[sharpcat]

Assuming you have an estimated win probability for two players to win a tournament is it safe to use those probabilities for drawing a match-up comparison line or is there too much variance between winning the tournament and winning a match-up to use those numbers?

[Optional]

Not the exact answer you want, but Ganchrow helped me work out calculations I use for assessing match-up bets that might help or be of interest here: http://www.sportsbookreview.com/forum/handicappe...-question.html

[u21c3f6]

Assuming you have an estimated win probability for two players to win a tournament is it safe to use those probabilities for drawing a match-up comparison line or is there too much variance between winning the tournament and winning a match-up to use those numbers?

This is something that I look at to find profitable wagers (hedges). The closer the tournament is to the final, the more highly correlated the odds will be. It is not that the odds are necessarily less correlated the further out you go, it is just that there are many more possible combinations of results that impact the odds.

When I do this, I am not looking to create a line, I am looking for comparative pricing inefficiencies that can be used to create profitable hedges. The problem with creating a line is that you don't necessarly "know" which line is correct, the current game line or the line for the game created with the use of correlated markets. All I want to find is a disparity in the pricing between the the highly correlated markets and then I try to figure out the best way to exploit that disparity that gives me the best risk/reward.

Joe.

[ForgetWallStreet]

This won't work very well for golf, if that's what you're wondering.

[buby74]

If the question is related to knock out tournament you cannot simply use the probability of team A or B winning the tournament as a prediction of team A beating team B in the first round.

This is because the odds of the stronger team (say team A) winning the tournament will also incorporate the higher probability of team A beating teams in subsequnet rounds than team B's chance of beating the same teams.

For example imagine a 32 team knock out tournament with a strong team A a weak team B and 30 average teams

Assume Team A has a 70% chance of beating B and a 60% chance of beating any of the 30 average teams. Team B has a 30% chance of beating A and a 40% chance of beating any of the others.

To win the tournamnet team A must beat team B and 4 average teams which is 70%*60%*60%*60%*60%=9.1%
Team B must beat A and 4 average teams which is 30%*40%*40%*40%*40%=0.8%

If you only looked at the ratio of tournament winning probabilities you could conclude that a 9.1% team will beat a 0.8% team 9.1/(9.1+0.8) =92.2% of the time. When in fact we know this should be 70%. So this approach will always overvalue favourites.

I have assumed that a team's ability stays constant in the tournament so if B happened to destroy A in the first round people wouldnt change their views of team B.

So how could you use the tournamnet winning odds to predict the match ups. You could use excel solver if you have a good model of how a teams "power ranking" relate to their chance of beating other teams. You could assume a log 5 type approach where in the above example team A is 150 team b is 66.67 and the average teams are 100 and your win probability of A beating B is 150 / (150 +66.67)

if you know the draw you can work out how likely different teams are of meeting each other in each round and use solver to set the power ratings so that the probability of winning the tournament equals the probabulity implied by the odds and then use the power ratings to handicap each game.

I am intrigued by u21c3f6 approach as I am trying to do something similar with my baseball model. I have used sabermetrics to predict how likely each possible score is in baseball once i know the power ratings of each team (taking account of the lack of the bottom of the 9th, the walk off run rule and the exception for HRs was a b**ch)

To actually work out the power ratings instead of assessing pitchers, park factors, trends etc etc i use the moneyline and totals and use excel solver to set the power ratings. This allows me to look for +ev in the runlines when i know what my model says they should be given the moneyline and totals. I am in essence assuming the runline will be inaccurate and putting my trust in the moneyline and runline.

THis brings us to u21cf36 issue of how do you know which line to trust when using it to derive a fair price for a related line. I go with the money! as more money is bet on the moneyline and totals then the runline I predict they will be more accurate. I will also look at alternative runlines who should be more variable as there is even less money bet on them.

For NFL I would go with the spread and totals and use that to calibrate the moneyline for the same reasons. I have tried a similar approach for NFL but my model must be wrong as it predicts massive +evs for almost every game. This type of approach should only give a samll edge in a few games or else it is almost certainly wrong.

Most days I find a 1-2% edge in the runline of one MLB game every day but I don't have a database of historical runlines moneylines and total lines and the final result to see if my model is correct or profitable.


james

[buby74]

Here is a simple spreadsheet for a 16 team tournament that allows you to calculate the odds of the round of 16 matchup based on the odds of winning the tournament taking the issues i raised in my last reply into account.