What follows is a partial extract of a conversation from our community where we go further under the hood as to how it works. TakingValue is a FI Trader/ Poker player and DCA is the developer here at IndexGain. It’s not necessary to have a full understanding of the below to understand the value of the Implied Dividends as a statistic, but for those interested and/ or for those that have a love of Maths or Poker, then read on……
Is there any info on exactly how the Implied Divs are calculated, I’m trying to understand how Akanji comes out at 0.33p when Carvajal has a higher PB Avg and Higher PB StDev but is only 0.23p. I understand Opta style data is being used for the calculation but I would have thought PbAvg and StDev correlate closely with the Opta scores as they are based on them.
My understanding of implied PB would be to look at the Opta data for a player over say 100 games and figure out what the players probability of their PB being a value of X is on any given day. Then the probability of any given match day being a single, double or treble should be calculated. Then the probability of X having the best PB score on that day. Hell of a lot of probabilities to compile for each player but I have always assumed with Opta data it is possible. But I would also have thought that PB and PB Stdv would correlate strongly.
it’s not using the Opta data to calculate a score / implied score – FI do that already. It’s taking the actual PB scores and rather than using the (all or nothiing) actual div received for that score (ie, 18p for the top player on a triple day, 12p for the 2 other position divs, and 0p for everybody else), it’s calculating an more representative div for that score that factors out the good/bad fortune of other results on that day. So for example, a 250pt score might be worth 18p on one day, 12p on another, and 0p on another depending on what other scores occurred that day. By doing a statistical analysis of that scores value based on all the historical scores, for that position type, and for that day type, we can calculate that a FWD scoring 250pts on a triple day has a 82% chance of winning the Pos div, and a 59% chance of winning the star div, and that the long run expected value of those 250pts for that position and that day type is 13.38p. Likewise, a score of 300pts by a FWD on a triple match day has an implied value of 17.23p.
Regarding the averages / Akanji vs Carvajal question, you can see the reason on the BuzzPro player dashboards….
Akanji Implied Divs
Carvajal Implied Divs
Akanji got more implied divs on his top score than Carvajal as he hit a much bigger score and therefore had bigger chance of winning the 18p. He also has two solid >200 double day scores to Carvajel’s one. They are both picking up similar on the < 200 scores. Like @indexkr mentioned, the day types matter, and Akanji has posted more / better scores where it counts. The way to think about this stat is as a better version of div’s received.
It removes the luck aspect of who else played on that same day, and gives you the long run expected value for the scores that were posted (allowing fail comparison and ranking), where as that isn’t something you could do with divs received or PB scores/averages in isolation.
Akanji posted more scores in the winning score range. You can see from below that he has one more score in the >200 range.
Fully with you now. I get how it’s being calculated. Thanks for that. So similar to poker maths.
He’s also played 6 more games than Carvajal also which helps. Exactly (like poker maths). Weighted average probability of winning the individual divs
Didn’t get that it was being done on a game by game basis before you showed me those screens. Thanks a lot. Very useful stat indeed.
It’s a nice stat. Divs received mean nothing due to the variability/luck aspect. Scores do mean things, but aren’t rankable due to day types/positions, etc. Is a 180 score by a DF on a single day better / worse than a 195 by a FWD on a triple day? Hard to tell. by putting the value on things we make it rankable/comparable and can calc implied yields vs the price
Great info. Most of this I had a slight grasp of but it’s good to know the full run down and complete understanding
The poker analogy is great, It really does boil down to, with this hand (ie, the scores), what’s the probability if it winning. As we know the div structure, we can take it further and turn those probabilities into rankable $$ values
That stat alone attracted me to IndexGain. Because I like poker and so I get implied probabilities. When Implied PB was talked about on FIG I signed up straight away.