Steamsharp Chronicles.
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
I will try my very best to answer these questions, it may take some time as they deserve intelligent responses.
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
First I will address Matty in several steps, and before that Congratulations! having a child is the best gamble you have ever taken!
Lets first develop the idea of WE and the thought behind it.
1) no Win Expectation is not win probability, by definition probability is defined through a probability distribution which is normalized. win expectations are not normalized and you might notice that sometimes they add up to greater then 100%. This is by intention because we are only interested in the number of random performances [consider a statistical set aka boxscore a performance] that a home team for instance might beat an away team from a vast cohort of known performances.
we full well know that hits, ops, walks etc etc do NOT add up to runs or even correlate positively. gamblers tend to have a confirmation bias to correlating ERA, WHIP and the like to runs but it just isnt true. we need to ask the question, if the expected boxscore (with a few exotic stats) is X, how many times will they beat Y drawn at random? that is the win expectancy. we as of yet DO NOT know how to change that to WP. its hard! that is because of outliers and the fact that past data often is a bad indicator of the future (read http://nassimtaleb.org/2013/09/nassim-t ... i0ssBuG0v0)
so to pick a winner, we want to see WE on the pitching or batting side add up so that one side gets a very strong advantage over the other team. Pinnacle often is forced to over juice teams like oakland (the 1.334 i was talking about) because the public over 12-14 hours is dropping 98% of the liquidity on the ML. in this case we hammer the underdog because of the "black swan" affect which looks a lot like PLP's superb RL/O bets on teams like MIA or NYM at 5-1.
if the WE looks reasonable, we bet it over the aggregate and by winning the heavy underdogs we generate 60% ROI on the year and beat pinnys CL's. to also address PLP, you must look at all 2100 games of data (i still cant post it as its under 2 megs of data) and look at the black swan affect over the aggregate. what you see is that baseball data is what is called "type 2" data! winning a 3.48 ML bet is massively influential to ROI and you simply should win those bets in any model.
also home teams are a fair deal stronger then away teams! the bias is quite strong for the home side!
Lets first develop the idea of WE and the thought behind it.
1) no Win Expectation is not win probability, by definition probability is defined through a probability distribution which is normalized. win expectations are not normalized and you might notice that sometimes they add up to greater then 100%. This is by intention because we are only interested in the number of random performances [consider a statistical set aka boxscore a performance] that a home team for instance might beat an away team from a vast cohort of known performances.
we full well know that hits, ops, walks etc etc do NOT add up to runs or even correlate positively. gamblers tend to have a confirmation bias to correlating ERA, WHIP and the like to runs but it just isnt true. we need to ask the question, if the expected boxscore (with a few exotic stats) is X, how many times will they beat Y drawn at random? that is the win expectancy. we as of yet DO NOT know how to change that to WP. its hard! that is because of outliers and the fact that past data often is a bad indicator of the future (read http://nassimtaleb.org/2013/09/nassim-t ... i0ssBuG0v0)
so to pick a winner, we want to see WE on the pitching or batting side add up so that one side gets a very strong advantage over the other team. Pinnacle often is forced to over juice teams like oakland (the 1.334 i was talking about) because the public over 12-14 hours is dropping 98% of the liquidity on the ML. in this case we hammer the underdog because of the "black swan" affect which looks a lot like PLP's superb RL/O bets on teams like MIA or NYM at 5-1.
if the WE looks reasonable, we bet it over the aggregate and by winning the heavy underdogs we generate 60% ROI on the year and beat pinnys CL's. to also address PLP, you must look at all 2100 games of data (i still cant post it as its under 2 megs of data) and look at the black swan affect over the aggregate. what you see is that baseball data is what is called "type 2" data! winning a 3.48 ML bet is massively influential to ROI and you simply should win those bets in any model.
also home teams are a fair deal stronger then away teams! the bias is quite strong for the home side!
