Competing in daily fantasy sports using generative models

Abstract

AbstractWe introduce a model for competing in daily fantasy sports competitions. Our approach combines modeling of players' fantasy points and subsequent lineup optimization. The distribution of fantasy points was estimated using an ensemble of generative models. We designed a mixed‐integer quadratic program to optimize not only the average fantasy points but also the variance and covariance of players' fantasy points to exploit the top‐heavy payout structures. We tested our approach on real‐world data. Over eight weeks, the model generated a profit of €2439 with an initial investment of €7133, resulting in a return on investment of over 34%.