Analysis 2024 AFL SC: Rolling Perfect Team [Post #242 onwards]
- Thread starter Beg2Differ
- Start date
- Joined
- 22 Oct 2014
- Messages
- 7,882
- Likes
- 41,731
- AFL Club
- North Melb.
Not surprisingly there is a new leader this week (so no head?) and it happens to be somebody who had Ben Mc as captain.
Interesting points
Interesting points
- Didn't captain Maxy in round 3
- Had a doughnut in round 5
- Was ranked 1557 after round 3
- Former leader (12th man) dropped to 29th and had the V on McDermott and also had Sam Elliott in the team who they could have used to loop; possibly because they also had Kurtis Patterson in the team and may have been counting on him to loop. Oops.
Last edited:
- Joined
- 22 Oct 2014
- Messages
- 7,882
- Likes
- 41,731
- AFL Club
- North Melb.
Given you have taken such an interest @warewolves I have run the following team for 2021 which enforces the guns and rookies approach for the starting team. This uses the 5% tolerance model which takes about 6 hours to run. I can run this overnight and get a result by the morning. This means it won't find the true maximum but should be good enough for this purpose. I added some constraints that meant the starting team must consist of at least 13 players prices $450k and above and at least 15 players that are priced at $200k and below. Note that I used $450k rather than $500k because I thought $500k was too restrictive on the forward and defender premiums.
I think this illustrates well the point I was making earlier in that simply enforcing a guns and rookies starting team still doesn't give you a realistic scenario to assess the G&R approach. There are any number of trade examples in this team that are unrealistic and impossible to predict. Effectively this is a GnR starting approach which then switches to value based trading approach which is what the model will automatically find unless you specifically prevent it. I haven't done this and not sure how you could without forcing the result you expect rather than letting the model find an objective solution. Too me it looks like it is funding its upgrades by trading out injured or high priced premiums and trading in the value. Note there are still no downgrades to rookies.
I think this illustrates well the point I was making earlier in that simply enforcing a guns and rookies starting team still doesn't give you a realistic scenario to assess the G&R approach. There are any number of trade examples in this team that are unrealistic and impossible to predict. Effectively this is a GnR starting approach which then switches to value based trading approach which is what the model will automatically find unless you specifically prevent it. I haven't done this and not sure how you could without forcing the result you expect rather than letting the model find an objective solution. Too me it looks like it is funding its upgrades by trading out injured or high priced premiums and trading in the value. Note there are still no downgrades to rookies.
I may need to think further about this, but my immediate response when I read your last couple of paragraphs was “That sounds a lot like BBL”.
To me the key differentiating trait of BBL SC vs AFL (aside from the byes and double game rounds) is the high number of trades - 3 per round, or 19% of our list, every round!
Perhaps a reason why this value based approach (1) works in BBL, (2) works in the perfect team AFL scenario, but (3) is uncommon and perhaps does not consistently work in AFL, is that it relies on either having a large number of trades (in BBL), or in nailing too high a proportion of them (as the perfect team will naturally do)?
Whenever I’ve experimented with a really midpriced team, I’ve been warned off by the high trade intensity of the strategy, and I’ve found this to be a valid warning.
Perhaps the above provides additional evidence supporting that?
To me the key differentiating trait of BBL SC vs AFL (aside from the byes and double game rounds) is the high number of trades - 3 per round, or 19% of our list, every round!
Perhaps a reason why this value based approach (1) works in BBL, (2) works in the perfect team AFL scenario, but (3) is uncommon and perhaps does not consistently work in AFL, is that it relies on either having a large number of trades (in BBL), or in nailing too high a proportion of them (as the perfect team will naturally do)?
Whenever I’ve experimented with a really midpriced team, I’ve been warned off by the high trade intensity of the strategy, and I’ve found this to be a valid warning.
Perhaps the above provides additional evidence supporting that?
- Joined
- 8 Aug 2012
- Messages
- 418
- Likes
- 2,699
- AFL Club
- Bulldogs
Many thanks again @Beg2Differ. The new run-thru looking for an optimum G&R starting line-up is very interesting and like @Darkie, I will be thinking about it for a while.
So now we can see that the best G&R line-up would yield a season-score in the region of 62,000 as compared with the optimum of 62,839. This is definitely a much smaller gap than I was expecting, so my championing of a value/mid-pricer strategy is diminished – if not entirely extinguished. My expectation (and hope) was a difference of 3,000 - 4,000.
I was convinced that the “value” starting picks in optimal team were the key to generating the cash that enabled the swag of big guns that the model subsequently traded in across the 2021 season. It now clear that most of this was equally achievable with a G&R structure – so it may be that the optimiser cannot really be used to argue my case.
The new G&R run suggests that good trading is more decisive than a good starting line-up. For me, it’s back to being a more open question – whether G&R is the better way to go – or not. Things to mull over a bit further..
