KENOSHA

In 2019, a paper by Bill Benter sent me down a rabbit hole. He had built a profitable horse-betting system using a statistical model to estimate each horse’s chance of winning. Could a similar approach still find an edge in another market, decades later? That became Kenosha: a personal research project adapting Benter’s multinomial-logit approach from Hong Kong thoroughbreds to European harness racing. I suspected any edge would be difficult to find: the data and computing tools available to me were available to other bettors too. To be profitable even in principle, a strategy would have needed to overcome both PMU’s deductions from the betting pool and Germany’s then-applicable 5% betting tax. I wasn’t about to quit my job over it. Building the whole thing nevertheless sounded like a blast: collecting and cleaning data, constructing features, testing models, and bringing live predictions into a race-day dashboard.