PSA: Large soccer underdogs = fake edges

[sharpedgepicks]

Great points slips. Although to some extent I find it offset by bettors overvaluing faves in soccer. In a game where a single goal can alter outcomes a big dog seems like value - rare to see 14+ pt dogs win in NFL or even CFB yet seems to happen regularly in soccer. Not saying you could beat bet fair just that the true VIG maybe on the fave. Just a theory.

[Slips]

Great points slips. Although to some extent I find it offset by bettors overvaluing faves in soccer. In a game where a single goal can alter outcomes a big dog seems like value - rare to see 14+ pt dogs win in NFL or even CFB yet seems to happen regularly in soccer. Not saying you could beat bet fair just that the true VIG maybe on the fave. Just a theory.

Please keep the discussion in this thread math focused. I am not interested in handicapping the largest soccer markets in the world.

[ProlinePlayer]

A very well presented idea here slips.

Everything you have stated has of course been know for a long time and it has been debated over whether a different method of calculating these percentages can be used. The problem is that no-one has come up with a better way of doing so.

The offshore lines really only identify the probabilities to a certain range. We use the assumption that the vig is evenly distributed to make a more exact estimate. But as you point out, this assumption may often be flawed.
Since the source of the errors are created by the vig, one obvious safeguard is choose a bookmaker with the lowest vig. Note that the 2% charged by Pinnacle in your example is lower than the vig charged by exchanges. If there was a switch to Betfair we may be exchanging more accuracy in some games, for less in others.

1) Matchbook at the same time had Czech Republic at +650 / -700, meaning a fair line of 12.9% and proline edge of just +3%. Because exchange lines (Matchbook, Betfair, etc) are two-sided (you can either back or lay any bet), it's generally safe to assume that they are more accurate than Pinny on liquid markets. Put another way, using PLP's fair line of 14.0% would mean that betting Czech Republic +650 at Matchbook has a 5% edge.

Two things to consider here. Why do you assume with Pinny that there is a bias in the distribution of the vig but accept that with the exchanges there is not? Also, I would question whether or not the true line is even between the 2 quoted prices is a safe assumption. I'm not sure what Matchbook charges, but if the line came from Betfair then the 'real' Betfair line is more like +615 / -710. With the true price being somewhere in that range.

A more accurate model is that market makers seek to minimize their exposure to sharp action, by setting the lines so that their best-estimate of the fair line lies at the midpoint. This is done by distributing the vig evenly (1/2 to each 2-way line, or 1/3 to each 3-way line) rather than proportionally.

Do you know this to be true or are you just guessing. Although I agree that there is a lot of reason to move more of the vig to the side of a large dog, in most cases it would seem to me, to be much more efficient to evenly distribute the vig.

2) The standard no-vig calculation used by PLP rests on the assumption that the vig is distributed proportionally across the 3-way lines. That assumption makes for convenient calculations, but does not generally hold true--especially for large underdogs. A more accurate model is that market makers seek to minimize their exposure to sharp action, by setting the lines so that their best-estimate of the fair line lies at the midpoint. This is done by distributing the vig evenly (1/2 to each 2-way line, or 1/3 to each 3-way line) rather than proportionally.

So the new fair-line calculation would be as follows:

-217 breakeven is 68.45% for Germany
+419 breakeven is 19.26% for the Draw
+600 breakeven is 14.29% for Czech Republic

These sum to 102%, meaning an excess vig of 2%. Instead of applying the vig proportionally, as the standard no-vig calculation does, we apply it evenly across the three lines:

2% / 3 = 0.67%

68.45% - 0.67% = 67.78% for Germany
19.26% - 0.67% = 18.59% for the Draw
14.29% - 0.67% = 13.62% for Czech Republic

This is an intriguing idea. I'm going to run some numbers and see what kind of difference there is using this on more equal games as compared to the accepted method of calculating the probabilities.

I'd also be glad of some input from the many pros here. What do you think of this method?

Thanks for sharing slips. An excellent post.

PLP

[MattyKGB]

2) The standard no-vig calculation used by PLP rests on the assumption that the vig is distributed proportionally across the 3-way lines. That assumption makes for convenient calculations, but does not generally hold true--especially for large underdogs. A more accurate model is that market makers seek to minimize their exposure to sharp action, by setting the lines so that their best-estimate of the fair line lies at the midpoint. This is done by distributing the vig evenly (1/2 to each 2-way line, or 1/3 to each 3-way line) rather than proportionally.

Do you have any source or theoretical basis for this model? It seems weird to me, and it breaks down completely in some extreme cases where odds-implied probability = true probability - (1/2 or 1/3 of vig) could reach zero or even go negative.

