Arbusers

ex-hft wrote: ↑

Mon Feb 05, 2024 8:38 pm

That depends on your edge and the prices which you are betting.
If your edge is good (5% or more) and the prices around equal, then it’s a pretty good number.

There are two ways to figure this out: if you’re familiar with statistics, then you probably know how to calculate the distribution of the average of 1000 Bernoulli variables - it is approximately Gaussian, with a mean determined by your edge, and a variance of p*(1-p) / 1000.

More generally - or if you’re suspicious of the math: you can simulate many games with your edge and prices which you bet, and see what the distribution of outcomes looks like.

francesco wrote: ↑

Wed Feb 07, 2024 6:17 pm

Is it possible that a statistical model works in one championship and doesn't in another championship?...I'm always talking about football.... I started comparing the odds with the % compared to those of the bookmakers and I'm seeing that in some championships it is more feasible and in some absolutely not

ex-hft wrote: ↑

Wed Feb 07, 2024 9:35 pm

francesco wrote: ↑

Wed Feb 07, 2024 6:17 pm

Is it possible that a statistical model works in one championship and doesn't in another championship?...I'm always talking about football.... I started comparing the odds with the % compared to those of the bookmakers and I'm seeing that in some championships it is more feasible and in some absolutely not

Maybe, it could be.
See what happens if you look at the championships where your models performed well in the previous season. Do they also perform well this season?

ex-hft wrote: ↑

Thu Feb 08, 2024 6:07 am

If you can't program, but are intelligent enough in order to build a statistical model, and serious enough to evaluate its results in depth like you do, then, I would offer that learning to program could be your best investment.
Over the course of the next six months, you will end up significantly ahead than if you keep on doing everything by hand, and the skills will stay with you for life.

Anyway - regarding number of games - the more the better. I think that it's basically impossible to build a model for a single league (=championship) and predict its behavior out of sample. You need good principles which work across many such leagues, and to win in average over them.
However - it might be possible to spot leagues which are better or worse than others for your model - for example, it only works in South America, or it only works on women's leagues, etc.

This is a pretty deep subject - balancing true and false discoveries, and not chasing results which in truth appear only due to chance.

Lastly - regarding which odds: opening odds are easier to beat, closing odds are where you can bet more money. I would try to beat the opening odds first.

ex-hft wrote: ↑

Thu Feb 08, 2024 11:57 am

The problem is that you don't have enough statistics.
If there is a theory that should only hold true for a specific league (for example - "In the Premier League they now add longer injury times, and therefore there will be more goals"), then you might have no choice - but still, you will not be able to tell very well.

If you have whatever random theory, and you look at a history of a few hundred games, then 10-20% of the time it will win. And then you look at ten different leagues, and say that it only "works" for one of them, when in fact - the result might be completely random.

ex-hft wrote: ↑

Thu Feb 08, 2024 2:34 pm

Good luck. The test on the second championship is very important, it’s all too easy to lie to ourselves and then crash in reality.