Monte Carlo simulation for football forecasting is it possible

[Gooner80]

The title says it all really, if it is possible has anyone had any success with it and is it worth the time,from my understanding if a variable is slightly wrong would that result in huge errors in the simulation

[Datapunter]

Re: Monte Carlo simulation for football forecasting is it possible Monte Carlo is not a prediction tool, it's a simulation of a situation whereby you would otherwise require huge time and/or effort to measure the same with real-life data. It can only tell you the past.

[Evertonboy]

Re: Monte Carlo simulation for football forecasting is it possible I think Im right in saying that the Racing & Football outlook paper use this technique to compile there own odds based on there ratings

[Gooner80]

Re: Monte Carlo simulation for football forecasting is it possible Hi data punter if monte carlo simulation can only tell you the past, could it be used to give you a more accurate view of the past like simulating a teams attacking performance thousands of times rather than taking an average

[Datapunter]

Re: Monte Carlo simulation for football forecasting is it possible Any Monte Carlo simulation has by definition an element which is random. It is the random aspect of the subject you are researching that prompts the simulation in the first place. If your information is fixed, or you can somehow define it's boundaries clearly then you can calculate to a reasonable degree of accuracy any relationship of cause and effect. If you have any aspect that is random or cannot be determined you have a problem as there is nothing to base your calculation on. You could try to find an answer by experimenting but that could prove costly, both in time and finances, if even at all possible. So you simulate with a model of the given situation and repeat within the set boundaries. Given enough repetitions you will find out what will happen on average. Here's the key, on average. It says nothing about the next individual situation. Flip a coin, heads or tails. Each having a 50% chance. Put this in a Monte Carlo simulation and run it 1 million times. It will confirm that on average you end up with 50% heads and 50% tails. Doesn't tell you anything about the next flip, only that on average it's 50%. Bit more practical now. You anticipate next football season to have 1000 bets. You start with a bank of 1000 units, stake 10 on each bet. With each bet you could win between -1% and +3% (random) What is the risk of going bancrupt in that season ? A Monte Carlo simulation will tell you that given 1 million seasons you will go bancrupt in 2% of them. Which is quite usefull to know. Doesn't tell you anything about how next season is actually going to go. Only that you have a 2% chance of going bancrupt. The key element here is "could win between -1% and +3%" any simulation is only as good as the data you put in. If you don't know your expectation or you have wrong numbers you're ******. In sports betting so many aspects are random or interlinked you cannot set simple boundaries.

[v-zero]

Re: Monte Carlo simulation for football forecasting is it possible I use monte-carlo simulation in sports with relatively complex scoring structures. I have tried it in football, creating a model that simulates games where the assumption isn't made, as it is with the poisson distribution, that the probability of scoring at any time in the game remains constant, however this did not return probabilities that were any more than very marginally different to poisson, so I use poisson, as it is faster and less computationally expensive. As for other sports. Tennis, NFL, NBA and some other sports have complex enough scenarios to devote monte-carlo methods to. For instance, I model an American football game based on about eight variables. I create two hypothetical league average teams, and then 'play' a set number of games to derive underlying probabilities within the game. For instance, from this I can get probabilities from point spreads. If I rate Green Bay as being 2.5 points better than Pittsburgh, then I can take that 2.5 points and use it as a variable in my simulation, and get back what that 2.5 points says about the win probability of each team. In that instance, 2.5 points better becomes a 57% chance of Green Bay winning. That's useful, that means I can bet in the match odds market, not just against the spread. These monte-carlo simulations, these complex probability distributions, as that's what they are, are incredibly useful in situations like this. Being able to take expected points for each team and convert them to win probabilities is very powerful, and allows you to model the scores in order to get probabilities, rather than using a method such as Elo that largely only cares if a team wins or loses. As such, I use monte-carlo simulations alongside iterative solutions to both total points/goals and the ratio of one teams expected scoring relative to the other.

[Gooner80]

Re: Monte Carlo simulation for football forecasting is it possible trawling the internet it seems a few are using a form of monte carlo simulation in their forecasting. To me it looks possible because you can plug probabilities into it, and I get what Data punter is saying, but my assumption is that most teams are inconsistent and if you have two teams although we know how they perform on average I was hoping to run several simulations of how Team A for example might perform whether it is, exceptionally well, average, or really poor and then do the same for team B and then put the various permutations together and see what comes out. obviously it is really time intensive but in my experience quality is always much better then quantity. (on a side note I have looked at draws in matches on a line of best fit and apart from at the very extremes of the scales they look/ APPEAR completely random and from what I have tested even the bookies seem to get the prices wrong) Im gonna come up with something

[Datapunter]

Re: Monte Carlo simulation for football forecasting is it possible Let me put it this way. if you want to build a house you'll need a toolbox. A whole toolbox with 100 tools. Can't build a house with just one hammer. Monte Carlo is one tool in the box that has it's uses, but also it's limitations. I can't see it as a prediction tool in itself, just as something that can be used to research some of the aspects involved in what you try to predict.