PB Power Rankings: 30-day simulations (Sep 22nd)
Introducing the PB Power Rankings
The past month has proven to be a tricky and volatile trading environment for all sorts of reasons. The outsized IPD payouts have acted a bit like Sirens to the market, offering insane value, but also distorting PB strategies as capital flowed into these shorter-term trades. I‘ve been enchanted myself, and ultimately burnt on afew occasions (Di Maria 🤦🏻♂️) so looking forward to default back to my tried and tested strategies - PB simulations. I expect the market to quickly flip back and focus here too, given the adjustment in intrinsic value we’ve witnessed after the flash crash and only partial correction. Expected yield on offer at these levels is off the charts.
I’m going to provide simulation results as the ‘PB Power Rankings’ going forward, where all players in all positions are ranked by the expected 1-month dividend return. I might push that prediction window out further in the future (let me know what you think in the comments) , but the models rely on the fixtures being accurate (since that massively affects probability of winning and therefore value). As a reminder, expected value here is probabilistically estimated using Bayesian inference (see below for how the model works)
So for the next 30-days, what do the PB models predict in-terms of simulated probability and therefore expected value?
The GOAT is head and shoulders above everyone. Man City aside, he retains real value at these prices and payouts.
RDP - Unclear what the situation with Leeds is, so steering clear for now, but his PB suitability is unquestionable.
Kimmich - Planning some follow up analysis on his transformation at CM. I’ve been steadily topping up over the past few weeks after some undeniable evidence of PB improvements. Didn’t see much intrinsic value at his price + div payouts when he was initially switched, his simulated chances of winning dropped dramatically. But that has picked back up in recent games and he’s clearly improving on most metrics and will benefit from Thiago leaving.
Barrow - Big fan of Barrow given his age and score distribution (see below - this is one of the main factors that ultimately drives the likelihood of winning divs)
Renato Sanches - Another Machine favourite. He hits peaks, even more so than Barrow (see below again) but faces greater competition as a Midfielder.
Lewandowski - Not a high base PB player, but of course capable of huge scores (peak of 320), playing as the talisman of a team in incredible form.
Jonathan Bamba - First time he’s ranked highly in my simulations, benefiting from a recent upweighting in the model given his big PB improvement since the restart (see below)
Suprising one here - Dominik Kohr. Looks like it’s driven by one Bronze gameday, so not putting a huge amount of weight on this, but he has won PB before fwiw
Cristiano - Held a lot during the IPD madness, and retain a significant amount for pure PB.
KDB - PB God, massively undervalued at this price.
One other thing to note - the drop in Trent (who barely makes the top 30) is quite worrying, the model has certainly picked up this decline in PB output which has been flagged on the timeline (Note Robertson rising above him now…). I’m still a big believer (and holder) though, I don’t think this is a structural change (he doesn’t look fully fit even) and at 21yo retains a huge amount of intrinsic value. Waiting for an upturn in form.
30-day Power Rankings
Modeling Approach:
The models run thousands of simulations for individual player PB scores, modeling each gameday score as a function of player, team, location and opposition effects, trends in form and many other variables. When each gameday passes, it updates the model parameters again (a process called Bayesian data analysis) meaning predictive accuracy increases over time. The overall goal of these models is to probabilistically estimate player dividends over a future fixture period, and therefore implied player value. PB is the core of the product, so my valuation methods are focussed on those.
Underlying data sourced from Index Edge. Modeling, simulations and analysis done by Index Machine
Hi, thanks for this, great read! Could you please advise why KDB is great value? Is the value for expected divs on your table the expected divs in pence, or is it a probability of winning? Excuse the question, new user and first time seeing this analysis. Looks like a cracking bit of work though!
Why is Neymar so low down (relatively) on the project list of returners for the next 30 days? Looking at last season's PB data, Neymar scored peak (i.e. PB winning worthy) scores at a greater frequency than Messi?