Handicapping/modelling AFL

[Butane]

I plan on looking into this for next year. Not sure how much value it will add though.

Extremely speculative I agree. That's why I think modelling the Brownlow is difficult. Of course, if it is difficult then it also difficult for the bookies/punters to get the price right so there should be value somewhere.

 

[robertbn]

I actually tend to think modelling the Brownlow wouldn't be too hard. There is enough data to see what kinds of games get votes (high possession, high contested possession?), and you could correct each player's votes according to their history (Judd +10 votes!). Unfortunately, because it's a reasonably small sample each year, the variance is quite high. But that can probably work in your favour. I suspect Ablett was underpriced this year and a good model would have liked Swan and Selwood's chances, and both would have been value at ~$13 and ~$5 (even though they didn't win!).

 

[Butane]

Just a few questions for robertbn

If you know your model is roughly 10% more inaccurate than the bookies(as judging by the line) and then they add an overround of about 5%(just a guess) than does that mean you are only betting when your model suggests 15% or more difference?

Also, have you run the model and checked the odds/percentages for betting against the line?

I'm just a little stumped on where to head now that I have a model that can predict games at about the same margin of error as the bookies. As far as I can tell, all this means is that I'm just as likely to be wrong as I am right and eventually the bookies vig will eat me away. Even if I'm cherry picking games when my model shows a big advantage isn't it just as likely that it's me who is has handicapped the game poorly?

 

[Duritz]

Butane if you think you can predict to a 26 point MAE in 2014, PM me and I'll turn it into piles of cash for both of us.

 

[Butane]

Butane if you think you can predict to a 26 point MAE in 2014, PM me and I'll turn it into piles of cash for both of us.

2013 was a highly predictable year for all model predictors and being somewhere around the 27-29 was easily achievable by anyone running a simple regression model. It's unlikely we'll have year similar to it next season. But what's unpredictable for models is also just as unpredictable for the bookies. I still have a few tweaks I'd like to add like quantifying subjective things like teams strength or working out if 'days rest' can be quantified (early signs show that it has no worth) whilst not overfitting my data.

Rourke has already stated in this thread that he got his MAE down to 25.5 - which in my books in better than most bookies. Maybe give him a buzz for some free cash?

I will say though Duritz, I believe we can at the very least draw level with the bookies line. What ever that may be next year. My question to robertbn was - if we assume we're 0.1 behind the bookie can we profit? When the model shows a there's a value bet is it only us that sees the value from our perspective. I'm not a mathematician and I'm sure this is high school stuff.

(I'm exploring chance and luck in sport at the moment so if anyone has any good reading material I'd love to read it)

 

[Duritz]

2013 was a highly predictable year for all model predictors and being somewhere around the 27-29 was easily achievable by anyone running a simple regression model. It's unlikely we'll have year similar to it next season. But what's unpredictable for models is also just as unpredictable for the bookies. I still have a few tweaks I'd like to add like quantifying subjective things like teams strength or working out if 'days rest' can be quantified (early signs show that it has no worth) whilst not overfitting my data.

Rourke has already stated in this thread that he got his MAE down to 25.5 - which in my books in better than most bookies. Maybe give him a buzz for some free cash?

I will say though Duritz, I believe we can at the very least draw level with the bookies line. What ever that may be next year. My question to robertbn was - if we assume we're 0.1 behind the bookie can we profit? When the model shows a there's a value bet is it only us that sees the value from our perspective. I'm not a mathematician and I'm sure this is high school stuff.

(I'm exploring chance and luck in sport at the moment so if anyone has any good reading material I'd love to read it)

You don't have to be more accurate than the market to profit from it. So long as you find things it misses.

 

[Butane]

You don't have to be more accurate than the market to profit from it. So long as you find things it misses.

Which brings me to my next question. How much do the bookies move the line? What I mean by that is are they 100% about covering all risk and simply balancing everybody's money so that just take the vig or do they say at some point - our models are better than where people are laying their money so let's sting them for a few more percent? Of course there's a limit to how far they'll go with the second option but nonetheless, I feel I can out do the collective 53% of the time.

