Patrician's Crystal Ball

[Patrician]

Starting a place for my own musings. For the last three seasons I have been using a stats based aproach to forecast expected points based on past performance and upcoming fixtures. It has been reasonably successful, two top 10k finishes, but I let myself down in a couple of key areas. First, I am terrible at getting a fast start, and I keep hitting the doldrums in mid-late season.

chart.jpeg

I feel the stats approach gives me a substantial advanatge in GW 3 to 10. This is the period when the stats illuminate what most people can't see - over (avoid) and under (buy) performing players. I usually Wildcard pretty early (GW3 this season) because my starting picks are often bad, and the stats tell a story after just a few matches.

Having hit ~ 3k in GW26, I was convinced that I would kick-on to a 1k finish, and it didn't happen. A slow decline started until I managed to play the chips well and over performed GW34 onwards.

Deeper analysis to follow.

Link to the GW1 team that would have won FPL in 1718, which underpins my philosophy.

http://www.fiso.co.uk/forum/viewtopic.p ... &hilit=GW1

FPL DISCIPLINES TO MASTER

There has been some emotional reaction on the main forum about the effect on FPL of the amount of statistics available via Opta in the last few years. This has resulted in a new breed of manager that makes decisions based primarily on data. Much lamenting of the good old days follows, when FPL was about football knowledge and a keen eye for watching football. This is a trend that goes way beyond FPL. There are parallels with investment banking, where in the good old days traders made millions from good stock market instincts, based on contextual understanding of the effect of a drought in one part of the world impacting the stock prices in an industry dependent on crop yields. Now, those traders are largely out of work because algorithms do the job better and don't expect bonuses. So is this the future for FPL? Can a very knowledgeable football fan beat an algorithm of the sort I advocate?
I think that question is the wrong way to think about it. Better to view FPL as a challenge with three distinct disciplines to master. You need to be good at all three.

Discipline 1: Statistics

I have put this first, because I do think it is the most important. In the end, this game is more of a math challenge than it is a football challenge. The data is available in massive depth of detail, and if we are not able to use that data to produce a decent points forecast for a player I believe we will be at a major disadvantage in FPL now and in the future. The other advantage of this discipline is the ability to assess every single player in a matter of minutes. To do the same by watching football, you would need to watch the full 90 mins of every game, every week. In fact you would need to watch every game more than once focussing on different players each time. It isn’t feasible!

Discipline 2: Contextual awareness

This is second most important, from the basics of fixture runs and injury news, to the more challenging aspects of reading the impact of personnel and tactics changes throughout the season. This discipline addresses the main weaknesses of a stats based approach which looks at averages of what has come before, and can’t predict the new points expectation when a player changes their role in the team, or the impact of a new manager, or a player comes back from injury. This is the weakest aspect of my game, and the one I will look to be better at this year.

Discipline 3: Watching football, but differently

Finally watching football. The key to this discipline is that you need to watch football with an FPL lens. As a fan I want to see players try the spectacular for the thrill of it, as an FPL manager the spectacular is just a shot with a low chance of success. We need to be able to watch a player miss chances, and not assume that means they are wasteful. We need to look for the number and quality of chances, the positions a player is getting the ball in. The eye test for FPL purposes is full of pitfalls that can lead us make bad decisions if we don’t watch the game right.

Bringing it all together

Let’s use Harry Kane as a case study. Normally a prolific striker, FPL gold in the last few seasons. Often a slow starter. Season 18/19 he gets off to his normal slow start. What has been different though is the underlying stats are also poor across the board which to the statistician means that his poor start looks like it will continue. Contextual awareness includes the obvious impact of the injury last year, world cup, coming back too soon, and a new team mate in Moura. Watching Spurs play we can observe that Moura is making a lot of runs into the box providing an additional outlet for passes from team mates and from Kane himself when he might otherwise have gone for goal himself last season.

Master all three disciplines, that’s my plan.


