Abstract
This paper introduces the first Expected Possession Value (EPV) benchmark anda new and improved EPV model for football. Through the introduction of theOJN-Pass-EPV benchmark, we present a novel method to quantitatively assess thequality of EPV models by using pairs of game states with given relative EPVs.Next, we attempt to replicate the results of Fern\'andez et al. (2021) using adataset containing Dutch Eredivisie and World Cup matches. Following ourfailure to do so, we propose a new architecture based on U-net-typeconvolutional neural networks, achieving good results in model loss andExpected Calibration Error. Finally, we present an improved pass model thatincorporates ball height and contains a new dual-component pass value modelthat analyzes reward and risk. The resulting EPV model correctly identifies thehigher value state in 78% of the game state pairs in the OJN-Pass-EPVbenchmark, demonstrating its ability to accurately assess goal-scoringpotential. Our findings can help assess the quality of EPV models, improve EPVpredictions, help assess potential reward and risk of passing decisions, andimprove player and team performance.