Playoffs 2018 Week 13!

Due to popular demand (thanks Nige) and personal interest we are back at it again with the playoff predictions!

You may remember that previously the predictions were based on a series of probabilities that each individual would be another but this year we have stepped it up another notch.

In order to predict which team will beat which (and how frequently) I have calculated the weighted average (reducing value by 3% per week) score of each player and the standard deviation of that score (excluding 0 score weeks). I then use that to pick the best player - according to the model - for each position.

For example, if you have Todd Gurley and Rashaad Penny their respective averages and standard deviations (Gurley: 23.13/7.30, Penny:4.88/4.95) would be used to calculate on average how often Gurley's score would be higher than Penny's and if Gurley scores higher in 90% whereas Penny is higher in 10% of cases then Gurley would be chosen for RB slot 1 and so on.

The formula for this assessment is:

Mu being the average
Sigma being the standard deviation

          μ:=μ1−μ2          σ2:=σ21+σ22.
          1 - ( - μ / σ)

The derivation/explanation of which can be found here:

https://math.stackexchange.com/questions/40224/probability-of-a-point-taken-from-a-certain-normal-distribution-will-be-greater 

The assumptions are that the values are normally distributed and independent (since there's no way I'm deriving the covariance of players on opposing fantasy teams who share real-life teams/matches).


This was rolled up on a team basis to give you your team's average points and standard deviation. The same method used above is then used to compare teams and calculate on average how often team 1 will outperform team 2.

Here are some example charts to show how team's performances stack up against each other (blue for team 1, red for team 2):

Gridiron Gang vs Patistan

Turn Down for Watt vs IKSAN

Tom's Tyrants vs Forge Flyers

As you can see, some teams have tighter distributions representing consistency whilst others have a wider spread of results. Sometimes (as in the case of Tom) this can be a good thing for overall scoring if the average is still high. Other times it leads to a bust, see Naithan's season.

The resulting scores were then plugged into the trusty monte carlo simulator to calculate the chances of each team making the playoffs. Here are the results:

Team	Playoff Odds
The Dream Team	100.00%
Andy Dalton Red Zone	100.00%
Tom’s Tyrants	100.00%
The Tuckfards	84.92%
Burdettinators	5.24%
Obi Wan Ganobi	3.80%
Forge Flyers	2.76%
Bear Necessities	1.92%
IKSAN	1.24%
Turn Down for Watt	0.00%
Patistan	0.00%
Gridiron Gang	0.00%

As you can probably tell I should have done this earlier in the year to increase the suspense. What we see is that the 3 teams topping their division are through to the playoffs (unsurprisingly) and The Tuckfards with their "points for" and game in hand lead are big favourites to join them. But hey, there's still a 1.24% chance I'll be there with them...

So the more interesting question at this point is how will each of the teams fair according to historic performance.

These are each of the matchups for the playoffs (team 1 in blue, team 2 in red and odds edge over/under opponent from team 1's perspective):

Tom's Tyrants vs The Tuckfards (+10.58%)

The Dream Team vs ADRZ (+8.26%)

ADRZ vs The Tuckfards (+0.68%)

ADRZ vs Tom's Tyrants (-9.28%)

The Dream Team vs The Tuckfards (+10.96%)

The Dream Team vs Tom's Tyrants (-1.16%)

As you can see the really tough game as it stands will be The Dream Team vs Tom's Tyrants but that's only if they make it to the final and with only roughly a 10% edge on their opponents it's very possible that we will see The Tuckfards or ADRZ or both in the final. The predictions will also evolve over the coming weeks based on each team and player's performance. With such a tight race for the final I will probably do another roundup of predictions in week 15 before the 4 teams above play. And hey, there's a chance that Phil's team will score 20 points and all of this will fall apart!

If you have any questions or want to look into the data let me know.

If you want to see the code it's here:


https://bitbucket.org/snippets/LOC_Code/Gea48X

It did require a bit of manual assistance since they discontinued the API for the fantasy site so I just copied out the taken players and their respective fantasy teams.