STARTING SQUAD PRINCIPLES UPDATED FOR 19/20

1) Assess contextual situation, e.g. VAR, Wolves in Europe, Burnley out of Europe, Chelsea in flux, ACoN
2) Assume a GW6-8 wildcard - prioritise teams with easy first 6 fixtures
3) Select two or three captainable players that rotate well for captaincy during first 6 fixtures
4) Focus on the first XI. Base price players for the bench, but ensure a playing 12th man.
5) Pick a premium priced player in each position except goalkeeper, for flexibility
5) Spread price points for flexibility
7) Identify and cover the key price points (e.g. 2+ good options), and avoid price points with just one good option (e.g. maintain 1 move transfers)



SEASON PRINCIPLES UPDATED FOR 19/20

Primary Goal: 2,700 points, which means quickly establishing a team which will on average score ~70 points per week

1) Relentlessly seek and drive towards the highest average xPts combination of players
2) Value for money without the armband comes at 6M and below, eleven of your squad should cost 6M or less (can break this rule for certain premium defenders like TAA and Robertson). If they cost more than 6M and you never plan to captain them, they have no place in your team.
3) Identify the core VfM players as soon as possible in the season (more on this later), get them in and stick with them.
4) Ignore the crowd. Never chase points unless they are supported by underlying stats
5) Don't hesitate to make the obvious plays (e.g. Aguero transfer in 17/18)
6) If the stats are supportive, be patient (e.g. Kane 17/18, Wilson 18/19)
7) Wildcard early (but try to have > four weeks of data) to take advantage of the models strength
8) Don't let the chips drive strategy, and don't be adverse to playing them outside of a DGW
9) Avoid hits unless there is a clear expectation of neutral xPts or better
10) Avoid early GW transfers, unless price changes really force the issue - wait until Fri/Sat
11) Maintain price point flexibility, a decent enough bench, and a strong 12th man. Injuries and rotation are inevitable
12) Monitor form closely (last 4 gameweeks), to spot emerging players and changes in role/tactics
13) Fixtures matter, especially for the premium players, so rotate for good fixture runs
14) Be flexible, rules are there to be broken, there is no such thing as a fundamental truth in FPL

CAPTAINCY PRINCIPLES FOR 19/20

1) Perhaps unsuprisingly, generally captain the best premium playing at home against the worst defence...
2) ...unless it is Sterling, who seems to do better against more average defences (know the player)
3) Generally favour midfielders. Big hauls come most frequently from the top midfielders
4) Don't forget TAA - he got 9 big hauls, 6 of which came at home for an average of 7.3

Finally: Don't get too excited by a good game week, or too depressed about a bad one. Complacency and rage transfers are the enemy

NEW LEARNINGS FROM 19/20

1) The principle about dead zone players needs changing, and was really an observation about the META of season 18/19
2) Need to add an early season objective about identifying the "season META" which is anchored in the price points of the under valued players that emerge. Don't use the first wild-card until confident this has been identified.

KEY RESOURCES

https://understat.com/ for quick player scouting
https://www.fantasyfootballfix.com/ spying on rivals, price changes, pts fcast sense check, heatmaps, first XI predictions
https://www.fantasyfootballscout.co.uk/ for good articles and primary source of stats for my model (members area), heatmaps
http://www.fplstatistics.co.uk/ for price change sense check
https://fantasyfootballhub.co.uk/the-de ... nt-theory/ Analysis of defender scoring trends
5 seasons of analysis Awesone stats reference last 5 seasons

[Patrician]

SEASON 17/18 REVIEW

A bad start

Lets look at the start, GW1 41 pts, OR 3.3M. Not what dreams are made of. Was it unlucky or was it a bad selection? It can be hard to look at a woeful GW 1 score. It is important to view it in context. Overall there were 7 gameweeks (18%) where my rank was > 3M, and 22 gameweeks where my rank was >1M (58%). How was the selection though?