I see the logic in your statement, that books would want to maximize their margin of error between the actual probability and the odds-implied probability. I think the problem is that the formula:

odds-implied probability = actual probability + vig factor

implies a linear scale, and those don't work very well for probabilities. E.g. under a linear scale, the difference between a 90% win prob and a 95% win prob is the same as the difference between a 5% win prob and a 10% win prob.

The formula that PLP is using,

odds-implied probability = actual probability x vig factor, or
log(odds-implied probability) = log(actual probability) + log(vig factor)

implies a logarithmic scale, where the difference between a 90% win prob and a 95% win prob is the same as the difference between a 9% win prob and a 9.5% win prob. I think this one is more defensible, although it may not be optimal either.

If I get some free time I'll run some tests with a linear scale vs a logarithmic scale vs a logit scale (where the difference between a 90% win prob and a 95% win prob is the same as the difference between a 10% win prob and a 19% win prob) vs a probit scale (where the difference between a 90% and a 95% win prob is the same as the difference between a 10% win prob and a 17.9% win prob).

[Slips]

The offshore lines really only identify the probabilities to a certain range. We use the assumption that the vig is evenly distributed to make a more exact estimate. But as you point out, this assumption may often be flawed.
Since the source of the errors are created by the vig, one obvious safeguard is choose a bookmaker with the lowest vig. Note that the 2% charged by Pinnacle in your example is lower than the vig charged by exchanges. If there was a switch to Betfair we may be exchanging more accuracy in some games, for less in others.

1) Matchbook at the same time had Czech Republic at +650 / -700, meaning a fair line of 12.9% and proline edge of just +3%. Because exchange lines (Matchbook, Betfair, etc) are two-sided (you can either back or lay any bet), it's generally safe to assume that they are more accurate than Pinny on liquid markets. Put another way, using PLP's fair line of 14.0% would mean that betting Czech Republic +650 at Matchbook has a 5% edge.

Two things to consider here. Why do you assume with Pinny that there is a bias in the distribution of the vig but accept that with the exchanges there is not? Also, I would question whether or not the true line is even between the 2 quoted prices is a safe assumption. I'm not sure what Matchbook charges, but if the line came from Betfair then the 'real' Betfair line is more like +615 / -710. With the true price being somewhere in that range.

I'm not sure what you mean that the 2% vig at Pinny is less than the vig charged by exchanges. The +650/-700 line from yesterday has less than 1% vig - and that's almost always the case for exchanges on major soccer leagues. If you mean because the exchanges charge commission, well there are two ways of looking at that. On the one hand yes, the posted offer isn't the true effective price available so you would need to account for that. The +650/-700 becomes +631/-712 after the 1.5% commission. Note that the effective vig on the line after commission is still only 1.4%, so still less than Pinny. On the other hand, you should also look at the lines posted from the perspective of the market maker. Matchbook charges 0.75% commission on posted offers, so the market maker is effectively booking at +655/-689 (0.6% vig). So overall it seems that the commission effects offset each other, especially considering exchanges frequently offer reduced commission to high-volume players, and I think it's safe to take the posted line (+650/-700 in this case) at face value as long as there is decent liquidity.

As for the distribution of the vig on the exchanges, as you noted the most important feature is that a lower vig means less room for bias. If we use +655/-689 as our basis, the implied fair line can only be somewhere between 12.6% - 13.2%, compared to the much wider range of 12.3% - 14.3% when using the Pinny 3-way line. Beyond that, the fact that the exchange lines are two-sided leads to a more competitive market and more accurate prices. If you think that the fair line for Czech Republic is +660, there is no way to fix that mispricing by betting at Pinny, but at the exchange you can simply post a lay offer to back your opinion. Thus the exchange lines can update to incorporate additional pricing information whereas the Pinny line remains untouched as long as it is accurate within the 2% margin of error provided by the vig.

A more accurate model is that market makers seek to minimize their exposure to sharp action, by setting the lines so that their best-estimate of the fair line lies at the midpoint. This is done by distributing the vig evenly (1/2 to each 2-way line, or 1/3 to each 3-way line) rather than proportionally.

Do you know this to be true or are you just guessing. Although I agree that there is a lot of reason to move more of the vig to the side of a large dog, in most cases it would seem to me, to be much more efficient to evenly distribute the vig.

I developed this pricing model in part after reading Flash Boys, which describes how traders in the stock market use a similar method to set the spreads for options. Of course it's just a model, and none of us know the true processes that bookmakers use to adjust their lines, but it seems to better match reality than the traditional no-vig calc.

