belmont3 Wrote:
> the Backyard Boys include Bud Fox and he
> has \'inside\' info that Mathcapper took the last
> train from Clarksville, had some bad chicken at
> Hatties, warmed up badly and begged Uncle Bill to
> be scratched.
>
> The Backyard Boys bet early placing their 65k on
> the nose of the Weatherman.
>
> Moby Dick\'s Orcaputer naturally spots the huge
> disparity in odds and plucks his $62,500 (or more
> in theory) on Mathcapper.
>
> Mathcapper spits the bit and the Weatherman romps
> home.
>
> Moby Dick gets beached
Moby should\'ve factored into his algorithm the evidence from my past performances, which showed that when I veer into that netherworld of nasty Hatties chicken and away from my standard diet of hard boiled eggs, I bounce badly next time out.
(loved the Bud Fox reference btw Bob - I can still recite just about every line of Gekko\'s, including the classic Teldar Paper speech).
> The questions raised are:
>
> At which tracks and which meets do whales operate?
Pretty much everywhere these days. They started in Hong Kong, which was fertile ground because of the huge pools and the self-contained horse population, but they\'ve long since expanded into the N.American market. The limiting factor for them is track handle - smaller tracks may not be worth their time.
> What \'pools\' do they typically bet into?
All pools. Basically any bet they\'ve computed to have a positive expectation. The natural tendency is toward the superexotics (both vertically and horizontally) because edges are multiplicative. I posted a little bit about this not too long ago:
Computer-Robotic Wagering
> Who creates the software and the algorithm they deploy?
There are many imitators now, but the original, and most widely known, was created by Bill Benter and Alan Woods. It took them 5 years and over a million lines of code.
> Are performance figures (such at TG\'s)
> incorporated into their algorithm?
I can only speak for the Benter/Woods model, but not performance figures that I\'m aware of, although I do believe they incorporate some sort of variant-adjusted final time. Benter\'s model uses something like 80 factors, which include \'native ability\' variables like normalized finishing position, race competitive level (ie. purse), final time , etc., along with \'preference\' variables like distance, surface, track, condition (wet/dry, hard/soft), etc. and \'incidental\' variables like weight, post position, trainer, jockey (normalized finish position or win%), etc. All variables are in the form of exponentially recency-weighted averages.
> How does an algorithm handle claims, trainer
> changes via private transactions, equipment
> changes, medication changes, feed chages (or how
> does the computer find out about the Weatherman?)
> jockey changes, surface changes, distance changes,
> etc. etc, etc. (How does the computer avoid
> Mathcapper?)
This was the biggest issue Benter/Woods ran into, and it caused them to go through their initial bankroll at least once before they were able to correct the problem.
When the first developed their program, they found that there was a very big bias in their estimates. The horses they deemed as overlays were not winning at the rate they predicted, they were underperforming badly.
What they discovered was that they needed to incorporate the public\'s probability estimate into their model (to account for the so-called \'smart money\'). When they combined the probability estimate from their fundamental model with the public\'s probability estimate (50% weighting to each input), they found that they now produced a set of probability estimates that showed very good correspondence to the actual frequency of winning. And the rest, as Benter says, was history.
> I find it difficult to imagine that such highly
> subjective intangibles could be incorporated into
> a program since, by their very nature, they are
> subjective.
They do have to quantify some of the more subjective \'preference\' and \'incidental\' variables. As they do with their other variables, I believe they use a normalized scale on the order of -0.5 to +0.5 or something like that.
> My feeling has always been that:
> I see the odds pool (exactas, DD\'s WPS etc.)
> before I bet
> I can see the Will pays going one race forward
> (pik 3 etc.)
> I have a pretty good notion of what the return
> will be based upon my wager.
>
> If there is perceived value you bet.
> If there is no value you pass.
Exactly. That is one of the reasons I always compute the odds based on the Will Pays, and then measure my expected value based on those odds, which are generally pretty close to the final odds. I\'m rarely surprised by a horse going from 3-1 in the gate to 9/5 down the backstretch. In fact, if the odds based on the Will Pay for such a horse was 9/5 and his odds
didn\'t drop, then I would be surprised, and likely in trouble if I bet him, as the lack of support in the win pool vis-a-vis the Will Pays is usually a bad sign.
> I suppose that the Orca may eating some of my
> profits but i wager based upon my perceived value
> proposition.
Yep. Benter himself estimated that the computer guys have the combined effect of raising the blended takeout by around 2 percentage points. So if a track\'s blended take is, say, 20%, the average handicapper would feel like he\'s playing against a 22% takeout instead.
Since it\'s a zero sum game, the fact that the computer guys are extracting this money from the pools makes it a little harder for everyone else, but as I mentioned in a recent post:
Odds Lines: Omniscient vs. Zero Knowledge, as long as you can create an odds line that is better than the public\'s (which is the same thing the computer guys are doing), and your overlays are performing like they should, there is still value to be had.
Best,
Rocky