GW1.jpeg

1) Every player apart from Quarner started - very good
2) 10 played the full 90 - not bad
3) Premium selections all did well over the season (Alonso, Kane, De Brunye)
4) Nailed the early 4.0 GK enabler

With the benefit of hindsight, here are the flaws

1) 2 x Spurs, playing their home matches at a new stadium, perma-captan Kane for the first few weeks. It was predicatble they may start slow
2) 1 x Man U who had an easy start, I knew it, and Lukaku goes off to a flyer
3) 2 x plus 10M players. I think I am coming around to the idea that only one +10M player. Yes De Brunye did well, but he was never captain material and the sacrifies elsewhere were too much
4) Not enough money in defence (lightbulb from the *Optimal Squad" thread)

Key Lessons Learned

1) 1 x premium player above 10M in starting squad
2) 2 or 3 three premium defenders
3) Situational awareness matters, e.g. early fixtures, world cup impact for next season
4) Don't panic, bad GWs happen even with good selections
5) GW1 Bench Boost would have been 9 pts vs 20 pts in GW34

[Patrician]

The GW3 Wildcard

A few players shone out in the first few weeks, notably Lukaku, Davies, Richarlison, Eriksen, and also Salah. I activated the wilcard to grab all of them (except Salah, was still suspicious of the Carlos Kickaball effect, and I read the "wasteful"comments here - he came in in GW10, his stats were off the chart, and this was early enough). Crucially I kept Kane despite his poor start, his stats were good. He blanked again in GW3 (Lukaku also missed a penno) but went on a huge run in the following weeks, while his ownership had dropped a lot.

GW4, first big score, 79, Gameweek Rank 17,040

GW4.jpeg

I then steadily climbed the rankings until GW 26 with a best OR of 3,597

Key Lesson

1) The stats do the job early in the season. Having an edge where you can spot who is scoring more or fewer points than they should be is a significant advantage before things stablise later in the season.
2) If anything trust the stats more, I delayed the Salah transfer too long due to reading comments here

[Patrician]

The one that got away

Documenting this howler for prosperity. It was the OBVIOUS move. I knew it, I hovered over the button, I KNEW IT. Agueros run GW24-27 was such an obvious play. I stuck with Kane for a run which included MUN, LIV, ARS.

Aguero Hauls.jpeg

This starts a steady decline which I need to look at closer. My gut feel is that I turned my attention to the DGW BGW DGW, and stopped playing my normal game. Maybe I just didn't get the run of the green. Maybe I tilted from missing Aguero. Also worth mentioning that I have a substantial bet on with a friend and we are neck and neck - he did Aguero(c).

Key Lesson

1) Don't hesitate to make the obvious play
2) Don't let the DGWs distract from solid day to day play

[Patrician]

DGW SGW DGW Chip Play

Overall I think this was a success. GW34 97pts GWR 129,667 with 20 pts BB. GW 35 70pts GWR 34,184 FH. GW37 79pts GWR 1.3M no chip.

GW34

GW34.jpeg

This was the boomerang strategy. Heavy on players with two DGW (LEI, TOT, CHE, MUN), with a FH for GW 35. The Leicester players paid off big time in the end, as they also went crazy in GW38.

GW 35

GW35 FH.jpeg

Kudos to FISO for the Lacazette tip

Lessons

1) Chips do well during this period...
2) ....but the teams that tend to have DGW are also the big squads with other commitments and will rotate
3) Teams like Leicester with small squads, and clubs fighting for survival are golden
4) On balance, despite getting a good result, GW 33 (rank 4.1M) and GW36 (rank 3.9M) probably negated the good chip play, not to mention the distraction in the build up.

[Patrician]

GW38 Mayhem

This was an epic week. I started 12 points behind my rival with the big bet. At one point I was about 20pts ahead, then Fabianski penno save, Burnley concede, Zaha scores, Jesus scores. I lose the bet by 6 points. 97pts, GW rank 39,936, OR 5,749.

GW38.jpeg

Lessons

1) Final GW is mayhem
2) Try and spot a fixture with two competent attacking teams who have nothing to play for (LEI v TOT) - it could go off
3) Back a teaam still fighting for champs league (LIV)
4) Listen to RuthNZ when he highlights players for the run in (Zaha)

[Aldershot Rejects]

1) Chips do well during this period...
2) ....but the teams that tend to have DGW are also the big squads with other commitments and will rotate
3) Teams like Leicester with small squads, and clubs fighting for survival are golden
4) On balance, despite getting a good result, GW 33 (rank 4.1M) and GW36 (rank 3.9M) probably negated the good chip play, not to mention the distraction in the build up.

Thanks for the write-up which I have enjoyed reading.

