Post

For anyone interested, here are some results from a study I did a few years ago Each dot represents projected pts yielded from 50 play-throughs of randomised versions of @fplreview.com's 20/21 data, with a decay rate (α) and FT value (β) selected using Bayesian optimisation (8 GW horizon)

I never shared these results in detail bc I think they do more harm than good if taken as gospel. My general thoughts are - Any rate 0.8-0.9 is useful - The "optimal" rate probably depends on game state - The best use for decay is to try various rates and learn from the differences between solutions

These results are from just one season, and before recent rule changes I don't think the question of optimal decay rate is closed, but I don't really think it's worth pursuing - it's very boring and noisy and there are other way more interesting & impactful things to study in FPL

The xMins were decayed as usual, so in my opinion decay rate performs some function besides to account for mins decay (cc @fplfutbolmetrix.bsky.social 😜)