[matthew919]

Interested to know what people think is the theoretical limit for success when betting against -110 odds. For major sports in particular, where lines are sharper, do you think something like 60% is sustainable in the long term, even in the theoretical realm?

I guess what I'm getting at is: what fraction of games do you think differ significantly from their true outcome probability, and what's the variance of the distribution of those discrepancies. And if you had a crystal ball, and were able to exploit the soft lines at will, what would you think is the upper limit of the success rate?

[roasthawg]

If we're talking about people simply applying new and unique methods to crunch the same readily available numbers and bet a small subset of games then maybe around 60%.

Gathering new information/numbers on your own you might be able to up that number.

Insider info regarding player health might be even more helpful.

[brettd]

My 2 cents:

On any individual play using publicly available information: low 60's%, say 62%-63%.

If you can find action on say 20% of the board on the NFL each week or busy NBA days...... and you're doing 60% lifetime in that context, you're one of the top 10 handicappers on the planet. I used to think entropy could have been higher, but not any more.

[Dark Horse]

I would say 60% is the longterm ceiling. If a model hits higher than that, it's most likely overperforming and will come down.

[matthew919]

So you're telling me my model that hits 82% is no good?!

Kidding. I used to believe it was somewhere around 60, but I'm starting to think that there's quite a few areas, even in the major sports, where it is much higher. Maybe even 70% in theory. But I'd agree- even with a crystal ball, 20% of the board is what I would guess too.

[Dark Horse]

You can hit higher in a sport for a season, but over three years I would say 60% is the ceiling. But only for the very best models. Sports betting is highly subject to short term fluctuations. You see it all the time in contests. One year a player hits 65%, the next he doesn't even break 50%. I have a very simple and short term model, with forward results only, in front of me that went 22-8 since I started it at the end of November. That's too high, so my first impression is that luck was involved. The next step is to go to the action points and see how many results were close. See what happens to your winning rate when you toss out the results that could have gone either way; for instance by using a 3.5 point cut-off point (both ways, so the impact of one TD in football).

[HeeeHAWWWW]

Something to think about: there is a site called pickmonitor which allows people to log picks, then sell through the site. That attracts a lot of good cappers, because it handles all the troublesome logistical stuff of distributing picks and collecting money.

Looking through these cappers, with minimum 100 picks, there are 16 guys above 60%. Min 200, there are 4. Min 500, there aren't any, and there are only 9 above 55%.

Make it 1000 picks min, and there is only one above 55%. He's 55.09% over 1048 plays, and almost all of his profit is in a recent hot period, previous months show minimal growth. There are a few between 54 and 54.5% that look to have more balanced bankroll growth graph, ie constant growth, not just one freak period.

This isn't too surprising, when you consider the size of bets allowed on major sports. Someone who could genuinely hit 57% year after year in any major sport would find a way to use pinnacle -105 lines (ie 11% RoI), and at half-Kelly would double their bankroll every 138 bets (median), and multiply it 12x over 500 bets.

10k into 125k over 500 bets? Don't see many people doing that. It keeps going too, by 1000 bets $1.5 million (although of course wager limits hit before then). Not very realistic. As for 60%? 10x bankroll growth every 200 bets, 100x every 400, 1000x every 600 ...... or turning $10 into $1million in 985 bets. Books would all be bankrupt years ago :-)

[easyliving]

In the longterm 60% sounds slightly high. And as stated above if your hitting 60% over thousands of bets laying -110 your probably one of the best in the world. I would think 58% or so would be more reasonable.

[chunk]

54 or 55% is more reasonable over large sample.

[HeeeHAWWWW]

Just to put some reasonable assumptions into the discussion: this is staking on a major sport at a decent book (ie not some rec book with soft lines who will boot you as quick as look at you), and we aren't simply skipping the lower edge bets to stat-pad (why pass up profit?)

[Sawyer]

54 or 55% is more reasonable over large sample.

Agreed..maybe %57.

%60 Over long haul doesn't sound realistic.

But also, it depends on which book you're placing your wagers..

You can hit %60+ if you're betting at square books, small sportsbooks/locals. (Slow to change line, don't know value of key numbers such as 3 to 3½ or 6½ to 7½ in NFL etc)

Hitting %60 ATS at Pinnacle? Not realistic. You can make %60 over 200-300 bets but after 1000 bets, it will land somewhere between %56-58 depends on your skill. I was able to hit %65 (Pinnacle Odds) in NBA Totals few years ago, going 133-83. (Full Documented on a betting forum) Next year, I end up somewhere between %53-54.

[hutennis]

Mature market's odds are initiated be very sophisticated methods developed by very smart people who are highly motivated to do their best
to be the best in a field of setting odds.
Then those initial odds are super fine tuned by by non stop input from countless millions of participants a lot of whom are very smart and also highly motivated.

According to Diversity Prediction Theorem these very diverse efforts combined are guarantee to produce results that are exceptionally close to actual fair value which is almost always lays somewhere within a limits protected by vig.

I have no idea under what assumptions can anyone project such a monstrous out performance.
55 - 60%% long term! What would allow for that to become available?
For someone to make that much this someone must be either one of a kind or the rest of the world must be clueless.

[Dark Horse]

I'm talking about separate models, not lifetime record. IMO, and experience, the highest longterm ceiling that can be sustained for an exceptional model is 60%.

[HeeeHAWWWW]

I have no idea under what assumptions can anyone project such a monstrous out performance.
55 - 60%% long term! What would allow for that to become available?

Have to agree. The key point comes in "longterm". A 55% player even over 500bets has a 95% probability of 253 to 296 wins (about 50.5% to 59%). 99% range is even wider, 49% to 60.5%.

[brettd]

I'm talking about separate models, not lifetime record. IMO, and experience, the highest longterm ceiling that can be sustained for an exceptional model is 60%.

I agree, but exceptional should be displayed as exceptional.



People underestimate what the term truly means.

[peacebyinches]

This is a tough question, there are a LOT of limitations, logistically and financially that you would need to clarify. The immediate financially based quandary that comes to my mind is whether you are betting for pure win percentage, or whether you sacrificing your win rate for more bets that counter intuitively actually increase your earnings.

What I mean by that is, lets say you want nothing more than to win the highest percentage of your -110 wagers that you possibly can. Now lets say you have developed a model where you know for sure the likelihood of winning any particular wager. Logistically, even if you managed to do this, unless you've stumbled upon some sort of handicapping holy grail, the model is only going to provide you with win rate confidences within a few percentage points more or less than 50% in for the majority of any -110 wager.
Sometimes you may find matchups with a confidence win rate of around 60%, but how frequently is your model actually going to find matchups where this is the case? 1 game a week in the NFL? 2 or 3 a week for NCAAF? 50 games for the entire NBA season? These are purely hypothetical situations of course, but the point is that as a handicapper, do you NOT bet on games your model predicts a 56/57/58/59% confidence of winning in order to maintain a more robust winning percentage? You would essentially be throwing away +EV games and missing out on potential earnings.

Hell, you could acheive a win rate of close to 100% if you only bet on games you know are fixed, but unless your an NBA referee or you work for the mafia good luck finding more than 1 of these in your lifetime.

[Miz]

It depends on which plays you are talking about ... as the size of the difference between the line and your prediction increases, so should your success rate (up to a point). The peak success rate at the best difference you can find (looking backwards) is unlikely to be greater than 58-59% against opening lines and maybe 56% against closers. There are also far fewer available plays to bet that have this type separation. A more realistic win percentage is 55-56% against opening lines and 53.5-55% against closers, betting 5% of the plays on the board. Some models produce far more plays than 5%, and still do pretty well against opening lines.