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
This will address Mattys question on "reasonable odds"
Our model does not exclusively bet dogs but searches for what we call the "logical underdog" in our math, if you bet every dog on a 50k bankroll at $500 a bet you would be -10k down. however if you calculate the WE with a slick little non-linear discriminator, you get a filter which pops those games out which are illogical . More specifically, given pinny's implied break even probability, 1/odds, (the WP you need to break even, "BE") if the delta between the WE and BE is smaller then the favorite, you bet the favorite. this way we generated 30k on $500 bets for 60% ROI. I am happy to email the data sheet if you want to look.
Suprisingly you make this money on 48% Winning percentage (WPer)
Our approach is to be EXTREMELY skeptical of the favorite on any game. we always start by saying, why does this team deserve to have juice on it! so the price of 1.334 or 74.96% WPer for BE to occur just smells rank to us. but you are correct, sometimes even that price is logical, but very very rarely. baseball teams at best win 100 games on the season which to me says anything above 62.5% implied Wper to BE is suspect at best.
I know our customers thrive off of this and often make the favorite juicer so that more money is piled up on that side (the trap game as some might say) because the books often do take small risks and hedge off in asia where you can throw a million down with no blocks or BS, (buyer beware, who you deal with is critical)
Our model does not exclusively bet dogs but searches for what we call the "logical underdog" in our math, if you bet every dog on a 50k bankroll at $500 a bet you would be -10k down. however if you calculate the WE with a slick little non-linear discriminator, you get a filter which pops those games out which are illogical . More specifically, given pinny's implied break even probability, 1/odds, (the WP you need to break even, "BE") if the delta between the WE and BE is smaller then the favorite, you bet the favorite. this way we generated 30k on $500 bets for 60% ROI. I am happy to email the data sheet if you want to look.
Suprisingly you make this money on 48% Winning percentage (WPer)
Our approach is to be EXTREMELY skeptical of the favorite on any game. we always start by saying, why does this team deserve to have juice on it! so the price of 1.334 or 74.96% WPer for BE to occur just smells rank to us. but you are correct, sometimes even that price is logical, but very very rarely. baseball teams at best win 100 games on the season which to me says anything above 62.5% implied Wper to BE is suspect at best.
I know our customers thrive off of this and often make the favorite juicer so that more money is piled up on that side (the trap game as some might say) because the books often do take small risks and hedge off in asia where you can throw a million down with no blocks or BS, (buyer beware, who you deal with is critical)
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
This post will address mattys thoughts on facebook.
in my back of the napkin calculation, I actually could show with math that the IPO price has a greater probability of falling (assuming volitility is bound at some number X) then rising. but then again, the data that drives such model has a dozen nasty pitfalls and fallacies so its all mental Mast*$&#tion.
indeed! you are so right. in trading (my former profession) some people like me say there is no underlying method at all you can learn from the past historical data the corporation releases. my thought was the higher and higer the IPO wen't the more likely the chance that a short will generate returns after IPO. there is actually a really cool method to calculate the "point of no return" using some basic calculus.This is a good example of "hindsight is 20-20". At the time of the IPO there were lots of people who thought it would go up and lots who thought it would go down. Of course one or the other is going to happen, but that doesn't necessarily prove the validity of any underlying method or logic
in my back of the napkin calculation, I actually could show with math that the IPO price has a greater probability of falling (assuming volitility is bound at some number X) then rising. but then again, the data that drives such model has a dozen nasty pitfalls and fallacies so its all mental Mast*$&#tion.
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
as we serve the books, we must calculate every game! but we don't necessarily bet it. we have two small funds, A) a total index fund like the Dow Jones where we bet every game (without fraud or blatent criminal manipulation of LIBOR lol) B) an alpha fund where we choose our best bets and ride them with betfair hedge.Please help me understand this as you appear to be contradicting yourself. You bet every game, but you only bet games where you feel the signal beats the noise. So what about the games where you feel the signal doesn't beat the noise?
