Algorithm revisited, after finals, week 3

[Hobbes]

It's a bit lame posting predictions for the first round while the round is nearly finished (as I type, Geelong are comfortably ahead of Freo in the last match of the round) but since the method is based on strict adherence to a formula, it shouldn't really matter.

Despite promising myself to refine it over the summer break, the method is the same as last year. The only change is how I'm handling last year's results - teams are given a null result for each round of the finals that they didn't play, so the results being used are the same for every team - from the last 9 results of the 2016 minor round.

The other, tiny change is that I won't tip draws - the numbers should spill a result 0.01 points one way or another, so I'll lean that way and tip a one point win.

The initial ratings are:

Sydney 32.8
Bulldogs 28.1
GWS 27.3
Geelong 17.8
Adelaide 15.8
West Coast 9.6
Hawthorn 6.8
Collingwood 4.58
St Kilda 4.57
Port Adelaide 1.5
Melbourne -4.4
North Melbourne -4.5
Richmond -6.3
Essendon -6.4
Carlton -6.6
Gold Coast -7.8
Fremantle -8.5
Brisbane -16.2

This would make the first round tips

Carlton v Richmond +1
Collingwood v Western Bulldogs +24
St Kilda +9 v Melbourne
Sydney +40 v Port Adelaide
Hawthorn +13 v Essendon
Gold Coast +8 v Brisbane
North Melbourne v West Coast +5
Adelaide v GWS +2
Fremantle v Geelong +17

That's looking a lot like 5/9 for winners. I'll keep track of win/loss and average error margin as the year proceeds.

 

[kaiserchief13]

It's a bit lame posting predictions for the first round while the round is nearly finished (as I type, Geelong are comfortably ahead of Freo in the last match of the round) but since the method is based on strict adherence to a formula, it shouldn't really matter.

Despite promising myself to refine it over the summer break, the method is the same as last year. The only change is how I'm handling last year's results - teams are given a null result for each round of the finals that they didn't play, so the results being used are the same for every team - from the last 9 results of the 2016 minor round.

The other, tiny change is that I won't tip draws - the numbers should spill a result 0.01 points one way or another, so I'll lean that way and tip a one point win.

The initial ratings are:

Sydney 32.8
Bulldogs 28.1
GWS 27.3
Geelong 17.8
Adelaide 15.8
West Coast 9.6
Hawthorn 6.8
Collingwood 4.58
St Kilda 4.57
Port Adelaide 1.5
Melbourne -4.4
North Melbourne -4.5
Richmond -6.3
Essendon -6.4
Carlton -6.6
Gold Coast -7.8
Fremantle -8.5
Brisbane -16.2

This would make the first round tips

Carlton v Richmond +1
Collingwood v Western Bulldogs +24
St Kilda +9 v Melbourne
Sydney +40 v Port Adelaide
Hawthorn +13 v Essendon
Gold Coast +8 v Brisbane
North Melbourne v West Coast +5
Adelaide v GWS +2
Fremantle v Geelong +17

That's looking a lot like 5/9 for winners. I'll keep track of win/loss and average error margin as the year proceeds.

Look at the results. That's 4/9 winners (providing Freo don't come back)

 

[Hobbes]

Look at the results. That's 4/9 winners (providing Freo don't come back)

Quite right. Fortunately, my computer is better at counting to ten than I am.

 

[Hobbes]

Round 1, results.

Carlton v Richmond +1 (42 out)
Collingwood v Western Bulldogs +24 (10)
St Kilda +9 v Melbourne (39)
Sydney +40 v Port Adelaide (68)
Hawthorn +13 v Essendon (38)
Gold Coast +8 v Brisbane (10)
North Melbourne v West Coast +5 (38)
Adelaide v GWS +2 (58)
Fremantle v Geelong +17 (25)

4/9. Not particularly impressive, but I think a few of these results were tough for tipsters. (Also, I expect my algorithm to improve with more data from this year.)

Average was about 36 points out. This will improve.

Ladder

1. Bulldogs 22.9 (+1)
2. Adelaide 21.05 (+3)
3. Sydney 21.04 (-2)
4. Geelong 17.8
5. GWS 16.3 (-2)
6. West Coast 12.5
7. Port Adelaide 10.5 (+3)
8. Collingwood 3.8
9. Melbourne 0.7 (+2)
10. Hawthorn -0.3 (-3)
11. Essendon -1.1 (+3)
12. St Kilda -1.4 (-3)
13. Richmond -1.8
14. Gold Coast -9.5 (+2)
15. North Melbourne -12.0 (-3)
16. Carlton -12.4 (-1)
17. Fremantle -12.9
18. Brisbane -13.5

(There's a bit of a positive skew at the moment, since the null results given to non-finalists were, effectively, a better result than they would have probably achieved while playing. This will ease with time).

