PREDICTING HORSE RACE WINNERS THROUGH A REGULARIZED CONDITIONAL LOGISTIC REGRESSION WITH FRAILTY

Conditional logistic regression has remained a mainstay in predicting horse racing out- comes since the 1980’s. In this paper, we propose and apply novel modifications of the regression model to include parameter regularization and a frailty contribution that exploits winning dividends. Additionally, the entire model was fit using state-of-the art parallelization methods on commodity graphical processing units. (GPU) The model is trained using 4 years of horse racing data from Hong Kong, and then tested on a hold-out year of races. Simulated betting produces a return on investment significantly higher than any other published methods involving Hong Kong races.