Observation (4) is interesting. For me the deciding factor in whether or not I played the chips well was not just the result in the week I played the chips but in the surrounding weeks as well. So GW33, 36 & 38 are pretty much as important as the other weeks. So setting up a team which will perform across the final few GWs of the season is the important thing, not just achieving good results in one or two weeks. You mention Zaha and he was a player without a double but with a very good run of fixtures. So he is far more attractive if you take the whole run of fixtures at the end of the season rather than seeing things through the lens of the chips and the dgws.

As it happens, having slipped considerably in the rankings, you made an excellent recovery after playing your WC in GW32 which suggests that you played that entire run of games pretty well. As you hint, it was that run between GW27 and your WC that really scuppered your hopes of a 1k finish.

[MoSe]

chart.jpeg
download/file.php?id=27448&mode=view

I see you took the graph from FFFix

I'm sorry, but being one who produces his own, I always foud their ones sloppy
for a series of reasons I won't detail unless someone is really interested

FPL 1718 Patrician.png

please, if you find this doesn't contribute to your blog and uselessly clutters it, just tell me and I'll remove it

[Spinynorman]

Interesting stuff and quite enlightening. Cheers.

[Ardrageen]

Nice blog, some interesting thoughts.

One small thing, I would probably not blame FISO for your decisions that didn't work out but each to their own.

I delayed the Salah transfer too long due to reading comments here

[Patrician]

Nice blog, some interesting thoughts.

One small thing, I would probably not blame FISO for your decisions that didn't work out but each to their own.

I didn’t mean it that way, so sorry if it read that way. What I meant is that I made a bad decision by breaking my own rule - don’t trust the eyes as much as the stats.

I also mentioned at least two occasions where FISO info was bang on (Lacazette and Zaha). One I capitalised on, the other I didn’t.



Sent from my iPhone using Tapatalk

[Ardrageen]

That's fair enough.

[Patrician]

I like your chart MoSe, what do the R^2 numbers mean?

[Patrician]

Initial Selection Deep Dive

This is very interesting, especially the comparison of my starting selection - quite a few high cost players and a lot of cheapies, vs the average cost of my team over the seaosn which is a much more balanced structure. So I think this is a key learning to pick a better balanced initial squad more in line with my season average structure.

Team Structure.jpeg

[MoSe]

I like your chart MoSe, what do the R^2 numbers mean?

OH, sorry, forgot them there

In a regression, R^2 is in short the linear correlation factor (this is not the scientifically exact definition, but one that a not expert statistician like me could easier understand and remember)
In practical terms, one of the regression says, what if Patrician ranking advancement was a straight line over the season? That is, if his improvement was constant each gw? The one in the graph is the closest straight line describing your season.
Then you might ask, how close is it to my actual ranking gw by gw? R^2 is that measure.
You could read it, as "that straight line explains 73% of Patrician ranking gw by gw evolution", "Patrician season is explained for 73% by a factor of constant improvement"; the rest, the ups and downs away from a straight line, has to be explained by something else, by factors with different relations.

That is not good or bad per se, it just describes how "straight" your season climb was. Mind, not necessarily how "steep", just how straight.
You could have a season going from 4M ing GW1 to top10 in the end, the line would be steep, but if you got there thru high peaks and deep troughs a straight line would be a poor description of it, and the R^2 would be low, despite the line slope is steep.
Or, you could just go up from 1M to 500k, but do it bit by bit every gw a small green arrow up the standings. That line slope would be flatter, but your actual season would be very close to it and you R^2 very high.
73% means pretty straight, but you can see both higher an lower. In 16/17 I went up rather straight and my R^2 was in the 85-90% IIRC. This one was a nightmare instead, shot up and slipped back more than once, then stalled with a final spurt, and it was less than 50% straight.

Then, you could check with formulas different than a straight line (y=ax+b), if they better approximate and describe your actual rankings.
There are several kinds, the "wavy" line in the graph is a polynomial of 6th order (with factors up to x^6 by gw). If we use such formula, the approximating wave we can find describes your season by 95%.