Tips for Round 2

Richmond v Collingwood +5
Western Bulldogs +7 v Sydney
Hawthorn v Adelaide +15
GWS +35 v Gold Coast
Brisbane v Essendon +6
West Coast +23 v St Kilda
Geelong +30 v North Melbourne
Melbourne +13 v Carlton
Port Adelaide +36 v Fremantle

 

[eth-dog]

Interesting. Will follow.

 

[JHF1870]

I will follow this too. I think your results will improve after round 9 this year, moreso for teams that have fluctuated from last season.

I was looking at several different methods for predicting tips and not one was able to beat the bookmaker.

Was something like 73% for the Bookmaker, 70% for some mathematician in Melbourne.

 

[JHF1870]

http://www.matterofstats.com/mafl-s...urprising-things-ive-learned-about-afl-so-far

 

[Hobbes]

If anybody cares, the algorithm was covered in some depth in last year's thread

https://www.bigfooty.com/forum/threads/team-rating-system-update-after-finals-week-three.1128857/

The main difference this year is the null rounds which I used to handle the final series and the break between seasons. It means that, in the short term, results from this year will have a much stronger impact than results from last year. The side-effect is that there's a positive skew - the Mark 2 method I used last year would have produced a negative skew because the last match for 13/18 teams was a loss (obviously, including 7/8 finalists).

 

[catstorm]

Round 1, results.

Carlton v Richmond +1 (42 out)
Collingwood v Western Bulldogs +24 (10)
St Kilda +9 v Melbourne (39)
Sydney +40 v Port Adelaide (68)
Hawthorn +13 v Essendon (38)
Gold Coast +8 v Brisbane (10)
North Melbourne v West Coast +5 (38)
Adelaide v GWS +2 (58)
Fremantle v Geelong +17 (25)

4/9. Not particularly impressive, but I think a few of these results were tough for tipsters. (Also, I expect my algorithm to improve with more data from this year.)

Average was about 36 points out. This will improve.

Ladder

1. Bulldogs 22.9 (+1)
2. Adelaide 21.05 (+3)
3. Sydney 21.04 (-2)
4. Geelong 17.8
5. GWS 16.3 (-2)
6. West Coast 12.5
7. Port Adelaide 10.5 (+3)
8. Collingwood 3.8
9. Melbourne 0.7 (+2)
10. Hawthorn -0.3 (-3)
11. Essendon -1.1 (+3)
12. St Kilda -1.4 (-3)
13. Richmond -1.8
14. Gold Coast -9.5 (+2)
15. North Melbourne -12.0 (-3)
16. Carlton -12.4 (-1)
17. Fremantle -12.9
18. Brisbane -13.5

(There's a bit of a positive skew at the moment, since the null results given to non-finalists were, effectively, a better result than they would have probably achieved while playing. This will ease with time).

Tips for Round 2

Richmond v Collingwood +5
Western Bulldogs +7 v Sydney
Hawthorn v Adelaide +15
GWS +35 v Gold Coast
Brisbane v Essendon +6
West Coast +23 v St Kilda
Geelong +30 v North Melbourne
Melbourne +13 v Carlton
Port Adelaide +36 v Fremantle

Hey Hobbes why didn't Geelongs rating change at all despite exceeding expectations.

 

[Hobbes]

Hey Hobbes why didn't Geelongs rating change at all despite exceeding expectations.

My method rates each past result on the current rating of the opponents - not their rating at the time of the match. My best guess is that they suffered a little for Sydney's loss, since their previous match was the prelim loss to Sydney. (There's also the mild paradox that the rating for the team you just beat went down, and you get punished for that.)

 

[catstorm]

My method rates each past result on the current rating of the opponents - not their rating at the time of the match. My best guess is that they suffered a little for Sydney's loss, since their previous match was the prelim loss to Sydney. (There's also the mild paradox that the rating for the team you just beat went down, and you get punished for that.)

Yeah I get that, what I mean is Geelongs rating went from 17.8 to 17.8 literally every other team in the league changed at least a little bit Geelongs stayed exactly the same.

 

[Hobbes]

Yeah I get that, what I mean is Geelongs rating went from 17.8 to 17.8 literally every other team in the league changed at least a little bit Geelongs stayed exactly the same.

