Hello boys!,
I'm trying to write some computer code to compute reliable probabilities for soccer matches, in particular I'm interested in the probabilities of the 1, X, 2 for a given match.
So, for each match I obtain something like:
milan-inter: 1: 0.3 X: 0.3 2: 0.4
Now I want to understand if my probability generator is good. The way I do it is the following: I consider the probabilities of the true result (the one that actually came out) for all the matches the generator worked on. Then I take the geometric mean of these numbers (the so-called likelyhood). This is an estimator for the goodness of the generator.
Obviously, for the most stupid generator the predictions are:
milan-inter: 1: 0.333 X: 0.333 2: 0.333
and the performance of the generator (the geometric mean...) is 0.333.
A perfect generator would always guess the true outcome and so its
performance would be 1. So the performance is a number between 0.333 and 1.
Now, testing some simple ideas to generate the probabilities, I obtained a performance of something like 0,36 that seems a poor one.
My question is:
-what can be the performance for a very good generator (like a human been)?
-what are the estimators commonly used for the goodness for a probability generator ?
-how do the best generators work?
-are they human or are they computer programs?
-how do the betting corporations compute their probabilities (and odds)?
-how do they perform?
Thank you for your attention and time!
I'm trying to write some computer code to compute reliable probabilities for soccer matches, in particular I'm interested in the probabilities of the 1, X, 2 for a given match.
So, for each match I obtain something like:
milan-inter: 1: 0.3 X: 0.3 2: 0.4
Now I want to understand if my probability generator is good. The way I do it is the following: I consider the probabilities of the true result (the one that actually came out) for all the matches the generator worked on. Then I take the geometric mean of these numbers (the so-called likelyhood). This is an estimator for the goodness of the generator.
Obviously, for the most stupid generator the predictions are:
milan-inter: 1: 0.333 X: 0.333 2: 0.333
and the performance of the generator (the geometric mean...) is 0.333.
A perfect generator would always guess the true outcome and so its
performance would be 1. So the performance is a number between 0.333 and 1.
Now, testing some simple ideas to generate the probabilities, I obtained a performance of something like 0,36 that seems a poor one.
My question is:
-what can be the performance for a very good generator (like a human been)?
-what are the estimators commonly used for the goodness for a probability generator ?
-how do the best generators work?
-are they human or are they computer programs?
-how do the betting corporations compute their probabilities (and odds)?
-how do they perform?
Thank you for your attention and time!
Comment