Adapting support vector machine methods for horserace odds prediction

ABSTRACT

The methodology of Support Vector Machine Methods is adapted to enable the analysis of stratified outcomes such as horseracing results. As the strength of the Support Vector Machine approach lies in the apparent ability to produce generalisable models when the dimensionality of the inputs is large relative to the the number of observations, such methodology would appear to be particularly appropriate in the horseracing context, where the often number of input variables deemed as being potentially relevant can be difficult to reconcile with the scarcity of relevant race results. The methods are applied to a relatively small (200 races in-sample) sample of Australian racing data and tested on 100 races out-of-sample with promising results, especially considering the relatively large number (12) of input variables used.

REFERENCES

• 1: R N Bolton , R G Chapman, Searching for positive returns at the track: a multinomial logic model for handicapping horse races, Management Science, v.32 n.8, p.1040-1060, Aug.1986 [doi>10.1287/mnsc.32.8.1040]
• 2: Edelman, D. (2001), The Compleat Horseplayer Sydney: De Mare Consultants
• 3: Platt, J.(1998), 'Sequential Minimal Optimisation: A Fast Algorithm for Training Support Vector Machines'. Microsoft Research Technical Report MSR-TR-98-14
• 4: Vladimir N. Vapnik, The nature of statistical learning theory, Springer-Verlag New York, Inc., New York, NY, 1995