Making a Model that Can Make Millions

Is it possible to construct a mathematical model of the performance of a golfer, a boxer, a football team, a cricket team, a snooker player, a horse, a dog, or whatever, that would predict well enough to allow us to earn a systematic profit over time?

The first problem is that any sporting event is influenced by random factors, which statisticians call ‘noise.’ In any given situation, this noise can overwhelm the best model to generate an unexpected outcome. That’s the bad news. The good news is that these factors tend to balance out over time, so any properly devised forecasting model which takes into account those factors which are predictable has the potential to perform very well indeed. Such models are known in the trade as ‘fundamental’ handicapping strategies, because they are based on fundamental information about performance.

The question is whether such a system exists which can actually turn a profit, or indeed whether it has ever existed. Ask Bill Benter how he made his millions at the Hong Kong racetrack and you have your answer. Basically, he constructed a computer model designed to estimate current performance potential. This involved the investigation of variables and factors with potential predictive significance and the refining of these individually so as to maximize their predictive accuracy.

In doing so he employed state-of-the-art econometric forecasting techniques, which he was confident enough to summarize in a classic paper entitled ‘Computer based horse race handicapping and wagering systems: a report.’ The basic question he seeks to answer in that paper is whether it is possible to construct a forecasting model which can generate a systematic profit at the races. Ever the practitioner, he provides the answer by constructing just such a model.

His method is to identify each individual factor that could possibly predict the outcome of a race and then to whittle these down to the most reliable and effective. Once he had a model that worked on past data, he tested it ‘out-of-sample; i.e. on a large sample of further races.

Sometimes he found that a variable was useful in predicting race outcomes but he really couldn’t understand why that should be the case. In such circumstances, he decided the best policy was not to care. Faced with a choice between a profitable model that he couldn’t fully explain, and an unprofitable model that he understood perfectly, he chose the former. Ideally, of course, you would work with a model that both works and you can fully explain, but Benter’s bottom line is that if it works, it doesn’t need fixing.

So what is that bottom line? Well, Bill Benter is now a very rich man even by the standards of most very rich men, and he made it on the basis of a sophisticated forecasting model which overcame track takes of about 19 per cent. In a masterly piece of understatement he concluded in his paper that “… at least at some times at some tracks, a statistically derived fundamental handicapping model can achieve a significant positive expectation.” Significant indeed! Why not try it some time?