Football Musings

[wearthefoxhat]

Took a look at the Dixon-Coles model, after watching Tom Brownlee's instructional you-tube video. (19.42 minutes)

He also gives a link to a free spreadsheet with all the formulas.


https://www.youtube.com/watch?v=SiBhLYf8YJ4

[Euler]

AI can do a great job of creating models now, as long as you ask it for specifics. It can also be produced it in a self-contained HTML file to make things nice as simple.

[wearthefoxhat]

AI can do a great job of creating models now, as long as you ask it for specifics. It can also be produced it in a self-contained HTML file to make things nice as simple.

Thanks, I'll take a look.

[wearthefoxhat]

Cobbled together the Dixon-Coles model as previously posted, and added it into my existing excel framework. Looking at the Championship games from last night and tonight, this was the output. The data (xG) is my own, based on certain in-play metrics, but they're similar to what's out there already on SofaScore...etc

Week 28.png

The ticks show the tissue to be greater than the live odds, but the bigger value ones >= 10% edge, are brought more to the fore.

The top 3 are highlighted for quick reference.

[Euler]

Football is one of those sports where variance dominates a lot. You’ve got very few scoring events, long periods where nothing happens, and then one incident, a deflection, a refereeing call, an early red card, completely changes the shape of the game.

From a modelling perspective that makes it extremely noisy. You can have the right read on a match but one one low-probability event swings everything.

That doesn’t mean modelling is pointless, but it does mean you have to be realistic about what it can deliver. Models are much better at describing long-term behaviour and edges across many matches than they are at predicting the outcome of a single game.

[wearthefoxhat]

Football is one of those sports where variance dominates a lot. You’ve got very few scoring events, long periods where nothing happens, and then one incident, a deflection, a refereeing call, an early red card, completely changes the shape of the game.

From a modelling perspective that makes it extremely noisy. You can have the right read on a match but one one low-probability event swings everything.

That doesn’t mean modelling is pointless, but it does mean you have to be realistic about what it can deliver. Models are much better at describing long-term behaviour and edges across many matches than they are at predicting the outcome of a single game.

It's been said, all predictive models are wrong. (but some are useful) - Prof George Box. (I suppose I'll find out in the long run)

I built in predictive elements based on Home / Away records and strength of previous opposition, but as you say variance for one game can put the kibosh on a well structured trade, ending up being right, but with the wrong result.

I have noticed, where there's "value" in one market, ie: BTTS, it carries over into the other goal markets too. (common sense really)