Fantasy Watch: Fixture Difficulty Rating

SUM of Difficulty	Game Week
Team	GW34	GW34+	GW35	GW36	GW36+	GW37	GW37+	GW38
Manchester City	78%	78%	86%	78%		74%	57%	70%
Liverpool	62%	75%	67%	62%		62%		70%
Manchester Utd	53%	31%	42%	65%		56%	50%	67%
Brighton	63%	41%	67%	22%	28%	73%	19%	38%
Newcastle	74%		35%	57%	45%	67%		37%
Arsenal	59%		37%	53%		67%		72%
West Ham	41%	8%	30%	32%		56%		40%
Chelsea	18%		48%	64%		9%	24%	34%
Tottenham	16%		57%	30%		48%		45%
Crystal Palace	30%		19%	43%		30%		47%
Aston Villa	21%		41%	42%		16%		33%
Fulham	8%	9%	41%	37%		42%		13%
Bournemouth	43%		25%	29%		20%		30%
Brentford	56%		13%	39%		25%		11%
Leicester	45%		31%	16%		13%		32%
Nottingham	19%		46%	14%		13%		26%
Everton	28%		13%	8%		26%		42%
Wolves	15%		31%	14%		46%		10%
Leeds	29%		4%	19%		19%		27%
Southampton	9%		27%	35%		10%		11%

elo generated probability based FDR - updated 2023-04-29 - (this will be the last update - the projections wont change a bunch over the next game weeks, one aspect to look out for is teams that are playing "dead rubbers".  For example my ratings are projecting West Ham in GW34 win odds at 2.42, however the market has drifted to 3.0.  West Ham would seem extremely unlikely to change their fate in the run in, safe from relegation and miles away from European places.  They could be playing vs extremely motivated opponents.)

The Problem (Part I)

There is a problem with the FDR's on many Fantasy content sites - even the official site.  Despite claims to the contrary its alot of words and complex sounding algorithms to arrive at the wrong ratings.

This is from the site:

FDR Explained

The FDR is based on a complex algorithm developed by FPL experts.

A set of formulas process key Opta data variables, along with each team's home and away form for the past six matches, to generate a rank for the perceived difficulty of each Gameweek opponent.

The FDR is designed to help FPL managers plan their transfer and team selection strategy. It is reviewed on a weekly basis and updated as the season progresses.

However it doesn't take long to pick it apart.

It looks like Wolves have an easy fixture against Everton away in GW17. Also Newcastle and Tottenham have trickier fixtures versus Leicester and Brentford.  This is not right.  Newcastle and Spurs have a higher win expectancy than Wolves.  With that will probably come a higher goals expectancy and probably a higher chance for a clean sheet.  In general there will be better scoring opportunities for the Newcastle and Spurs players in comparison to Wolves players, even though Wolves have the "easy" fixture.  The win odds at time of writing are 2.0 for Spurs, 2.3 for Newcastle and 3.3 for Wolves.  Implying Spurs have a 50% chance to win and Wolves only 30%.  Its quite a significant advantage for Spurs and the FDR would imply the advantage was the other way around.

We can easily pick out more discrepancies.  For weeks further out as well, although there are no odds available to quickly prove it we can still see contradictions.

Below we see Man City will have "hard" matches against Chelsea (A), Man Utd (A), Spurs (A) and Arsenal (A) in the next 7 game weeks. Meanwhile Bournemouth's matches against Palace and Forest are "easy".

We already saw Man City go off odds on favourites against Liverpool (A) earlier in the season.  It is reasonable to expect that they will be equal or in most cases a stronger favourite against these other teams.  What about Bournemouth's easy matches, well we believe that they will be 3.6 vs Palace and 2.5 vs Forest. The only match this season that Bournemouth started favourites was against Southampton (H) - 2.71 to win, 3.3 to draw and 2.86 to lose, so just about favourtie. You can expect them to be underdogs vs Palace and perhaps similar price to that Southampton game versus Forest.  

If you plan to bench/sell your Man City assets in this spell of games it might be a mistake.  They may be some of Man City's hardest matches this season but we still expect City to score more and concede less than most other teams, even those teams on an "easy" week.

So much for the experts.  So much for planning transfers and team selection.

The Problem (Part II)

So why not just base the FDR on the match odds as we've used in the comparisons above.  Well that is the idea, however the books and the exchange generally will not publish odds more than a week in advance.  Sometimes two weeks in advance although these will have low confidence, low limits and low liquidity.

We will have to generate weeks ahead by ourselves.

The Ratings

So we develop ratings that match the odds.  Everything gets baked into the match odds - form, ability, strengths, weaknesses, recent results, xG, stats, advanced metrics etc.  We don't have to go back to collect data, or analyse which data gives the strongest signal, or even build a model based on this.  We can just listen to what the odds tell us.  Are the teams equal, is one team better than the other, a lot better, a little better.

I am reading an elo rating for each team off the odds.  Usually elo rating are generated from match results and used to calculate future match probabilities, so this is just the reverse, the market tells us the probabilities and we infer the team ratings from that.

The odds will only tell us our ratings at kick off.  Information/performance after this is not captured in the ratings.  The standard elo method will normally update team ratings after each result.  We have to anticipate that the markets will too.  So we still use an elo ratings update based on the result.  I am using k = about 5 as it seems to agree closest to how the market updates.  Many elo systems will update by a factor of 20 or 40, however it seems we are reading a more accurate rating at kick off for the odds that we shouldn't and don't need to adjust quite as much on the result.

The Result

So once we have the ratings its just a case of plugging them into an elo calculation (the right way around this time).  This will give our win probabilities for all matches in the schedule.  This is used then for the FDR.  Is it accurate?  Probably not, but it comes close enough.  Below is a quick comparison between the elo generated odds vs the betfair odds at time of writing.  The biggest discrepancy probably is highlighted, but in general I think its more accurate than the bog standard FDR's currently available (except drafthound seem to have a decent one, although behind a paywall so I cannot see their ratings for weeks in advance, just the current).

The Caveats

So as you can see its never 100% accurate and sometimes there are significant enough differences.  These will be exagerated when there is injury news, or a player returning from injury, perhaps new signings coming in at the windows, maybe players rested at busy periods.  The elo generated odds are based only on the most recent closing odds, the most recent match result and the elo ratings of each team before that match.  Any contributing factors that differs in the coming weeks or further into the future weeks will not be accounted for.

The full FDR is available at the top of the post, which I may or may not update as the season goes on.