Creating automated oddsline

[harnesslover]

Does anyone currently create an automated oddsline using fractions or speed figures? I am not so much interested in the exact criteria of the automated oddsline, rather the math of it.

Is it possible to accurately create a decent oddsline based on fractions or pace figures alone? If so, how would one do this?

[Overlay]

If you were using just the one factor, I would think you would want to find the pattern of figures or fractions (such as last race only, an average of recent races on the same surface, or whatever) that produced a smooth flow of results in terms of winning percentage or impact values based on ranking the horses in a field from top to bottom using that pattern, and that also produced the most pronounced effectiveness (the largest disparity) in differentiating the top-ranked horses in the field from the bottom-ranked. Then base your odds line on the winning percentage associated with each rank.

Although I think that using any type of odds line associated with a factor is better than just playing the factor blindly with no regard to the odds of the horses involved, using only one stand-alone element for that purpose might result in not isolating enough overlaid horses to produce an overall profit, since there would be a greater probability of too many other bettors keying on that same factor.

[prank]

You can get started with logistic regression (or, as some call it, "ordinal logistic regression"). From there, you're on your way. Then, try Benter's article "Computer Based Horse Race Handicapping and Wagering Systems: A Report".

From that point you'll have learned most of the math fundamentals.

Actually, starting with Benter's article isn't a bad idea, then you can work out all the unfamiliar things in one go.

PRank

[levinmpa]

Do a search for CJ's oddstool. It's an excel sheet that allows you to enter up to three figures of your choice, weighing them the way you want. It will go through it's calculations, simulating 1000 races, and create an odds line.

[jfdinneen]

While adding my voice to Overlay's wise counsel on the dangers of projecting from a single factor and endorsing Prank's advice on following Benter's guidelines to integrating additional variables into the public oddsline, I have outlined below an adaptation of Steven Skiena's player superiority probability calculations from Calculated Bets (http://www.amazon.com/Calculated-Bets-Steven-Skiena/dp/0521009626) that provides a realistic oddsline. Though it is by no means a mathematically optimal solution, it is a satisfactory one.

P(A,B) = (1 + [Rating(A) - Rating(B)]^alpha) / 2, if Rating(A) >= Rating (B)
P(A,B) = (1 - [Rating(B) - Rating(A)]^alpha) / 2, if Rating(A) <= Rating (B)

Alpha is a 'fudge factor' for fine tuning the probability function and you need to determine its value from your records or by trial and error. For Jai alai, Skiena uses 0.4.

Note the scores column is calculated from the product of the row probabilites mutiplied by 1000 and that the projected oddsline is an approximation.

Best wishes,

John

[richrosa]

This looks real good, but I'm a little lost.

How do you get the numbers in the grid? I can't seem to reconcile the math in this example.

[jfdinneen]

Rich,

First, I include two examples below with different Alphas (1/e, 0.23) to illustrate the effect of the fudge factor on the overall shape of the probability function.

Second, with respect to the grid calculations, the following example should help clarify any confusion (apologies):

For Alpha = 1/e = 0.37 (approx)

IF Rating(A) >= Rating (B),
THEN P(A,B) = (1 + (Rating(A) - Rating(B))^alpha) / 2,
ELSE P(A,B) = (1 - (Rating(B) - Rating(A))^alpha) / 2

P(1,2) = P(105,96) = (1 + ((105/809 - 96/809)^Alpha)) * 0.5 = 0.595552 (Excel)
P(2,6) = P(96,99) = (1 - ((99/809 - 96/809)^Alpha)) * 0.5 = 0.436215 (Excel)

Finally, in the original posting I corrected the projected odds to the nearest odds fraction for visual impact. On this occasion, I have given the original values.

Best wishes,

John