B can be very very volitile as sports data is type 2 data, or data that often pays no heed to historical trends on any given day. in the end the "B" fund makes more money but has very wild swings. sometimes i lose 6-7 in a row and go on twitter rants!
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
this in my opinion is false. (please take with a grain of salt) the problem is that people magically think pinnacle is the true win probability to break even when they post lines. I can show you a direct proof by contradiction that this is just false.It's not retarded but it may be unwise if you're attempting to be profitable in the long run. Let's say you determine that each side is equally likely to win a game, and Pinnacle is offering -104/-104. There is no possible way for you to bet this game profitably. So why bet it at all?
I don't know what others think but by being on the other side (seeing how books servers work), I know the books are just reacting to liquidity and trying to heap up bets which have negative win expectancy so that sharp money is rendered ineffective and they will either win a crapload of money when the fave loses or payout next to nothing when they win due to hedges and smart pricing. they know the true WP as well as anyone and the closing lines are almost always not that. (at least we have deluded ourselves into believing that and we will gladly open a bottle of 25yr McClellan's with you in Van if you drop by to debate it)
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
Thanks to our friend Arch, we are actually opening our algos up for use and putting them into a sandbox for free!steamsharp wrote:but lo and behold after 2000 bets we are up 50k on an 100k bankroll. average odds bet = 2.24. win accuracy = 47.43% $500 a bet.
This is very impressive, congratulations! I'd like to learn more about the models behind your pitching and batting WEs if you care to share?
we will launch the website in late OCT or NOV for basketball day 1 and cover all N/A sports.
since we cant give away the secret sauce, we can at least open the results up via API and make it available on a sweet website that tracks bets. all we want is for betters to make more money and experiment with more modern mathematics. The math syndicates are using is stuck in the 80's and some cool new stuff has happened since GPU's have come out. our philosophy is make things free to use and cheap and let 100 flowers blossom.
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
you bet. and home teams actually are stronger then away teams. i think they win close to 55% of games.I notice from your files that both your pitching and batting WEs average around 55-56%, resulting in calculated edges that are positive much more often than they are negative. Is this intentional?
also sometimes that way features are created for machine learning algorithms are inconsistant. sometimes even us PhD doofuses screw up and ill have to see if you caught something here.
often the awya team gets fairly low WE's (well its hard to win on the road) and the home team gets a little boost. data is always a work in progress.
also you are looking at 30 games of data. you really must look at 2000+ games to get a feel
-
steamsharp
- Posts: 38
- Joined: Mon Aug 26, 2013 8:34 pm
Re: Steamsharp Chronicles.
hey PLP i see what you are saying here is my thoughts on that.
building a WP from WE's is really really hard because of 3-4 biases in type 2 data. in fact i have to say "i don't know" how to do that.
my heuristic is to take a training set of 10000 away games and through the statistical configuration onto that tableau of data and see how many away teams that home team (or vice versa) will beat given an implied cumulative distribution function that is meticulously calculated with a couple machine learning routines. its like the expected box score is a snapshot and I want to know how many known snapshots it should beat. building expected values when the braves can get 0 hits or 10 hits on any given day is HARD, and it takes approaches like monte carlos and copulas to do this effectively.
your calculation is not correct for WP and here is why:
A) each WE is derived from a different heuristic which is based on different data (home and away + team specific). creating a joint probability is bad because its like saying you want to predict the chances of a dice turning up 6 twice in a row except you use a wood dice on throw 1 and a lead dice on throw 2 and maybe the lead dice has 2 sixes on it instead of 1.
