NO blind system can beat the books in the long run. This will turn into disaster this year. Stick to handicapping methods explained in bettingresource that evolves with time and keeps the winners rather than fading away like stupid blind systems.
-
Comment -
I agree with 747. things change and it's best to evolve with those changes. players injured, lockouts, etc throw off blind systems. maybe not over the course of 1000 season but do you have that long?Comment -
when checking to see if I beat the closer should I measure against the actual line I bet or what the opener was to see which direction the money moved? Or track both and use both in observations? Often times it moves during the day where if I bet a couple hours earlier or later I'd have beaten the closer but instead got a price that didn't beat it.Comment -
Either way; the only difference is adding a few cents to represent the margin. Intuitively, it makes more sense to me to compare to the same side, rather than the no vig.
Ask yourself which statement is more informative, from a modeler's standpoint: if I were to conclude that I beat the no vig closer 40% of the time, by -2 cents on average; or if I conclude that I beat the same side closing line 56% of the time, by 3 cents on average?
They're essentially the same numerically, and although you're almost certainly doing better than coin-flipping, you're also unlikely to be profitable. However, the second set of numbers is a lot easier to process, and get an idea of where you are performance-wise. I'd argue that doing a binomial test in the second case is also more in line with testing what you really want to know, with the null hypothesis being that you are coin flipping. In the no vig case, you'd be testing whether you beat the no vig closer more than half the time. The problem is, in the development stages of a model you don't really expect it to perform that well immediately. A hypothesis test with a lower stringency is still informative, and usually a lot more helpful in the short term.Comment -
A couple of thoughts:
Usma, that is indeed a very nice percentage, but I'm having trouble understanding how much of those results are based on forward modeling rather than only back-testing.
Also, I happen to agree with the thoughts on incorporating flexible parameter estimations in models and don't feel that this invalidates the model... necessarily... What I think people have a beef with is not the dynamical nature of models, its that people so frequently choose to back test an already manipulated set of weightings on old results and falsely believe that this accurately reflects the predictive power of their model. Implementing any sort of dynamical modeling technique it still has to be tested in a forward manner.
Another quick note about dynamic parametrization:
I'd argue that this is almost a necessity to incorporate in handicapping models for a variety of reasons, but most intuitively because sports change... Take NFL/NCAA football for example, if it was the 1960's/1970's, your model should be valuing run offense/defense a whole lot more than it would for the 2013 season, not even because teams (especially in the NFL) themselves focus so much more of their offense on passing now (although thats part of it for sure) but because of how pass interference is called and how much QBs/WRs are protected.Comment -
Thank You for your comments...
I understand what everyone is concerned about. My hypothesis behind my model have changed dramatically from 2010. I started with a blank spreadsheet and now I have a fully function--yet untested model that I can update and churn out results in less than 15 Mins... Week to week. After looking at the statistics and trying several different methods... I believe that I have settled on my approach. It took me around 5-6 Weeks into the 2013 season to get there. I tested it each year in 2010 2011 and 2012 and failed miserably each year to finally get to this point. Will it work this year? Who knows... What I do know is that much thought and time has been put in to get to this point. And though I have optimized and back tested ... I think my relationships make sense. And will continue to make sense as this year moves forward. But again, I will know Week 5-Bowl Season. I do not believe that I will hit 60% but I am predicting a winning season. I have already had one winning season with college basketball.Comment -
My system is far from blind but I get the point. However, I have chosen to believe that I can create one... and believe that I already have created one. We will see shortly. I think I can beat the books... or I would have never started it in the first place.Comment -
You really need to read this:
If you test on games which were used to create/modify/tweak/etc/etc your formula, your testing is meaningless. Your testing is essentially giving you an unfair advantage (i.e. knowledge of future game data) that you have absolutely no way of replicating in the real world. You have the capability of properly testing your formula; why wouldn't you want to do it before risking your own money?