The biased distribution of vig on the underdog is a common pattern that I have picked up on over 10 years of betting for a living. For example, SBObet (another relatively sharp book for soccer) consistently posts laughably bad prices for underdogs on the 3-way line. They had +520(!) on that game yesterday for Czech Republic. And they're currently offering +270 on Liverpool and +420 on Watford for next week's EPL, whereas Pinny has +334 and +492, respectively. Many other books offer similarly terrible lines on the 3-way underdogs. I don't have any inside info on the books' reasoning for doing this, but I would guess it's a combination of risk aversion (not wanting to get burned on large underdogs), and also taking advantage of unsavvy customers who favor betting the underdogs with little regard to the odds. See the recent Mayweather/McGregor fight for a striking example of this.

This pattern frequently comes up in college sports as well, where Pinny or another sharp book will have a moneyline of say +450/-600, and +500 is easily found at other offshore sites. The traditional no-vig calc would say the fair line is +471, meaning +500 is a very good bet (5% edge). But any market that is even mildly efficient would never have 5% edges so readily available.

[Slips]

Do you have any source or theoretical basis for this model? It seems weird to me, and it breaks down completely in some extreme cases where odds-implied probability = true probability - (1/2 or 1/3 of vig) could reach zero or even go negative

I think it only breaks down on 3-way lines (in very extreme cases)? And even then it's only cases where one outcome is essentially impossible, like a 3-way line of -10000, +4000, +50000? You can just ignore the outcome that is drawing dead and treat it as a 2-way market, then the problem goes away.

[MattyKGB]

Do you have any source or theoretical basis for this model? It seems weird to me, and it breaks down completely in some extreme cases where odds-implied probability = true probability - (1/2 or 1/3 of vig) could reach zero or even go negative

I think it only breaks down on 3-way lines (in very extreme cases)? And even then it's only cases where one outcome is essentially impossible, like a 3-way line of -10000, +4000, +50000? You can just ignore the outcome that is drawing dead and treat it as a 2-way market, then the problem goes away.

Say you're a bookie and you want to take a -10000/+10000 vig free line and add 2% vig. The -10000 becomes -infinity.

[Slips]

Do you have any source or theoretical basis for this model? It seems weird to me, and it breaks down completely in some extreme cases where odds-implied probability = true probability - (1/2 or 1/3 of vig) could reach zero or even go negative

I think it only breaks down on 3-way lines (in very extreme cases)? And even then it's only cases where one outcome is essentially impossible, like a 3-way line of -10000, +4000, +50000? You can just ignore the outcome that is drawing dead and treat it as a 2-way market, then the problem goes away.

Say you're a bookie and you want to take a -10000/+10000 vig free line and add 2% vig. The -10000 becomes -infinity.

True, fair enough. But like you said those are extreme examples, so while they show some small holes in this model on a theoretical level, I don't think they detract from its general usefulness in ordinary situations.

Hell, I'd even trust a naive american-odds midpoint calculation (i.e. for +450/-600, the midpoint = +/-525 fair line) over the traditional no-vig calculation.

[richhhh]

A visual of the two methods with different odds. Really see the craziness the long odds do with slips method (which are not relevant with proline offerings)

SEPT 2ND.PNG (30.51 KiB) Viewed 12700 times

SEPT 2ND 2.PNG (30.8 KiB) Viewed 12700 times

[richhhh]

So in conclusion, PLP should
1) use the lines from betting exchanges and
2) use a more accurate no-vig calculation for soccer 3-way lines. There have been days where PLP showed +10% soccer edges that went OTB, that were actually -5% losers according to Matchbook. So it's pretty dangerous to rely on PLP numbers until this gets fixed - we need to be a lot more conservative with playing soccer edges.
sincerely,
Slips

Right now plp is showing one edge for today .

sept 3.PNG (20.19 KiB) Viewed 12833 times

Lots of books offering lines really but really only one that means anything is pinny. The rest are recreations books like bet365 or a handful of legitimate books like sbo which take more juice then pinny.

sept 3 2.PNG (86.03 KiB) Viewed 12833 times

Then there is Matchbook showing us your above example.Problem is you can only bet 39$ at those odds. Not enough liquidity as you say.

sept 3 4.PNG (35.25 KiB) Viewed 12833 times

If Pinnacle moves its line the whole market will move including Matchbook ( users will edge play Matchbook to pinny like we do with OLG) . Why is Matchbook so great? Lets say for example Matchbook is the boss is plp supposed to gauge if its liquidity is efficient? Also if Matchbook was boss then pinny would moves its lines towards Matchbooks. The Back and Lay concept is nice to reduce the noise of three way vig but I dont think Matchbook solves this known issue.