QUESTION ON PROBABILITY

Sorry, should have read
1 to 10
2 to 11
3 to 12

Do they ever DRAW?

Baseball

DOUGHHHHHHHHHHHHHHHHHHHHHHHHH

The maths guys will know.....I'm just another "guesser"

GL

Still, would like your views:

Would you divide the 2?
If so, would you divide 70 and 60 percent
or
78 and 60 percent?

Would you just ride the most recent?

I personally wouldn't just use 10 games as you will always get a lot of noise in the rating. Ignoring recents results would be worse than ignoring old results, but why not use all the available data?

I don't mechanically "rate" FOOTBALL teams......and tend to go by their recent games record, taking account of the strendth of their opponents and whether they were home or away. I look for flaws in these scores...eg a team that constantly leaks a goal when playing away or at home.....I look for strengths....do they score regularly, and against either "tough" or "weak" opponents.

I take their league position into account re what happens if they win or lose or draw.

I look at their next few games to see if and CL or other CUP games looming up, and just how important such games are to that team.

If their are INTERNATIONAL matches looming up where a team will have players thinking about such games going into a match etc. Also, take into account teams that might be affected by players returning from international duty.

I then have a view on how the game might be played out.....and which team might score/concede ...be highly motivated, or not etc etc

As I deal with CS LAYS mainly, then whether a team wins or not is not as important as whether they will score or concede a goal.......and also their opponents likelyhood of scoring/conceding .



Racing about to begin, so preoccupied with that now......

The above "puzzle" only gives me a sore heid when glimpsing at it.....so best to rely upon MATHEMATICAL type brains to indicate a path to tread.

I probably should be RATING teams in FOOTBALL matches, but no time to do so at present....or likely to have in near future.

Wish you well on the above.

GL with bets.

Also look for a team's style of play, and how their style might fare against opponents....

Lol

You'd want to use a system that takes into account all information without a hard "cut-off" point, but also has a weighting so that recent information is given more weight than old information.

There are several rating systems that try to achieve this, but all have some weaknesses.

A quick method would be to just use a beta-binomial model, although this doesn't actually weight more recent results as more important.

You gave 11 results for the 10 results form in your post, so I'm a bit confused, but let's say that the sequence of 10 is:

W W W L L W W L W W

You'd distribute p as Beta(2,6).

After the next result (loss) you'd update this to p~Beta(2,7).

If p~Beta(a,b) then each win you add 1 to a and each loss you add 1 to b.


That's using a naive prior, but the end result is the trivial probabilities already mentioned. You can start with values for a and b which smoothes the likelihoods.

But you might yet be better off using a different system. For example instead of just using sum(wins) / sum(wins+losses) you could use the following multipliers:

n-games ago	10	9	8	7	6	5	4	3	2	1
multiplier	0.3487	0.3874	0.4305	0.4783	0.5314	0.5905	0.6561	0.7290	0.8100	0.9000
Win	1	1	1	0	0	1	1	0	1	1
Loss	0	0	0	1	1	0	0	1	0	0

And instead use the sums of wins/losses multiplied by the multiplier to work out the probability.

Now to update you just use wins = 0.9*wins for a loss and wins = 1+0.9*wins for a win, and the same formula with losses as a loss.

Then just use p = wins/(wins+losses).

Instead of a=0.9 you can use any other number

10 games must be a tiny sample for any sport.

One of the reasons is that fixtures may be particulary easy/difficult in a 10 game run.

1 to 10 sequence coupled with loss in game 11 means 78% chance of winning game 12

The reasoning that they win 70% of the time so if they lose their next match they can be expected to win 7 of the 9 after is deeply flawed .

If all their opponents are of equal ablity (which they won't be) and they win 70% of the time then they would be 1.42/43 to win their next match, but you are saying if they lose it they would then be 1.28/29 (7/9) to win the game after! That is a huge difference and so there is no way that both prices could be correct.If anything for the 2nd fame they ought to be bigger odds if anything as the defeat is  (slight) evidence that they are not as good as thought.

The main problem is you seem not to consider what their oppponents sequence is

10 samples IS tiny, which is why the beta model is nice. With only 10 samples the variance on your estimate is very high. Your confidence interval would cross the price you're offered.

tobermory: The main problem is you seem not to consider what their oppponents sequence is

I was going to delve into this in "part two" once I gathered views on the simplicity of the example.

I'm also aware I am showing "one side of the match" and using percent chance was the "rating" of the team which, agreed, is flawed BUT this is what the public gravitate towards when betting ("they've won 80% of their games at home" but don't look at the fact that combined with home and away, they have lost their last four ("they're bound to win this time") and they won most of their games against teams in the bottom of the standings.

My ratings do factor in their opponent and how they performed against them (i.e. my tennis ratings, for example, factor in straight sets v 2-1 when best of three)

I also do weigh up the two ratings and do math from there that then tells me if there is value in the bet, my spreadsheet often putting lines through matches for me based on my own criteria of value.

Once the game is over, the ratings are adjusted immediately.

My chose of ten games for baseball is it factors in the pitching rotation plus home and away performance (though some road trips do last ten to twelve games).

It also comes down to betting on the immediate game at hand.

