are all just trying to make as best a guess as we can, and the guy with the best guess takes home the money. Please no more posts about waves or contours.

There are very strange phenomena in the world of science, which these math guys have never heard of. So I don't quite see why a discussion should be confined to only what they're aware of.

Many of us have realized that each season is different, and that the same methods don't necessarily carry over from one season to the next. I find that interesting. Would it be strange to think that teams, as the season progresses, become more aware of each other, of the type of season in progress, and of the relationships to other teams within the season? Clearly not. Could that mean that they start to gradually 'agree' with each other as to how the game is played for that particular season? Possibly.

Undoubtedly, this type of hypothesizing is beyond the comfort zone of the few unimaginative minds that have taken it upon themselves to police the think tank. But even they should be able to comprehend that a hypothesis is a starting point, and not a final conclusion. Some hypotheses are dead end streets, others lead to new insight.

[ziptoe]

this is what i do in soccer,

find a game with even odds for the under. wait for the over to raise even (usually 10-15mins) then place your bet.

always been good to me.

place your bet on the over or the under? not clear.

[pedro803]

As for the waves/season contours hypothesis -- I would be careful that any of these waves, or countours, are uncovered in real time b/c if they are discovered in past seasons by looking at the whole data set after it has been completed then there would be a real danger of data mining being involved -- that is, it is easy to just go back and look at a set of games and say oh this or that correlates to winning, but when the games are already decided you run the danger of using a variable to explain itself so to speak -- just my two cents

this is why it is important to understand that for backtesting you need longitudinal data -- data that will let you replicate each pick that the model would have made with only the data that was actually at hand at the time that the pick would have been made. i.e. using only the data from the previous games up until that point.

[evo34]

There are very strange phenomena in the world of science, which these math guys have never heard of. So I don't quite see why a discussion should be confined to only what they're aware of. Many of us have realized that each season is different, and that the same methods don't necessarily carry over from one season to the next. I find that interesting. Would it be strange to think that teams, as the season progresses, become more aware of each other, of the type of season in progress, and of the relationships to other teams within the season? Clearly not. Could that mean that they start to gradually 'agree' with each other as to how the game is played for that particular season? Possibly. Undoubtedly, this type of hypothesizing is beyond the comfort zone of the few unimaginative minds that have taken it upon themselves to police the think tank. But even they should be able to comprehend that a hypothesis is a starting point, and not a final conclusion. Some hypotheses are dead end streets, others lead to new insight.

It's not uncomfortable. It's just b*llsh1t. If you have to resort to "very strange phenomena in the world of science" to describe your strategy, it's fairly safe to say you're not making money.

[chunk]

Sports betting, however, is not an exact science is it? I know that there are those that claim that they have the probabilities nailed down to the proverbial gnat's arse, but I'm sceptical of that. After all we aren't flipping a fair coin or rolling a fair pair of dice. There are many dynamics that come into play that could influence the outcome of an event that are difficult if not impossible to quantify. I do believe that the math is first and foremost, but not the end all when it comes to handicapping.

[mathdotcom]

There is no money to be made in the philosophy of gambling

[Maverick22]

Can we all just agree that Herbert Spencer and Charles Darwin said it best with "Survival Of The Fittest"?

[Dark Horse]

It's not uncomfortable. It's just b*llsh1t. If you have to resort to "very strange phenomena in the world of science" to describe your strategy, it's fairly safe to say you're not making money.

And just who might you be? lol

I've done this for years, and I know that researching the 'unusual' is anything but bullshit.

Meanwhile, as Wrecktangle also noted, the OP is long gone. Another victory for the sad people.

[evo34]

And just who might you be? lol

I'm a guy who makes money. I'm quite sure that outside of bonus whoring and asking for other handouts, you're well in negative territory lifetime. How's your groundbreaking NFL turnover research going? Never would have guessed a correlation between past turnovers and past winning...so counter-intuitive. As soon as you figure out how turnovers are "created," fill us in. Whatever your tea leaves end up telling you, don't let the data get in the way. It's only for sad people who are good at math.

[bztips]

Come on, everyone knows that turnovers are a prime example of DH's wave theory, and it becomes easier to predict them later in the season as their occurrence becomes more "reliable".

[Justin7]

Oh, perhaps you are right, I get you [MonkeyFocker] and Mathdotcom confused at times as you sound so much alike.

Try to keep them straight. Mathy understands theory, and wins on his own merits. I can't remember MonkeyFocker ever making a serious contribution to HTT. He wants to play with the big boys, but hasn't done the work yet. If he is a winning player, he has shown no evidence of that here.

Quote:
Originally Posted by Dark Horse
"It's a good question. I don't presume to know the theory behind the phenomenon. The underlying idea is that each season is different and that specific contours or outlines of the season will become clearer as the season progresses. (This also means that it would be a mistake to rely too heavily on previous seasons). I find this one of the more interesting themes in sports betting, but until now have never really thought about the why (that may change). In general the 'waves' appear to be more reliable towards the later part of the seasons."

Gambler's fallacy mumbo jumbo FTW.

If you don't understand what Dark Horse said, you need to reconsider how you are doing your modeling. If scoring goes up 5% a year for 2 years, you're screwed if your models don't pick up on that -- even if it is stat driven, you'll get horrific results when both offense and defense are in alignment to make a higher or lower scoring game. Even in the course of a single season, the smaller changes can make a winning model useless.