 

[Duritz]

Which brings me to my next question. How much do the bookies move the line? What I mean by that is are they 100% about covering all risk and simply balancing everybody's money so that just take the vig or do they say at some point - our models are better than where people are laying their money so let's sting them for a few more percent? Of course there's a limit to how far they'll go with the second option but nonetheless, I feel I can out do the collective 53% of the time.

It depends which bookie, and it depends from whom the money came.

 

[arkie]

What do most use to build their AFL models? I was thinking Excel, but perhaps Python might be a better option?

 

[Butane]

What do most use to build their AFL models? I was thinking Excel, but perhaps Python might be a better option?

I don't have the technical nouse to explore other options besides excel. Although there does seem to be a large amount of people using 'R' - http://en.wikipedia.org/wiki/R_(programming_language)

I don't understand the pros and cons of using such programs but for me, I just get an idea and then just put together in excel. I realise that this probably means I'm doing 10 times as much work and with greater scope for error but it does mean I don't have to learn any new skills for something that isn't anything more than a hobby.

For those that have used Python or 'R' - how difficult is it to pick up the basics?

 

[d3ft]

R is pretty easy and there are loads of resources, it's also free. I use a program called SAS, which I suppose is like the corporate version of R. You can think of them like the framework that sits under Excel (the GUI). They don't offer the point and click interface of Excel and almost all functions must be performed by writing code, even simple things like a pivot table.

The main benefits these programs offer over Excel are:

• Faster handling of large datasets - Excel slows down to a crawl when you are looking at a lot of data, even on a beefy computer
• Much better handling of simulations
• Lots of built-in functions so you won't have to write them yourself in Excel.

That said, if your model is fairly simple, Excel is much more intuitive. I tend to do the heavy lifting in purpose-built data analytics software (R/SAS) then move it into Excel for manipulation and presentation.

 

[arkie]

I'd like something to predicts margins and odds based on historical data between two sides that factors in home ground advantage. I don't want to include other stats at this point.

I can code in Ruby on Rails so Python won't be too difficult to learn. Would you recommend Python over Excel?

 

[CamTinley]

I plan on looking into this for next year. Not sure how much value it will add though.

I'm doing this atm. It's a pretty long and involved process and I'm only about a quarter of the way through. From what I've done so far a couple of interesting things have popped up that I was half expecting but it's too early to be sure.

A random unexpected vote swung this years overall result.

 

[brett128]

R is pretty easy and there are loads of resources, it's also free. I use a program called SAS, which I suppose is like the corporate version of R. You can think of them like the framework that sits under Excel (the GUI). They don't offer the point and click interface of Excel and almost all functions must be performed by writing code, even simple things like a pivot table.

The main benefits these programs offer over Excel are:

• Faster handling of large datasets - Excel slows down to a crawl when you are looking at a lot of data, even on a beefy computer
• Much better handling of simulations
• Lots of built-in functions so you won't have to write them yourself in Excel.

That said, if your model is fairly simple, Excel is much more intuitive. I tend to do the heavy lifting in purpose-built data analytics software (R/SAS) then move it into Excel for manipulation and presentation.

Excel is excellent for prototyping ideas before they are written in hard code. Whether that hard code ends up as written in R, SAS, or C# code. It's also handy to know a good DB language like Mongo or SQL (or in my case, have a programmer that knows DB languages). If you model for long enough, you will quickly realize Excel has its limits.

Palisade's 'Industrial Decision Suite' is also an excellent addition to Excel, and is a product I use frequently for sports modelling.

 

[pmalcolm1984]

Sharing ideas/info/data massive in this space. I have a brownlow model that is relatively accurate. However, my assumptions around its inaccuracy was overstated this year. Was pushing me to back guys at longer odds in the team votes as an example. If anyone is keen to take conversation off line, I'd be up for it. My programming is poor but I've been able to build good models using concepts from other sports, finance, etc.

 

[Kynge of Begrem]

Palisade's 'Industrial Decision Suite' is also an excellent addition to Excel, and is a product I use frequently for sports modelling.

An endorsement akin to Nero pimping fire prevention techniques or Amy Winehouse endorsing a guide to healthy living

 

[iluvparis]

Solid Nero reference