Reply to thread

I'm attempting to revive the algorithm. I've made a few changes and I'll see how it pans out.

I'm not recalculating old results - this results in recursion issues and plays havoc with my spreadsheet.

Teams will have a rating which reflects an expected result against an average team on a neutral ground. Home ground advantage is assessed as two goals for interstate match-ups, nothing for intra-state. The Grand Final is considered neutral. The predicted result will be your current rating, minus the other team's current rating, and the home state bonus.

When assessing a result, I shrink results over 50 points (by half of the amount over 50) and add/subtract the home-ground bonus, and then add the opponents' rating before the game. So, a +20 win at home against an interstate team with a rating of -5 would be valued at +3. Away, it would be rated at +27.

I consider Tasmanian venues as home ground for Hawthorn/North Melbourne against interstate teams. I'll have to check if they play any Victorian teams there to see whether I need to make a decision. Other matches played outside the home team's home state will be considered neutral.

I'm currently degrading results at a regular rate, and I'm trying varying rates of degradation from 10% to 20%. I'll see which one has the best predictive rate so far this season. (So far, it only matters for roughly one match per round). In theory, old matches can continue to count toward current predictions forever. I've only entered results from the beginning of last year. With (functionally) 30 rounds in a season, including the opening round, break before the finals, and four weeks of finals, at 10% it takes 96% of its rating from the last year, and over 99% at 15% or 20%. After 13 rounds this year, 77-96% of the current ratings are taken from this year alone.

For long-term ladder predictions, I add the predicted margin to 50%, capping at 95%/5%. So, if I predict a 20 point win, I think the team will win 70% of the time. I'm applying the degradation for future tips so that projections for later in the season will be weaker - margins will shrink and predictions will tend to favour the home team a bit more.

I won't predict draws - if that's the predicted result, I'll drag it out to a 1 point tip for whichever side is rated slightly better.

Currently, with 10% degradation, the ratings are:

Collingwood 23.4

Geelong 21.4

Adelaide 20.4

Bulldogs 19.5

Brisbane 11.8

Fremantle 8.7

Gold Coast 7.9

Hawthorn 6.5

GWS 3.0

Carlton 1.1

St Kilda -4.5

Sydney -8.1

Melbourne -8.4

Essendon -9.56

Port Adelaide -9.59

North Melbourne -16.0

Richmond -22.6

West Coast -25.8

And the predicted ladder is something like this (with fractional wins)

Collingwood 17.26

Geelong 15.23

Brisbane 15.15

Adelaide 14.77

Gold Coast 14.21

Bulldogs 13.66

Fremantle 13.24

GWS 12.72

Hawthorn 12.37

Essendon 10.87

Carlton 10.39

St Kilda 9.78

Melbourne 9.43

Port Adelaide 9.32

Sydney 9.19

North Melbourne 7.48

Richmond 6.77

West Coast 5.13

If I just take the favourite for each match, we end up with:

Collingwood 19

Adelaide 17

Bulldogs 17

Geelong 17

Brisbane 15.5

GWS 14

Fremantle 14

Gold Coast 13

Hawthorn 12

Carlton 12

St Kilda 10

Port Adelaide 10

Sydney 9

Essendon 9

Melbourne 7

North Melbourne 5.5

Richmond 3