I wouldn't put much on it tho.
These are just standard default tools excel offers as optional features when you have a line graph, and it's just a curiosity to check how straight or how wavy your season was, easily adding with a few clicks the regression line(s) from the 4 or 5 kinds available
_______

Anyway.
The chart is mainly the same as the FFFix season one you had posted.
Only, better rankings go UP, not down and their log scale is more fine-tuned.
And bars show your gw rankings as well, in the same scale, rather that Fix one's points (which have to be compared with average/top10k points to be meaningful)

[MoSe]

So I think this is a key learning to pick a better balanced initial squad more in line with my season average structure.

This should be sensible.
But as usual, I like to also consider some off-the-box alternaive explanation (I then might not be able to draw the correct conclusions, that's a different matter )

Maybe an "unbalanced" initial selection, with more cheapies and pricier premiums, offers more flexibility, considering that some changes will surely have to be done after the first gws.
If you have an 11m and a 4m player, it's easier to change the 11m to an emerging 10 or 9m if needed, with a single FT, then you'll reallocate the spared budget next gw.
If you have a 10m+5m, or a 9m & two 4.5m, you won't have margin if you realise later that you needed that 11m player after all, and you'd need a 2nd transfer to fund it, probably costing a hit.

What I'm saying, you could consider an unbalanced gw1 selection as a specific tactic knowing you'll likely change your team anyway following what emerges after gw2 or gw3.

Of course it will all have to be weighed and assessed with pros and cons

[Patrician]

It is a good point which I think I normally cover by starting with the most expensive player in each position. The problem with my starting squad seems to be the expensive 2nd forward and expensive 2nd and 3rd midfielder. Just too top heavy when I can only captain one player.


Sent from my iPhone using Tapatalk

[Billy Bongo]

Excellent blog interesting read.

One word of caution, placed like this are great, but have their down sides. Many of the noisiest members can take over and they aren't always right, I'm referring to the label 'Carlos kickaball'

It's only a term I've only heard of in heard of in here. Some in here because they didn't know of Salah just labelled him. Arrogant in the extreme. He had a superb season at Roma and many of us started the season with him.

It's the unusual thing about this place, if people don't know they just guess, and use labels like Carlos kickball to try and prove themselves right.

Salah was never that, and that's my next point, stats are great, but so is actually watching football and following it , not just on paper. Salah is the perfect example

Sent from my SM-G965F using Tapatalk

[Patrician]

The Carlos Kickaball commandment emerged in the early days of FISO during a period when quite a lot of big name players came to the EPL with big reputations and went on to be very underwhelming. In those days I played TFFO, and have been burned many times myself picking that type of player. TFFO was a game with quite limited transfers, so it was a lot more important to not waste them.

I think it is true, maybe more so back then, that the EPL has high of level of intensity, more competitive, while being less technical than the other big European leagues, which means a player that shines elsewhere may struggle here. Vice versa too, a player that shines in EPL might struggle in Europe.

So I normally do follow the commandment, but I fully agree that it is lazy to just slap a label on a player, and assume that as truth. Kudos to those that picked Salah from the off. From my stats perspective it was obvious by GW3 that he was potentially a great pick.

My word of caution on watching football is that looks can deceive. For example, consider two players, and assume neither has EPL history.

Player 1: scores a hat trick from 3 shots
Players 2: scores 1 goal from 10 shots

Watching those two players, you could conclude that player 1 is lethal, and player 2 is wasteful, especially if some of the misses were easy chances. The facts are that forwards convert ~ 15% of shots on average. Player 1 will likely be a 1 goal every 3 games player. Player 2 will likely be a goal every game player. This situation happens quite a lot in the early part of the season.


Sent from my iPad using Tapatalk

[Billy Bongo]

Yes a balance of both is wise, fantasy football requires data wherever you get it. But there are many opinions on social media, you have to work hard to find just enough glitter amongst the chicken feed, so say George Smiley. You have to be as cunning as him to weed it out

Sent from my SM-G965F using Tapatalk

[Patrician]

Transfer Discipline

This graph is a bit embarassing, as I am sure that the best time to make transfers is aas late as possible, but I get twitchy fingers every Saturday. I guess it has to be worth a good 20 points over the course of a season, so this is one weakness I will try and iron out next time.

Transfers.jpeg

[Patrician]

Captain Picks

This is an area where I would expect my algorithm to perform well, and it seems to be born out by the results. It wasn't a particularly challenging season to get the captain right, as a lot of the time the right choice was Salah. I did a little worse overall compared with the top 50, but had I got the Aguero hauls which I really should have done, it is not hard to see how I could have over-performed substantially.

