Remodelling Exercise

The reason I have not blogged for a while is due to the fact that most of my time has been absorbed dealing with all the stuff life throws at you interspersed with betting and trying to grind out a profit for my efforts.

Those of you who have been following my ratings have noted a steady improvement in their accuracy, although I am still not satisfied.

For a while now I have been trying to get my head around William Benter’s seminal paper “Computer Based Horse Race Handicapping and Wagering Systems” (Efficiency of Racetrack Betting Markets) where he outlines a practical and successful computerised horse race handicapping and wagering system he developed. I’m not a mathematician or statistician so I needed to develop a basic understanding of these skills to develop my own version of his model; hence another reason for not blogging recently.

One major area of discontent with my model was the accuracy of the final rating and how it translated to winning probabilities. This is function of the limited factors I was using in my model and as such I had many failed attempts at creating a semi accurate betting forecast which also factored in the favourite longshot bias. The majority of computer generated tissues do not take account of this bias and in my mind fundamentally flawed.

In addition, the other area of contention was that even with most sophisticated model on the planet, I could not hope to include all handicapping factors into the model that out performed the crowd’s powers of prediction thus

The chart is old however you can create a modern day version if you take horse racing results over a significant period and create a scatter plot of observed proportion of races won by horses verses the probability implied by their betting odds. You will see that the “fit” is remarkable; an R squared of around 0.98. What does this chart mean to non stats junkies?

The Public Is Very Good At What It Does
Not only can the public tell the difference between two low odds horses say a horse that should be even money (1/1) versus a horse that should be say 2/1 we might be able to do that at a decent level, but it can tell the difference between high odds horses say a horse that should be 20-1 versus a horse that should be 25/1.

Now how do you do that? They are indistinguishable in the form yet the public can separate them in the long run. The horse that is 20-1 will win more than the horse that is 25-1.

If you are not convinced by the predictive powers of the crowd take a moment to watch this video.

Benter recognised the public’s implied probability estimates generally correspond well with the actual frequencies of winning. He realised that what was needed in his model was a way to combine the judgements of two experts, (i.e. the fundamental model and the public).

After thinking and modelling different approaches to absorb pre- race market opinion into my own model, I have decided to implement one from today and try it out over an expended period of time using real races. I have the old model to revert to if everything goes Pete Tong, although my gut feel tells me that I must embark on a modification of the current model to increase its accuracy.

Related Posts

• No Related Post