It's a Mug's Game Part Deux

comment by Mr. Eboue Emmanuel (U12374)
posted 7 minutes ago
With these things the more data you have the better. Only if your model is good of course.
----------------------------------------------------------------------
Within reason, https://betegy.com use a load of different factors, I'm sure it goes into the dozens. Football is so inherently influenced by the random it makes prediction difficult and the search for minute variables is probably of limited value Vs say using the bookmaker who offers the best odds. It's counter intuitive but it's not about picking winners, it's about picking winners that win more frequently than the bookmakers odds would suggest, but might in principle lose more often than win.

Of course - what I meant is that if you have a stable good model and you feed it with more data of the same kind it will get better and better. Lots of small random variables would rather confuse your model but more of the same meaningful stats is always better. At least in ML. Feed it with bs data and you get bs results.

comment by Admin1 (U1)
posted 16 hours, 18 minutes ago

----------------------------------------------------------------------

I was wondering this but then looked at the figures and decided it wasn't really worth it.

Take Inter for example - the best team to bet against since the creation of JA:

You make £62 profit over 251 games.

That's 24 pence profit per game...

Over a season it would be worth it (e.g. Chelsea this year) but you don't know the info at the start of the season.
----------------------------------------------------------------------
That's an absolutely incredible return. Its 25% which is astounding, If i could hit a 25% return id retire within a couple of years.
----------------------------------------------------------------------

When you put it in percentage terms, that is a very attractive return I agree.

Suppose you need to bet big to make it worthwhile, which rules me out as I'm the most casual / smalltime betting man ever, but if you're doing it seriously then yeah a consistent 25% over a few years worth of markets would be gold

comment by NotSoMagicJuande (U1913)
posted 1 minute ago
comment by Admin1 (U1)
posted 16 hours, 18 minutes ago

----------------------------------------------------------------------

I was wondering this but then looked at the figures and decided it wasn't really worth it.

Take Inter for example - the best team to bet against since the creation of JA:

You make £62 profit over 251 games.

That's 24 pence profit per game...

Over a season it would be worth it (e.g. Chelsea this year) but you don't know the info at the start of the season.
----------------------------------------------------------------------
That's an absolutely incredible return. Its 25% which is astounding, If i could hit a 25% return id retire within a couple of years.
----------------------------------------------------------------------

When you put it in percentage terms, that is a very attractive return I agree.

Suppose you need to bet big to make it worthwhile, which rules me out as I'm the most casual / smalltime betting man ever, but if you're doing it seriously then yeah a consistent 25% over a few years worth of markets would be gold
----------------------------------------------------------------------
Have in mind that those are after the fact numbers. They can change in the future. Survivorship bias at full strength here

I actually tried a betting strategy like that (in simulator of course) and it didn't make any profit. It just placed bets on the most profitable teams using data for a given period back. No value for me there.

This is a Leyton Orient WUM article

comment by Mr. Eboue Emmanuel (U12374)
posted 25 minutes ago
comment by NotSoMagicJuande (U1913)
posted 1 minute ago
comment by Admin1 (U1)
posted 16 hours, 18 minutes ago

----------------------------------------------------------------------

I was wondering this but then looked at the figures and decided it wasn't really worth it.

Take Inter for example - the best team to bet against since the creation of JA:

You make £62 profit over 251 games.

That's 24 pence profit per game...

Over a season it would be worth it (e.g. Chelsea this year) but you don't know the info at the start of the season.
----------------------------------------------------------------------
That's an absolutely incredible return. Its 25% which is astounding, If i could hit a 25% return id retire within a couple of years.
----------------------------------------------------------------------

When you put it in percentage terms, that is a very attractive return I agree.

Suppose you need to bet big to make it worthwhile, which rules me out as I'm the most casual / smalltime betting man ever, but if you're doing it seriously then yeah a consistent 25% over a few years worth of markets would be gold
----------------------------------------------------------------------
Have in mind that those are after the fact numbers. They can change in the future. Survivorship bias at full strength here

I actually tried a betting strategy like that (in simulator of course) and it didn't make any profit. It just placed bets on the most profitable teams using data for a given period back. No value for me there.
----------------------------------------------------------------------
Totally agree. Though the entire investment banking industry is built upon the same fallacy. The superstar funds and fund managers always face regression to the mean given long enough

comment by Nostraduncus (U11713)
posted 21 hours, 49 minutes ago
just watch who you bet with.

the inlaws do theirs at Willy hills and i do mine on sky bet.

ive checked their bets on sky and the difference is huge.

they were getting 800 quid at willi hills but 1200 for exactly the same on sky bet
----------------------------------------------------------------------

Do they do their bets at Hills in shop? I've always thought Hills offer very competitive odds online, but in shop (as per all bookies) they generally print terrible odds on their coupons.

comment by Mr. Eboue Emmanuel (U12374)
posted 11 hours, 22 minutes ago
comment by Admin1 (U1)
posted 13 hours, 8 minutes ago
comment by Mr. Eboue Emmanuel (U12374)
posted 24 minutes ago
Don't know much about those will definitely look at them. Is there any python library that support those? I've been playing for some time throwing opta data at machine learning algos with variable degrees of success. For example trying to build expected goals models using stuff like distance/angle to goal, how the shot was assisted (counter attack, free kick, dribble), current game state, etc. I was thinking that if I derive expected goals for / against and then plug them in a poisson equation would be able to get something closer to true match odds. My best results sees to be using gradient boosting / xgboost getting to 3-4% of the bookmakers odds which is not bad but of course not good enough. It's a fun hobby but sadly don't have much time these days.
----------------------------------------------------------------------
Not sure about python, my initial version was written in C/C++ and the algorithms the same(http://lancet.mit.edu/galib-2.4/Overview.html etc), but each week it was taking me an hour or two of prep work to get data loaded and stuff. So i bit the bullet and started redoing it in java, and been at it since early 2016ish
----------------------------------------------------------------------
Thanks. Never used genetic algorithms before. I'll definitely give them a try. It would be interesting if they are better than ML at this task.
----------------------------------------------------------------------
For Python these are worth a look https://developers.google.com/optimization/

Would definitely check it out.

Any updates?