Actually, they're down about 0.02. I don't normally bother with the second decimal place unless it's necessary to clarify standings.

 

[catstorm]

Actually, they're down about 0.02. I don't normally bother with the second decimal place unless it's necessary to clarify standings.

Huh that's interesting I thought you must have missed them, guess it's just an anomaly.

 

[Hobbes]

Richmond v Collingwood +5
Western Bulldogs +7 v Sydney
Hawthorn v Adelaide +15
GWS +35 v Gold Coast
Brisbane v Essendon +6
West Coast +23 v St Kilda
Geelong +30 v North Melbourne
Melbourne +13 v Carlton
Port Adelaide +36 v Fremantle

8/9 winners, not too surprising on a weekend which largely adhered to the form guide. Average margin error was 27.7, not bad given that more than half of the total was the two big blowout results from GWS and Port.

1. Bulldogs 24.9
2. Adelaide 22.67
3. GWS 21.5 (+2)
4. Sydney 18.9 (-1)
5. Port Adelaide 15.2 (+2)
6. Geelong 12.8 (-2)
7. West Coast 12.3 (-1)
8. Melbourne 2.3 (+1)
9. Essendon 2.0 (+2)
10. Richmond 1.9 (+3)
11. Collingwood -0.1 (-3)
12. St Kilda -1.5
13. Hawthorn -2.5 (-3)
14. North Melbourne -9.3 (+1)
15. Carlton -13.3 (+1)
16. Gold Coast -15.7 (-2)
17. Brisbane -16.4 (+1)
18. Fremantle -18.2 (-1)

Sydney +24 v Collingwood
North Melbourne v GWS +26
Richmond v West Coast +2
Geelong +10 v Melbourne
Port Adelaide v Adelaide +8
Fremantle v Western Bulldogs +34
St Kilda +21 v Brisbane
Carlton v Essendon +15
Gold Coast v Hawthorn +7

 

[free at last]

Richmond v Collingwood +5
Western Bulldogs +7 v Sydney
Hawthorn v Adelaide +15
GWS +35 v Gold Coast
Brisbane v Essendon +6
West Coast +23 v St Kilda
Geelong +30 v North Melbourne
Melbourne +13 v Carlton
Port Adelaide +36 v Fremantle

8/9 winners, not too surprising on a weekend which largely adhered to the form guide. Average margin error was 27.7, not bad given that more than half of the total was the two big blowout results from GWS and Port.

1. Bulldogs 24.9
2. Adelaide 22.67
3. GWS 21.5 (+2)
4. Sydney 18.9 (-1)
5. Port Adelaide 15.2 (+2)
6. Geelong 12.8 (-2)
7. West Coast 12.3 (-1)
8. Melbourne 2.3 (+1)
9. Essendon 2.0 (+2)
10. Richmond 1.9 (+3)
11. Collingwood -0.1 (-3)
12. St Kilda -1.5
13. Hawthorn -2.5 (-3)
14. North Melbourne -9.3 (+1)
15. Carlton -13.3 (+1)
16. Gold Coast -15.7 (-2)
17. Brisbane -16.4 (+1)
18. Fremantle -18.2 (-1)

Sydney +24 v Collingwood
North Melbourne v GWS +26
Richmond v West Coast +2
Geelong +10 v Melbourne
Port Adelaide v Adelaide +8
Fremantle v Western Bulldogs +34
St Kilda +21 v Brisbane
Carlton v Essendon +15
Gold Coast v Hawthorn +7

Excuse my ignorance but for example what is sydney +24 v Collingwood actually mean?

 

[Hobbes]

Excuse my ignorance but for example what is sydney +24 v Collingwood actually mean?

I'm tipping Sydney to beat Collingwood by four goals.

 

[Vikingnz]

Hey Hobbes,

Can the algorithm easily be converted to a % chance of victory? Like each +1 equates to +2% chance?

 

[Hobbes]

Hey Hobbes,

Can the algorithm easily be converted to a % chance of victory? Like each +1 equates to +2% chance?

If I was attempting a probablistic ladder prediction, I'd probably work with 50% plus/minus the predicted margin.

 

[Hobbes]

Sydney +24 v Collingwood
North Melbourne v GWS +26
Richmond v West Coast +2
Geelong +10 v Melbourne
Port Adelaide v Adelaide +8
Fremantle v Western Bulldogs +34
St Kilda +21 v Brisbane
Carlton v Essendon +15
Gold Coast v Hawthorn +7

Well, the "round of upsets" got me like it did a few others. 4/9.