If the approach of (for lack of expression) the "overlapping ten" games, then I'm considering (and views on this welcome, as well) knocking out the best and worst rated games and basing their performance on eight of the past ten games.

If maths worked for results, everybody would be doing it, and bookies would be skint

The last game a team played bears no relation to the game they are due to play

This is why all the top clubs in every division never go undefeated for a season

... apart from arsenal

twomatchpoints.  Ratings are NEVER the final check on the list.  Just an indicator. 

But they can often insight into the relative strengths of two teams.

Something I came across three years ago:  I built a basic filter and ran it against several leagues using past results.  So entire seasons done in an evening.

The 'edge' it offered was nothing across the board BUT on some of the leagues (where the teams are rated between 0 and 1000), if two teams rated over 750 AND the home team was rated atleast 5% better, the game would be OVERS.

Now, bear in mind, because a team is rated high doesn't mean it is in the top of the league.  The top 6 on my ratings is rarely the same as the league standings.

But here's where it works:
LEAGUE / STRIKE RATE SELECTIONS / LEAGUE AVE ALL GAMES

Australia A / 85% / 51%
Italian A / 74% / 47%
French 1 / 70% / 44%

Average odds / all leagues 1.96

So, do I do these blind?

If there are two selections on the leagues in a weekend, usually, yes, since the strike rate when a league has one is pretty solid.

Three, I'll bet on all of them but will balance out my stakes after researching (i.e. 1.5, 1, .5 points)

More than three, I'll spread according.

But that is where ratings can have a bearing rather than just results.


The assumption on ratings is that people see it as a 1X2 decision maker.  It's not.  It weighs out the match being looked at.  How many times have we seen the SKY TV hype of "Two Giants in football go head to head" blaring at us for a week.  Well, my ratings will sometime point out one is actually a dwarf and the unders play is on.

MadDogs
26 Apr 12 10:17   
... apart from arsenal

and Preston.... even if it was 120+ years ago

Looking only at results is too simplistic. You need to include the underlying performance. In football a team can dominate, yet lose, and in baseball the difference between a win and a loss can be one error or bad pitch. Using the suggested approach in baseball is flawed anyway, since the pitcher is key. Look at the difference in prices on consecutive days - same teams, with one player changed - 1.7 one night, 2.2 the next.

cpfc4me - I might have mentioned this on the thread, I'm using W L for the example of approach.

I rate each game that a team plays.  And based on the criteria, I do have teams that lose actually have a better rating.

So, no, I don't just use W L.  So really, my question is how to distribute weight against performance.

A LOT of weight is put on the pitcher so rating a team overall can sometimes find you a bit of edge.

Baseball is probably the easiest sport in the world to build a model for because it's a sequence of percentage chances that you can assess reasonably well on a player vs player (vs outfield) basis.

It's been proven that you can't play to the score very well (Unlike tennis for instance where the better players can raise their game for more important points) which is also conducive to easier model making.

*The large sample size being the key.

Agree.  There are so many games and the number of teams playing against each other to create a balanced overview of the league very quickly (as opposed to football or U.S. football where the teams play once a week).

don't the above comments from Lori and Shapeshifter just mean that:
the matches are more predictable
the odds are more efficient
rather than saying anything about how possible it is to beat the bookie/market?

That would be the case, except the odds for baseball tend to be built around the Vegas line, and the Vegas line is efficient in making money, not in pricing up the result of a game.

There is a small but important difference between the two which means the thinking punters can still get an edge.

The bookies will focus the public on the pitcher and their record and push the odds in accordingly.  Because of pitcher rotation, they get to do this every six or seven games.  A team can be performing mediocre and their "star" will be on the mound.  This will be enough to take a team that should be 5/4 based on their form and push them into 4/5.

A friend of mine taught me: don't just look at the pitcher but look at the record the team has when they start since, in baseball, a pitcher can leave the mound without a win or loss on his record if the game is resolved while their in the showers.

The bookies will use the results to put up a market.  Again, a team will be pushed in but it is a matter of "looking" at the results to 'rate the win'.  From there, as a punter, I need to gauge the value.

As for "predictability" - there are some match ups that are "predictable" but the bookies will corner you on value.  Like a 1/5 horse race, sometimes better to let them win or look at the forecast, in the case of baseball over under on the runs or -1.5.

I'll give you one small aspect I have started looking at to rate the teams:

If the game is won in "extra innings", I factor it into ratings as a "tie/draw".  The top team in the league is playing a mediocre team and it took them "extra innings" to win.  What does that say?  Sort of similar if Man U take 87 minutes to beat QPR and score 2 goals.  The result says 2-0 but how did they perform?


It's a matter of making your assessment and looking for value.  The ratings are just a guide that cuts down the list from dozens of games a week to a handful worth having a punt on.

Are all the OPTA stats available to the public again yet?

Would love to get my hands on all the pass completed type stuff

re: MONEYBALL.  I haven't read the book but the movie does point out one clear message about pitchers and performance - the movie never mentions pitchers in the buy/sell scheme of baseball.

That is interesting to me. The book did mention pitchers in their draft but never went into great detail. I suspect Beane only told Lewis what he wanted him to know, and didn't give away the keys.