B) the right way to deal with this is to use bayes rule, as the WP calculated from the WE for CHC vs a random opponent is surely depending on that opponent being above average or below. the dodgers are probably above average. we have been painfully stupid here just looking at the deltas between the WE's and saying that delta has some kind of predictive correlation to winning if its high enough (it does) we are working on how to do WP properly but its hard!
let me know if you think this is valid.
data driven approaches are super hard (like correlating ops, pitches, hits/pitch and various sabermetric values to expected runs scored) because the forecasts on the data often encounters the black swan effect. like that parlay you crushed Where MIA scored 7 runs in a row a few days back and was leading 7-0. the black swan effect is HUGE and most importantly drives 6-8 times more profit then betting 1.5 faves. I know where I want to be.
our philosophy is to capture the black swans and try our very best to bet all the other games well enough so that we avoid too much risk by picking far too many underdogs.
admittingly we just aren't good at this. I am not a pro better, just a pro math guy with a good team. that's why we are opening up our data via a website for free to bettors. we want to learn from the best.
building a WP from WE's is really really hard because of 3-4 biases in type 2 data. in fact i have to say "i don't know" how to do that.
my heuristic is to take a training set of 10000 away games and through the statistical configuration onto that tableau of data and see how many away teams that home team (or vice versa) will beat given an implied cumulative distribution function that is meticulously calculated with a couple machine learning routines. its like the expected box score is a snapshot and I want to know how many known snapshots it should beat. building expected values when the braves can get 0 hits or 10 hits on any given day is HARD, and it takes approaches like monte carlos and copulas to do this effectively.
your calculation is not correct for WP and here is why:
A) each WE is derived from a different heuristic which is based on different data (home and away + team specific). creating a joint probability is bad because its like saying you want to predict the chances of a dice turning up 6 twice in a row except you use a wood dice on throw 1 and a lead dice on throw 2 and maybe the lead dice has 2 sixes on it instead of 1.
B) the right way to deal with this is to use bayes rule, as the WP calculated from the WE for CHC vs a random opponent is surely depending on that opponent being above average or below. the dodgers are probably above average. we have been painfully stupid here just looking at the deltas between the WE's and saying that delta has some kind of predictive correlation to winning if its high enough (it does) we are working on how to do WP properly but its hard!
let me know if you think this is valid.
data driven approaches are super hard (like correlating ops, pitches, hits/pitch and various sabermetric values to expected runs scored) because the forecasts on the data often encounters the black swan effect. like that parlay you crushed Where MIA scored 7 runs in a row a few days back and was leading 7-0. the black swan effect is HUGE and most importantly drives 6-8 times more profit then betting 1.5 faves. I know where I want to be.
our philosophy is to capture the black swans and try our very best to bet all the other games well enough so that we avoid too much risk by picking far too many underdogs.
admittingly we just aren't good at this. I am not a pro better, just a pro math guy with a good team. that's why we are opening up our data via a website for free to bettors. we want to learn from the best.
Re: Steamsharp Chronicles.
Great Insights Steve as usual. We all appreciate you coming back and answering all the qustions u were asked. I know u been busy on your honeymoon and I thank you for taking the time to reply to the many questions the good lads had waiting for you.
Only thing I'll add is $ goes were its respected, and while pinny is no where near accurate for many small market sports and props,The Majors do provide accurate WP% from "special" ppl shaping the odds especially 1hr-30mins before game
Just like in the financial world risk is best priced towards final hour of trading because thats when the most $ is usually traded at exchanges like NYSE or CME.
I agree and have warned many times about treating pinny as "God" but it can be argued that they may be Gods mathematicians LOL
Theres no question that ever since 2007 pinny has seen a decline of liquidity in NAmerican Sports for obvious reasons which may correlate with ur views regarding MLB, I do share similar views for obvious reasons Ive stated to others aswell
Only thing I'll add is $ goes were its respected, and while pinny is no where near accurate for many small market sports and props,The Majors do provide accurate WP% from "special" ppl shaping the odds especially 1hr-30mins before game
Just like in the financial world risk is best priced towards final hour of trading because thats when the most $ is usually traded at exchanges like NYSE or CME.
I agree and have warned many times about treating pinny as "God" but it can be argued that they may be Gods mathematicians LOL
Theres no question that ever since 2007 pinny has seen a decline of liquidity in NAmerican Sports for obvious reasons which may correlate with ur views regarding MLB, I do share similar views for obvious reasons Ive stated to others aswell