Selections April 19

[traynor]

Blue Chip Selections - April 19
Greyhound, Southland, 04/19/2006, Race 4 - #4 Opalescent
Greyhound, Dairyland, 04/19/2006, Race 11 - #6 Westvirginia Bop
Greyhound, Dairyland, 04/19/2006, Race 13 - #7 Jit Dodge Meteor
Greyhound, Victoryland, 04/19/2006, Race 1 - #7 Bget Jim Jim Jim
Greyhound, Victoryland, 04/19/2006, Race 3 - #8 Gucci Rush
Greyhound, Victoryland, 04/19/2006, Race 5 - #1 Bee Bee
Greyhound, Victoryland, 04/19/2006, Race 9 - #7 Flying Plaintiff
Harness, Balmoral, 04/19/2006, Race 1 - #2 Point Of Attack + exacta 23
Harness, Balmoral, 04/19/2006, Race 2 - #8 Dazzling Artie
Harness, Balmoral, 04/19/2006, Race 4 - #5 Miss Alyssa + exactas 53 & 54
Harness, Balmoral, 04/19/2006, Race 5 - #3 Kabanero
Harness, Balmoral, 04/19/2006, Race 6 - #2 Free
Harness, Balmoral, 04/19/2006, Race 7 - #4 Rita Hall
Harness, Balmoral, 04/19/2006, Race 8 - #4 Chevie Nitrous
Harness, Balmoral, 04/19/2006, Race 9 - #3 Oakie's Ride Witme
Harness, Balmoral, 04/19/2006, Race 10 - #7 Collegezone Dotcom
Harness, Balmoral, 04/19/2006, Race 11 - #1 Exit Hanover
Harness, Balmoral, 04/19/2006, Race 12 - #5 Yo Ho Rojo

Selections for Cal-Expo will be posted Wedneday morning at
http://www.tekwrytrs.com/Home/FreeSample.aspx

All selections based on positive expectation model; minimum
33% win percentage, minimum 125% ROI over time.

Free Selections:
for Greyhounds April 15 through April 30
for Harness Races April 15 through April 30
for Thoroughbreds May 1 through May 15

Good Luck

[traynor]

Selections for Cal-Expo are being posted now at
http://www.tekwrytrs.com/Home/FreeSample.aspx and should be available shortly.
Good Luck

[Koko]

Traynor,

I took a look at your site and read through it in full. I found your descriptions quite understandable in spite of not being highly versed in programming, and other related subjects. I commend you for making it as clear as possible to the "layman".

I do have two questions if you don't mind. I understand the concept of not using older data in spite of it's ability to give you a larger data base. Apparently you're convinced that enough of the conditions that dictate where the value lies in the pools changes dramatically enough that the old data becomes a liability instead of a help. It would seem to me that the betting habits of the public would tend to be fairly stable over long terms, at least on a track specific basis, so I'm presuming that track conditions or other factors are the ones that are most prone to change over time, more dramatically, is that correct?

Assuming your approach makes sense, to not use older data and having fresher but smaller databases to work with, doesn't the relatively small databases of a few hundred races or less, of specific grades, tracks etc. cause the confidence level to be affected greatly?

[traynor]

Koko wrote: <Assuming your approach makes sense, to not use older data and having fresher but smaller databases to work with, doesn't the relatively small databases of a few hundred races or less, of specific grades, tracks etc. cause the confidence level to be affected greatly?>

I think the key point is in "layering." Remember that the "few hundred races" are very specific to track, distance, and class level. While there may be some grouping of adjacent class levels, the layers are mostly discrete (separate). The model size is not arbitrarily selected; it is what we have found to be the most predictive, and is essentially a compromise between the "big bucket" of all races for the season (or year) at a track and the "small bucket" of what the pace handicappers generally call the "Brohamer Model." One is too big, the other too small.

We keep a lot of date-coded samples for comparison. That means we use a model in June, and compare the results of that model with a similar model developed from the whatever-number-of-races a month previously, two months previously, etc. That allows us to detect anomalies, or short runs that are atypical, or even models that are totally whacked.

For example, I posted selections for a model at Wheelng that was kicking out 143% ROI, longest missout 7 races, win percentage over 40. A few days after I started posting the selections, it went into an 18 race missout--the longest in more than a year of models for Wheeling. That missout was followed by several back-to-back wins, a short missout, several more wins, and now is back to "normal." When I compare the model for March 15 to April 15 with the model (using the same number of current races) for February 15 to March 15, it looks (approximately) the same.

While we set up our "handicapping" apps to automatically truncate after a specific number of races, and the modeling apps to only use a specific number of races, we are fairly confident that the approach is sound, and that it is considerably more predictive (based on a LOT of alternative approaches) than using a larger base of races.

One of our primary reasons for using the limited model is that we correct for outliers. That enables us to use a smaller model without the distortion that would be created by one or two unusually high payoffs. (The LAST thing you want is to bet into a misleading model that depends on a few high mutuels for profitability.) By truncating payoffs to one-and-a-half times the "mean of the interquartile range" we can use smaller models without the distortion that would otherwise be introduced.

Lastly, the betting patterns of the public often go through dramatic and substantial shifts. While generally the same over time, certain factors such as an infusion of winter visitors at Florida tracks create unusual situations that can corrupt a model if mixed in carelessly with other results.
Good Luck

[traynor]

Apologies for missing part of your question. I think it is the "few hundred races" that is misleading. That is, a track, distance, and class specific "layer" of a sample may include several thousand races, representing (for thoroughbred and harness races) two to four months, and about six to ten weeks for greyhounds (that frequently race at four day intervals). Each layer may be 150 - 200 races, in an overall sample of 2000 - 3000 races.
Good Luck

[Koko]

Traynor,

What simulcast dog racing options might you know of for California Residents?
I know that the few books in the states aren't allowed to offer accounts to CA. residents, and I haven't found an offshore book that does either. I think there was one I found on Rosnet2000 that might have been offshore if I remember correctly, but they wouldn't serve CA. residents either. Also, what's the liklihood (like the Pinny mysteries) that if you're beating them regularly that they'll dump your bets into the mutual pools from your experience, if you have relevant experiences to share?

[traynor]

Greyhound, Southland, 04/19/2006, Race 4 - #4 Opalescent [typo error on name - #4 won at $8.60]
Greyhound, Victoryland, 04/19/2006, Race 1 - #7 Bget Jim Jim Jim won at $4.20
Harness, Balmoral, 04/19/2006, Race 5 - #3 Kabanero won at $6.00
Harness, Balmoral, 04/19/2006, Race 6 - #2 Free won at $5.20
Harness, Balmoral, 04/19/2006, Race 7 - #4 Rita Hall won at $4.40
Harness, Balmoral, 04/19/2006, Race 8 - #4 Chevie Nitrous won at $3.40
Harness, Balmoral, 04/19/2006, Race 9 - #3 Oakie's Ride Witme won at $3.60
Harness, Balmoral, 04/19/2006, Race 11 - #1 Exit Hanover won at $6.20
Harness, Cal-Expo, 04/19/2006, Race 5 - #6 Jordan's Planet won at $23.20
Good Luck

[traynor]

There may not be any, aside from Australian or UK racing, and even that may be interpreted as "illegal." It seems best to avoid offshore books, because of ambiguous legislation regarding legality. Specifically, we do not use offshore books at all--everything goes through tracks or track-affiliated sites. One alternative is to spend more time in Nevada. That is not meant to be funny; some people do well betting only on weekends, and use that as an excuse to spend almost every weekend in Nevada. Something like Andy Beyer's "immersion" at Gulfstream years ago.
Good Luck