[goty0405]

Hey guys,

As I mentioned in my other probability thread on NRL I have been building a roughly generic sports modelling tool to analyse data and return estimated probabilities for upcoming sports matches. It has been going alright for NRL and I am looking to expand into AFL.

I have an alright background with AFL but definitely not as much knowledge as I have with the NRL so I thought I would see if any AFL experts have opinions on the following:

1. The significance of the last X results - should the last match this team played be given more weight than the match before that which is given more weight than the match before that? If so by how much is each recent match worth compared to the previous?
2. The significance of a result at home or away - If the model is considering A v B should it give more weighting to A's home matches and B's away matches. If so how much weighting? Or are all past games created equal?
3. Home much is home ground advantage worth? Does it differ depending on the teams (i.e. a WA team traveling to Melbourne)

I have some ideas on this myself but I'm playing around with the data and the results don't always match my expectations, so I am interested in any ideas people might have. Also if you have an idea on other things that may contribute to predicting AFL matches let me know. I will be doing a list of future additions to make so any idea is a good one.

[donkson]

1 - Basically spot on, possibly give or take a little due to travel factors etc
2 - Interstate records are still quite important, but probably not as important as many seem to think. Good teams have good records, bad teams have bad ones, nothing amazing. But what usually happens is that poor teams in general will have shocking interstate records.
3 - Yes, but again don't get too carried away with it. But the WA teams to tend to have a more pronounced advantage playing in Perth than what Sydney would at home.

[goty0405]

Thanks Donkson. My thoughts are confirmed and I've been backtesting the simulations and getting alright results. I will post some details when I am arrive at settings which are good enough to be considered the "base" model.

[goty0405]

Ok let's put up the simulation results for the current round. I just had a quick check at the AFL opening spreads thread and it looks like all of my predicted favourites are actual odds favourites so that is positive sign. The key point will be to see if my predictions can find an edge in a few games where the odds favourites have been massively over-rated.

Round 18

St Kilda vs Adelaide
S: 60.84%
A: 35.58%
D: 3.58%
Avg Points: 168.5

North Melbourne vs Brisbane
N: 64.38%
B: 32.27%
D: 3.35%
Avg Points: 187.5

Sydney vs Western Bulldogs
S: 64.59%
W: 32.16%
D: 3.25%
Avg Points: 176.5

Gold Coast vs Collingwood
G: 0.78%
C: 99.00%
D: 0.22%
Avg Points: 177.5

Essendon vs Carlton
E: 29.07%
C: 67.52%
D: 3.41%
Avg Points: 178.0

Geelong vs Richmond
G: 71.25%
R: 25.67%
D: 3.08%
Avg Points: 187.0

Melbourne vs Hawthorn
M: 42.15%
H: 53.99%
D: 3.86%
Avg Points: 176.5

Fremantle vs West Coast
F: 28.08%
W: 68.69%
D: 3.23%
Avg Points: 179.0

[donkson]

Interesting to see how this goes, a couple of those numbers are odd, but in general looks good.

[goty0405]

Which numbers do you think are odd? I'm very interested to get feedback as I've only been partially keeping track of the AFL this season so I'm not as clued in on picking the outliers in my numbers.

[benrama]

You should google "Stephen Clarke", he's been modeling the AFL for 30 odd years.

He has a model that takes into account a teams last 8 games to come up with a ranking from first to last. It then gives a relative home advantage. This is briefly described here:

Things to keep in mind with weighting ground advantage:

- Subiaco and Skilled Park are disproportionately advantageous to Freo/WC and Geelong
- Home ground advantage needs to be adjusted for inter vs intrastate
- Etihad and MCG are home games for multiple teams, so home advantage is less when both teams are playing

Specifically regarding your questions:

1) I would weight each of the previous X games equally, as many anomaly games occur even when a team is on a good or bad run. Could be worth testing an alternative though

2/3) For a home team, I would take into account ALL home games, but for away games you need to choose what is relevant - I would do what Clarke did which is combine this into one "home ground factor"

[donkson]

Which numbers do you think are odd? I'm very interested to get feedback as I've only been partially keeping track of the AFL this season so I'm not as clued in on picking the outliers in my numbers.

To be fair there is only one game that I'd suggest is 'wrong' as such, the Hawthorn/Melbourne match. I'd say Hawthorn are very very strong favourites there.

