Comparing the Effectiveness of One- and Two-step Conditional Logit Models for Predicting Outcomes in a Speculative Market

This paper compares two approaches to predicting outcomes in a speculative market, the horserace betting market. In particular, the nature of one- and two-step conditional logit procedures involving a process for exploding the choice set are outlined, their strengths and weaknesses are compared and their relative effectiveness is evaluated by predicting winning probabilities for horse races at a UK racetrack. The models incorporate variables which are widely recognised as having predictive power and which should therefore be effectively discounted in market odds. Despite this handicap, both approaches produce probability estimates which can be used to earn positive returns, but the two-step approach yields substantially higher profits.