PDA

View Full Version : The scientific calibration of footy tipsters


wadood
05-09-10, 07:17
There are many good tipsters in this forum and probably in some others.
But the method of scoring is not scientific.
A uniform method is needed as well as a scientific one to measure their performences.

The method to achieve this is every tipster should give his predictions as percentage probabilities like this:

match A: 70% - 20% - 10%
match B: 45% - 35% - 20%

The result he wants us to bet could perhaps be indicated by bold, or red, or underlign.

What is the method of calibration then ?

Suppose we have two forecasters as follows:

Tipster 1:

match A: 70% - 20% - 10%
match B: 45% - 35% - 20%
match C: 15% - 30% - 55%
match D: 60% - 30% - 10%

Tipster 2:

match A: 90% - 05% - 05%
match B: 30% - 30% - 40%
match C: 20% - 40% - 40%
match D: 70% - 25% - 05%

and suppose the result line is 1-X-2-X

Then tipster 1 score is = (0.70 x 0.35 x 0.55 x 0.30 ) ^ (1/4) = 0.4484 or 44.84%
and tipster 2 score is = (0.90 x 0.30 x 0.40 x 0.25 ) ^ (1/4) = 0.4054 or 40.54%

Because the series of numbers thus obtained may be long we use logs:

score = exp [ SUM OF ( log probability ) / no of matches with known result ]

This is the scientific method so we know who are the best tipsters without dispute.

Effectively we are determining what might be called "a mixed mean Bernoulli number" for each one. Every match is a Bernoulli probability experiment after all and what the tipster does is strive to approach certainty with his predictions. Certainty lies at infinity of course and requires all the predictions to be of the 100-0-0 type and all correct.


* It is quite possible and quite likely indeed that the same folks who are now nominated as champion tipsters will emerge as champions again, but we should be given a real figure of merit, using the above rigorous mathematical formularism, rather than ad hoc.

grefstad
05-09-10, 10:43
You got a good point here, but adding such calculations would be difficult for BA; since the betting software we use for the forum contests is third party software.
We dont have any say on the programming in this software, although I guess we could perhaps influence it a bit.

The betting contests in this forum (I am talking about the score forcasts now) is anyway a bit based on luck, because predicting scores is far more complicated than just predicting the final uutcome based on 1,X,2.
However, mathematically most football matches are determined by a single goal margin, so if you bet on 1 or 2, it is probably wise to bet on a 1 goal win for either team, since you will get 2 points if you bet in the correct outcome with the correct goal difference.

Points for a completely wrong bet : 0
Points correct match winner: 1
Points correct winner and goal difference: 2
Points correct winner and result: 3

wadood
06-09-10, 03:45
We bet on everything, even yellow cards so my scheme is for final result only.
I have done it in the past though, for a national newspaper. Once with a method such as yours, the second time with my method.
Our so called tipsters were also doing it for advertising purposes, because they were paying some money for display space for their businesses in the journal.
In our case it seemed more like a case of those who were keen doing well and those who were lazy doing bad.
But it was useful.

This is the recent work of French university on the subject:

http://www.econ.ku.dk/det/workshop/Papers/Gergaud.pdf

I have done such work myself.
It is a strong prediction method.
Some people don't like because they say this: "I want to be involved with the real machinery myself i.e. study the facts and figures of football. It's not fun to follow others".
I agree with them also, but as I say what really matters in the end is winning or losing.