Geelong is basically a lock, but I can see why your model suggests otherwise, they have been average in the past few weeks.

[benrama]

A read you might find interesting, it was by someone that studied under Clarke:

http://www.google.com.au/url?sa=t&so...t43bKy92MZ5akg

[donkson]

Nice link thanks, I used to follow Clarke about 2 years ago or so, when he used to actually post odds and more analysis. Unfortunately he's toned down what he posted.

[benrama]

Yea, I think you now only get more detail from him by paying for his picks, e.g. through smartgambler, which I'm never going to do. Been burned once paying for someone's picks, never going to do it again.

I still think situational capping works best, and choosing your spots. I use primarily , look at how the teams have fared during the season, check for who's in and who's out, evaluate relative home ground advantage and go from there. Don't track properly how my results are throughout the season but my account is healthy.

[goty0405]

Wow some great resources provided there. Thanks! I'll be busy reading up on it for a while now

Interesting to see that the Swinburne computer system is tipping at 68%. I've got my model up to about 65% (when excluding the first 3 rounds) and about 75% when using specific rules to pick bets (making about 80-100 bets per season).

Thanks for the feedback on how to model the home/away factors. It is pretty much what I was thinking and although my model is still very simple I think one of the next additions is a more complexity in the Home/Away ratings, so you've given me good food for thought.

[Lookingtostart]

Have you thought about including the Perth factor?

[goty0405]

Have you thought about including the Perth factor?

I don't currently specify the "Perth factor" but I do cater for more significance of home performances when a team is player at home and vice versa. I think actually paying attention to the specific venue seems to be the best idea for future changes I should make.

[goty0405]

I'm only making plays on this when the odds present value, so there was no action last night as the Saints were very short.

So far today I've gotten on the Swans @ 1.59 and there might be a few other plays to come.

[goty0405]

I was out with my lady friend last night so I didn't get to update here but there were a couple of plays that were good value:

Collingwood (vs GC) @ 1.04
Carlton (vs Ess) @ 1.51

And today there is only one value play:

West Coast (vs Freo) @ 1.53

[goty0405]

West Coast managed to fulfill the perfect round for me...but only just!

The results:
Predictions: 8/8 (+2.39 units)
Actual Bets: 4/4 (+1.67 units)


A very auspicious start but picking 8 favourites isn't something to get too excited about yet. Let's see how it goes next week.

[goty0405]

Round 19...

North Melbourne vs Carlton
NM: 47.35%
Car: 52.65%

Western Bulldogs vs West Coast
WB: 42.91%
WC: 57.09%

Geelong vs Melbourne
G: 65.60%
M: 34.40%

Gold Coast vs St Kilda
GC: 27.75%
StK: 72.25%

Fremantle vs Hawthorn
F: 41.61%
H: 58.39%

Collingwood vs Essendon
C: 70.58%
E: 29.42%

Adelaide vs Port Adelaid
A: 58.04%
PA: 41.96%


Some pretty interesting predictions there. I had a quick look at the opening lines and it seems that all my predictions are favourites but interestingly most aren't predicted as strong as the opening odds indicate. So, if the books are the "source of truth" then it looks like my model may be under-rating some of the favourites. We'll see how it pans out.

[goty0405]

No play tonight as there is no value in Carlton ~1.30.

In fact it looks like there wont be any plays for the weekend based on the current odds on offer.

[goty0405]

Not much on tonight too. I am hoping WC and Hawthorn drift out a bit to my value threshold but otherwise nothing doing.

[goty0405]

Another round and another series of good predictions. There were no bets as I'm only betting when value is on offer when compared to my simulation and this week saw a lot of short priced favourites.

This Round:
Predictions: 7/7 (+2.44 units)
Actual Bets: 0


Overall:
Predictions: 15/15 (+4.83 units)
Actual Bets: 4/4 (+1.67 units)

[donkson]

Shouldn't you be taking some dogs if you're only betting value?

[goty0405]

If I was betting every game then yes I'd be on the dogs a lot...and the recent results show why that isn't a very good idea.

I'm only betting on the team that my simulation says is the most likely winner and only if I get value for them.