No one knows whether or not your formula will work. But just know that what you're doing is exactly the same as betting your formula without ever doing any testing at all.Comment -
usma, what's the point of opening this thread if you are going to ignore what everyone says anyway? i agree with what's been said about the way you built your model, but it's pointless as you don't really care either way and just want to start betting your model. so you do that, and good luck..Comment -
It is a scientific fact that most people (pretty much all except those who make a real cognitive effort not to be like that) when look for information, don't really look for information.
They look for CONFORMATION of their existing beliefs and preconceived notions.
Pleasure they derive from conformation, often faulty, of them being "right" they cherish way more than benefits they can get from stopping being wrong.Comment -
I am not completely ignoring what everyone has said. However, I do believe at some point the numbers don't lie. Let's suppose, I have 7 seasons of 500 games a piece of completely backtested data that yields 57%... are you honestly going to tell me of the scan of 3500-4000 games that I haven't stumbled upon a relationship that holds true. Even if it is backtested... At some point, don't the numbers ... mean you have a reasonable chance of having found a relationship that works...And to say it has no value until it proves itself to be successful doesn't make sense to me. My college basketball program worked... do I stop there... absolutely not... I have to continue to refine the program and make it better even though it worked...My original question is ... at what point do the number of games actually mean something...Comment -
At no point pattern that worked in the past means something by itself.
Unless, of course, you are getting paid for finding patterns that worked in the past.
And yes, I am honestly going to tell you that after whatever thousands of games
You have stumbled on nothing.
You are doing it backwards.
Instead of establishing rules for what to look for first and then scanning data trying to find it,
you scan data being clueless on what you are looking for, and then make up rules to fit your random findings.
Did you give any thoughts to sharpshooter story I told you a few posts back?
If not - you freaking should!
That is exactly about you.
Instead of drawing a bullseye first and then trying to put as many bullets as you can in a center of it,
You randomly spray thousands of bullets and then draw the bullseye around largest clusters produced by randomness. And then you wonder why no-one even begins to consider you any kind of shooter, never mind a good one.Comment -
As Matthew already mentioned, though, once you start using your model in real-time, if you track the number you got against the closing line, you'll be able to determine if your model is any good a lot faster than by W/L alone.
Neither closing line value nor a p-value matter until that time, though, because your method of backtesting inherently produces invalid results.Comment -
Holy crap, just make your bets using your model and quit worrying about what other people think. Most everyone else can see the telltale signs that it's a completely jiggered collection of mined data, but that doesn't matter to you and you don't care. And that's perfectly OK. It's your money and you can wager it however you like.
Track the plays made using your formula this season, and when January comes you'll have a clearer idea of whether or not you built something worth having.Comment -
Modeling up date....First two game this week...
My projections beat Vegas significantly each time. I had 48 points vs. 66 they scored 42. I had 84 points on Fresno Game ... Vegas had 69. They scored 81 points. You can say what you want about not being predictive... but it is coming around. I start betting next weekend. Enjoy Dave... P.S. I realize that it is only 2 games, but Vegas was way off... I was not. And my NFL projected the Kansas City under ...enjoyComment -
Don't worry ... Hu will be able to tell me I told you so soon enough...
If I view this from a half full point of view. I have already made it. My college basketball program not only went under the same scrutiny and mindset last season but was said to be not worth it at that time. It slayed it as a predictive model. So, using hutennis premise. I have one model ready to print money which is fantastic. My goal, however, is to have a year round solution that hits between 56-60% with minimal effort. Hu can quickly tell me I told you so soon. I start betting this week and will publicize all of my bets. I have nothing to hide. I have spend too much time and energy. If it doesn't work, I would rather be truthful with myself and try to fix it than lie to myself and say I am there. Enjoy Dave
Comment -
No, no, no!!!!
I never tell anyone personally "You are dumb and stupid"
That would be a personal attack and I don't do that.
I will gladly, though, make your argument and/or position look and/or feel dumb and stupid (if it is warranted ofc.)
There is nothing wrong with that. That's what makes (at least, it is intended to make) people think, re-think, learn, become smarter, stop being a shark bait, save money, make money and so on. And it goes both ways. You can ridicule my arguments all day long if merits are there.
And if you are right, I will learn from you and tell you "Thank you very much" That's how it works, or at least supposed to work among rational and reasonable people.