1. GWS 23.3 (+2)
2. Adelaide 21.5
3. Bulldogs 15.4 (-2)
4. Port Adelaide 13.9 (+1)
5. Geelong 12.5 (+1)
6. Sydney 10.9 (-2)
7. West Coast 9.2
8. Richmond 5.0 (+2)
9. Collingwood +2.5 (+2)
10. Melbourne +0.6 (-2)
11. St Kilda -1.5 (+1)
12. Essendon -3.1 (-3)
13. Gold Coast -5.6 (+3)
14. Carlton -9.6 (+1)
15. Fremantle -12.7 (+3)
16. North Melbourne -13.4 (-2)
17. Hawthorn -16.0 (-4)
18. Brisbane -17.1 (-1)

Sydney are slipping from their original lofty perch, but still rate as a good team. Hawthorn, on the other hand, appears to be going down the toilet.

West Coast +11 v Sydney
North Melbourne v Western Bulldogs +29
Melbourne +22 v Fremantle
GWS +18 v Port Adelaide
Carlton +2 v Gold Coast
Adelaide +31 v Essendon
Collingwood +4 v St Kilda
Brisbane v Richmond +18
Hawthorn v Geelong +29

 

[sverik25]

Hobbes would you mind quoting your explanation of how your algorithm works?

 

[Ice-Wolf]

Hobbes would you mind quoting your explanation of how your algorithm works?

60% of the time it works everytime

 

[RightYouAreThen]

Hobbes would you mind quoting your explanation of how your algorithm works?

Wait - we will get a response from Bill Vlahos shortly...

 

[Hobbes]

Hobbes would you mind quoting your explanation of how your algorithm works?

There's a link to last year's thread in my fourth post above in which I describe it fairly completely.

The Reader's digest version is:

I use only the match margins as data. These are modified by: crunching big margins (halving the excess over 50), expanding small margins (adding half of the difference between the match margin for wins under 12) and modifying for home ground advantage when the teams are from different states (the total of the two teams - Vic 1, NSW 4, SA or Qld 5, WA 8.

Then I adjust every score against the current rating of the opponents (not the rating at the time of the match) and reiterate four times. This gives each team a rating, which is like an expected result against a hypothetical average team.

The predictions are just subtracting one team's rating from the other and adjusting for home ground advantage.

 

[Fadge]

Roby, is that you?

 

[Hobbes]

West Coast +11 v Sydney
North Melbourne v Western Bulldogs +29
Melbourne +22 v Fremantle
GWS +18 v Port Adelaide
Carlton +2 v Gold Coast
Adelaide +31 v Essendon
Collingwood +4 v St Kilda
Brisbane v Richmond +18
Hawthorn v Geelong +29

6/9. Gold Coast and Fremantle seem to be tough to pin down right now.

1. GWS 26.1
2. Adelaide 24.4
3. Geelong 17.0 (+2)
4. Port Adelaide 13.5
5. West Coast 12.9 (+2)
6. Bulldogs 12.1 (-3)
7. Richmond 10.8 (+1)
8. Sydney 7.5 (-2)
9. St Kilda -0.1 (+2)
10. Collingwood -0.4 (-1)
11. Gold Coast -3.2 (+2)
12. Melbourne -3.6 (-2)
13. Fremantle -8.0 (+2)
14. Essendon -9.0 (-2)
15. North Melbourne -9.9 (+1)
16. Carlton -15.2 (-2)
17. Brisbane -22.5 (+1)
18. Hawthorn -23.7 (-1)

Adelaide and GWS clearing a gap at the top, Hawthorn in freefall.

Port Adelaide +35 v Carlton
Western Bulldogs +41 v Brisbane
Gold Coast v Adelaide +18
Sydney v GWS +19
Fremantle +11 v North Melbourne
St Kilda v Geelong +17
Hawthorn v West Coast +28
Richmond +14 v Melbourne
Essendon v Collingwood +9

And, as a new feature, a predicted (probablistic) ladder.

1. Adelaide 17.3
2. GWS 16.3
3. Geelong 15.3
4. Port Adelaide 14.7
5. Richmond 14.6
6. West Coast 13.9
7. Western Bulldogs 13.8
8. Sydney 10.6
9. Gold Coast 10.40
10. St Kilda 10.36
11. Mebourne 10.1
12. Collingwood 10.0
13. Fremantle 9.3
14. Essendon 9.0
15. North Melbourne 7.5
16. Carlton 6.5
17. Brisbane 5.2
18. Hawthorn 4.2

This looks like seven settled finalists, and an almighty shitfight for eighth place, where six teams are within a match and a half of the finals.