So now we can see that the best G&R line-up would yield a season-score in the region of 62,000 as compared with the optimum of 62,839. This is definitely a much smaller gap than I was expecting, so my championing of a value/mid-pricer strategy is diminished – if not entirely extinguished. My expectation (and hope) was a difference of 3,000 - 4,000.
I was convinced that the “value” starting picks in optimal team were the key to generating the cash that enabled the swag of big guns that the model subsequently traded in across the 2021 season. It now clear that most of this was equally achievable with a G&R structure – so it may be that the optimiser cannot really be used to argue my case.
The new G&R run suggests that good trading is more decisive than a good starting line-up. For me, it’s back to being a more open question – whether G&R is the better way to go – or not. Things to mull over a bit further..
- The way I see it is that a good value line-up is “buying” more points with my $10M – this can’t be a bad thing – but may be hard to achieve – is it worth persisting?
- The flow of rookies thru the season always seems to be important – but doesn’t feature in either optimising solution.. interesting..
- The optimal value solution “beats” the optimal G&R solution by about 800 pts – at first it doesn’t seem like much but perhaps it is worth holding on to.
Your first bullet point is something I used to grapple with a lot - if I can get a mid priced player at a discount, and get more points out of my $10m, surely that puts me at an advantage … so why wouldn’t I do that?
The thing that I was probably missing is that rookies are typically the best value players in the competition, assuming they are playing. They’re typically priced at 23-38, which is a very low bar.
The other things with a true GnR structure are:
(1) It forces you to pick a larger number of rookies, perhaps a much larger number - and if they’re the best value players, you’re putting a lot of value into your side, and giving yourself the maximum opportunity to get the best cash generators.
(2) You start with a side that’s much closer to a full premium side (say 12 top premiums, rather than 10 decent ones and some dodgy ones). This means you need fewer trades to get it properly upgraded, and your finished product is much better than some of the compromised sides that we’ve all ended up with at times!
If you view a GnR structure as providing you with maximum opportunity to get the top rookies, which are the best value and best cash generating players in the comp, I think it makes it more conceptually appealing.
I usually allow myself c. 2 midpriced exceptions, for those I am most confident on/excited about, so I don’t run a strict GnR structure - but I’ve had a good experience with it, and I’m much more of an advocate of it than I used to be.
The thing that I was probably missing is that rookies are typically the best value players in the competition, assuming they are playing. They’re typically priced at 23-38, which is a very low bar.
The other things with a true GnR structure are:
(1) It forces you to pick a larger number of rookies, perhaps a much larger number - and if they’re the best value players, you’re putting a lot of value into your side, and giving yourself the maximum opportunity to get the best cash generators.
(2) You start with a side that’s much closer to a full premium side (say 12 top premiums, rather than 10 decent ones and some dodgy ones). This means you need fewer trades to get it properly upgraded, and your finished product is much better than some of the compromised sides that we’ve all ended up with at times!
If you view a GnR structure as providing you with maximum opportunity to get the top rookies, which are the best value and best cash generating players in the comp, I think it makes it more conceptually appealing.
I usually allow myself c. 2 midpriced exceptions, for those I am most confident on/excited about, so I don’t run a strict GnR structure - but I’ve had a good experience with it, and I’m much more of an advocate of it than I used to be.
- Joined
- 22 Oct 2014
- Messages
- 7,882
- Likes
- 41,731
- AFL Club
- North Melb.
- Joined
- 8 Aug 2012
- Messages
- 418
- Likes
- 2,699
- AFL Club
- Bulldogs
Following posts from @Darkie & @Beg2Differ I have attempted to pull out some stats to help our analysis of G&R vs Value/Mid-price starting structures.
The two groups are split by their 2021 starting salaries – Group 1 are rookies (up to 200K) and Group 2 are cheap mids (201K-375K). I have chosen to chart the outcomes based on the prices after Round 8 has completed (so prices going into R9) and selected players who gained at least $80K over that period. This is an arbitary point but I reckon by then you would be needing to cash out - can change this if necessary.
The blue line is the demarcation between the 2 groups. The 3rd column shows the first round that the player was selected – so those after R1 are not strictly speaking in our starting squads. The final column shows the average score for the players over the first 8 rounds of the respective seasons.
The two groups are split by their 2021 starting salaries – Group 1 are rookies (up to 200K) and Group 2 are cheap mids (201K-375K). I have chosen to chart the outcomes based on the prices after Round 8 has completed (so prices going into R9) and selected players who gained at least $80K over that period. This is an arbitary point but I reckon by then you would be needing to cash out - can change this if necessary.
The blue line is the demarcation between the 2 groups. The 3rd column shows the first round that the player was selected – so those after R1 are not strictly speaking in our starting squads. The final column shows the average score for the players over the first 8 rounds of the respective seasons.