Unreasonable, irrational and not very bright among us will cherish their preconceived notions and believes more than truth and even more than money. That's how strong that fallacy is. They take attacks on stupid and dumb fantasies in their heads personally, which is stupid and dumb in and off itself. Vicious circle.
They are not interested in information on why they can be wrong, even if being wrong will cost them dearly.
All they need is CONFORMATION on why they are right even if this "conformation" is nothing more but a
another degenerate and useless "encouragement" like
"Looks interesting, dude. Keep on posting. Would you mind sharing that data? PM me. Lets work together on getting books out of business!" or some similar nonsense.
I'm trying to expose it, which should be welcomed b/c it is a form of help.
But I guess pipe dream is much more in order.Comment -
Go over to GL.com and get in touch with Rito or MF'r and see if they can offer you some advice. Both are actually very helpful despite what some around here think. I think they got burned out because honestly very few here could hold an interesting conversation with them once Ganch and J7 left. They bash J7 but they liked him because he wasn't as big an idiot as the rest.
But they'll both be able to help your knowledge grow by leaps and bounds. MF'r has been a programmer for over 10 years and a winning gambler.Comment -
what's gl.com?Comment -
I imagine that you simply don't understand his posts, especially in this thread, or you would have added something constructive
that counters what he said. But that's not really possible, as he and YourAllAmerican are spot on with their additions/advice here...
good luck tracking the plays made using your formula this season, usma1992Comment -
HU... and others fire away...my picks are in and bet...
in 2010 Week 5, I recognized that I hit 50% which wasn't fantastic, 2011 and 2012 were great. I believe that I need to eliminate FCS games next year and go through the previous seasons and account for them. Unfortunately, I don't have the time and energy to do that right now. Anyway, ... my picks are out their... my system is out there ... critique away.... THE ONLY THING I ASK... is if you have a problem with a game. Tell me ahead of time and why. I will look into it. Beating on me or the system really doesn't add any value...TY in advance...
Comment -
My system is far from blind...Comment -
Hu... the problem I have with your comments are that you don't fully appreciate the analysis that has gone into the place I am today. My guess is that you look at me as a guy just looking at numbers. Far from the case. I have spent 3 years looking at where points come from when determining my over/unders. I don't ever use the amount of points they actually score. I know it seems counter productive, but I believe it has a large determining factor of my results. When I finally develop my program and finalize, I will be successful. I maxed out the Graduate Record Exam analytical portion when it existed for a reason. The strength is not in the stats... the strength is in my analysis of the stats... College Basketball proved me correct... TOO Many people said that there were too many variables to even attempt it... I narrowed it down... simplified it... back tested it... than crushed it...I am attempting to do the same thing on 4 other sports... and I am on my way...Comment -
It's often difficult to determine what is real and what isn't in these forums, but if you're on the up and up, I parallel your situation closely. When I found that I finally had enough time to devote to it, I back tested and refined my methods for an entire year in '09. Have been testing in real time since 2010 and have not had a losing season (cbb, cfb, and nfl). Overall ~56% with 1000+ plays ATS. I've been modestly successful long before this, but didn't keep meticulous records like I do now. One difference being I do sides as apposed to totals.
I think that Hu is actually rather harmless. Likes to play devil's advocate often times and sometimes can be somewhat amusing. I will say, however, that he seems to have tunnel vision and doesn't do well "outside the box".
Good luck with your plays.Comment -
No offense, but where is your evidence of not having a losing season since 2010? It takes no time at all to enter plays into a site like pickmonitor.com. Those who don't wish to create a public record of their success probably shouldn't go around publicly boasting about it...Comment -
For the record, using the data you used to create the model to test it DOES NOT invalidate the model. I wouldn't recommend it, as it subjects a "predictive" model to overfitting, but it doesn't completely invalidate it (just ask any global warming scientist). That said, it's not that hard to backtest your model with new data, so just do it. Otherwise, people will be skeptical of your methods, as they should be.Comment -
To be an effective predictive modeler, I would worry much more about collinearity than I would about overfitting. Though honestly, both are reasonably fixable.Comment