[goty0405]

St Kilda vs Fremantle
S: 73.64%
F: 26.36%

Carlton vs Melbourne
C: 75.61%
M: 24.39%

Geelong vs Gold Coast
Ge: 84.17%
GC: 15.83%

Essendon vs Sydney
E: 36.03%
S: 63.97%

Port Adelaide vs Collingwood
P: 11.81%
C: 88.19%

Brisbane vs Adelaide
B: 53.25%
A: 46.75%

Hawthorn vs North Melbourne
H: 75.07%
N: 24.93%


I've made a slight modification to how the simulation works over the past week. The results have been good when back-testing and would have even resulted in a bet or two last week. Fingers crossed the good form continues.

[kingsr]

Swans is the play!

[goty0405]

Swans is the play!

Yep they definitely looking under-valued in the market compared to my simulation results. Go Swannies!

[goty0405]

Swans will be the play today but just watching their price at the moment to see if they drift any further. The market history image also shows a positive tick for the concept of starting off with massively inflated prices when using an exchange just in case you get some silly action...


So if you liked the Swans you could have thrown a bet in at 2.50 and now you'd be sitting back with a pretty big +EV bet.

[goty0405]

Ah what a game. I mus admit I don't watch too much AFL and especially not if there is NRL on at the same time. But I started the night flicking over to get score updates and ending up watching the final quarter. And when Goodes took that final mark I was pretty confident.

Oh well, I'm very happy with my bet (I ended up getting the Swans @ 2.0) and I think that those odds were good value, even though the result did not go my way.

[goty0405]

This Round:
Predictions: 6/8 (-1.16 units)
Actual Bets: 0/1 (-1 unit)


Overall:
Predictions: 21/23 (+3.67 units)
Actual Bets: 4/5 (+0.67 units)


Two games went against me this weekend but the good news is that they were predicted to be the two closest games from my simulations so that's quite positive. The Brisbane-Adelaide games was estimated to be pretty much a coin-flip (53:47) so the only real problem was Sydney losing as I had them as much stronger favourites (64:36) than the game suggested.

[goty0405]

St Kilda vs Collingwood
S: 30.81%
C: 69.19%

Hawthorn vs Port Adelaide
H: 88.08%
P: 11.92%

Fremantle vs Carlton
F: 24.46%
C: 75.54%

Western Bulldogs vs Essendon
W: 57.43%
E: 42.57%

Brisbane vs Gold Coast
B: 63.25%
G: 36.75%

Melbourne vs West Coast
M: 18.55%
W: 81.45%

Richmond vs Sydney
R: 29.61%
S: 70.39%

Adelaide vs Geelong
A: 18.9%
G: 81.1%

[goty0405]

Looks like there will be a couple of plays today with Carlton and Western Bulldogs showing value in the market.

[goty0405]

This Round:
Predictions: 6/8 (-0.65 units)
Actual Bets: 2/3 (-0.11 unit)


Overall:
Predictions: 27/31 (+3.02 units)
Actual Bets: 6/8 (+0.56 units)

[goty0405]

Round 22...

Carlton vs Hawthorn
C: 46.27%
H: 53.73%

Gold Coast vs Adelaide
G: 31.61%
A: 68.39%

West Coast vs Essendon
W: 65.41%
E: 34.59%

North Melbourne vs Fremantle
N: 63.09%
F: 36.91%

Collingwood vs Brisbane
C: 91.61%
B: 8.39%

Sydney vs St Kilda
Syd: 40.08%
StK: 59.92%

Port Adelaide vs Western Bulldogs
P: 27.27%
W: 72.73%

Melbourne vs Richmond
M: 46.73%
R: 53.27%

[goty0405]

This Round:
Predictions: 7/8 (+1.51 units)
Actual Bets: 0


Overall:
Predictions: 34/39 (+4.53 units)
Actual Bets: 6/8 (+0.56 units)

[goty0405]

Fremantle vs Collingwood
F: 12.65%
C: 87.35%

Hawthorn vs Western Bulldogs
H: 67.45%
W: 32.55%

Geelong vs Sydney
G: 74.98%
S: 25.02%

St Kilda vs North Melbourne
S: 51.45%
N: 48.55%

Brisbane vs West Coast
B: 26.40%
W: 73.60%

Adelaide vs Richmond
A: 53.90%
R: 46.10%

Melbourne vs Gold Coast
M: 71.54%
G: 28.46%

Essendon vs Richmond
E: 78.79%
R: 21.21%