Captain Stats.jpeg

Captain Picks.jpeg

[Patrician]

Casting my mind to the new season

Starting squad principles.

1) Don't select any players who have gone deep in World Cup. Avoid Spurs entirely
2) Assume a GW6-8 wildcard - prioritise teams with easy first 6 fixtures (M City, Everton, Chelsea, C Palace, Bournmouth, Burnley, Liverpool, Southampton)
3) Select just one 10m + player (Candidates: Salah 13M, Aguero 11M, Aubameyang 11M, Sanchez 10.5M)
4) More emphasis on a premium defence than I normally go for
5) Spread price points for flexibility

[Patrician]

Who to pick as my one Super Premium?

First a quick explanation of how I assess players. This table show the conversion stats by position for shots in and outside the box, and chances created to assists across all players last season.

Conversion.JPG

The simpifying assumption I use is that I calulate expected points per game based on how many shots/chances created using these average conversion rates, and base my decisions on that. To bring it to life, lets have a look at my Super Premium candidates.

Super Premium Comparison.JPG

Immediately I can dismiss Aubameyang, and relegate him to the watch list. Yes, he was scoring at at a very impressive rate last season (and so far in pre-season), but look at the underlying stats - his scoring was based on an inhumanly high conversion rate for both shots and assists. With a more normal conversion rate I expect him to be more like Arnie, who I include for comparison.. I wouldn't pay 11M for Arnie, so I am not paying 11M for Auba.

Kane is just there for comparison, as a well understood player. Note that he has more "normal" conversion rates, and his actual pts per 90 is quite close to my forecast pts per game.

So it comes down to Salah vs Aguero. My calulations assess both as similar potential, at about 240 points per season if they play every game. They have very similar underlying stats. Therefore, it is hard to jusitfy an extra 2M for Salah over Aguero, if I am confident that Aguero starts the season. Salah is more assured of his spot though and it is easier to sell him than buy him, so is it worth the premium? Auba is an easy swap for Aguero if it turns out I wrong about him, and Jesus comes easily into consideration if Aguero is injured. There is also a relatively easy two transfer move to get Salah (Aguero to 7M striker, 9M midfielder to Salah). There isn't much in the fixtures for the first 4 games, after which I expect to wild card.

I don't think there is much in it, so I will bank the 2M and opt for Aguero (gulp!).

Finally why just one Super Premium? Look at the double pts/£M line. A super premium captained is fantastic value for money. Look at the potential pts/M line. Withought captaincy I would rather have two Arnies plus 1M than Aguero plus a 4.0.

[FranckKessie]

The positional advantage of Salah somehow isnt translated into points?

[Patrician]

I could add a few expected points to Salah by assuming the conversation rates of a forward, but I think my conclusion would stand: he is a 250ish player, 300 unlikely to strike twice.

Bit scared not to pick him though!


Sent from my iPhone using Tapatalk

[LamebrainEddy]

Can't the Aubamayang conversion be explained by the hypothesis that he had a higher proportion of quality chances?

He overachieved his xG by only 0.83, which is why I make the point

[Patrician]

You could be right. I use my own algorithms, and when I see extreme conversation rates, I assume it is luck until more data proves otherwise.

I just don’t believe that Auba is twice as lethal as Kane for whom I have more data.


Sent from my iPhone using Tapatalk

[Patrician]

When you add the fact he has two tough games to start, it is an easy decision.


Sent from my iPhone using Tapatalk

[Patrician]

Funds allocation

GK 8.5M – I think the 4.5 plus 4.0 second choice keeper is a perfectly good tactic
DEF 28M – three premium, one mid-price, one cheap
MID 38.5M – two premium, two mid-price, one cheap
FWD 25M - 11.0, two mid-price or 11.0, one premium, one bench fodder

Formation is 4-4-2, or 4-3-3

So this is where I get to on funds allocation once I apply the principle of a "more premium" defence than I usually go for, together with a spread of price brackets for flexibility. I feel this structure will give me a much better balanced squad than I normally achieve (last year I started with three 10M+ players I think it was).