Testing Services

[Myhorse1_X]

Does anyone out there know of a company that will test an algorithm against a database of races for a fee??

It seems to me that there was one out there that would test an
algorithm for you against a database of 20,000 races for a reasonable fee.

MyHorse1

[Bob Harris]

Jim Bayle in Las Vegas was offering a service where he would run a customer's system thru his database for a fee...I don't remember how much or if he still does it. His e-mail address is mail@sportstat.com

[TG44]

I have a database of 50,000 odd races.

Could test for you depending on the factors you are looking at. If it's a simple enough query I'll do it for free!

[PMANN1]

I've tried to contact Jim Bayle twice recently. I had spoken with him a couple times in the past. The last time I talked with him a few months ago he was moving on to other things.
If you hear from him let me know if he's still doing queries.
Maybe he's just been on a long vacation?

[Topcat]

i understand Jim Bayle doesn't do queries for the public anymore
but manages his profitable longshot service

[PMANN1]

TG44,
What type of speed and pace ratings do you have in your db? I may have you do one or two things for me if the price is right and you are willing. An example would be:
I've heard a lot recently about averaging the last 2 races and the last 5 races. If the average of the last 2 races is less than the average of the last 5 the horse is a false favorite. If it's true this could be very valuable information if you knew what conditions it did best in. You could also try last 2 or last 2 on todays surface within 1 furlong (whichever rating was greater).

[TG44]

can't say I believe much in the last 2 vs last 5 theory:

Odds<3
Must have 5+ races

Ave last 2<Ave last 5
6629 winners - 22139 horses
Win: 29.9% 0.81 ROI (per $1)
Place: 50.5% 0.86
Show: 65.6% 0.89

Ave last 2>=Ave last 5
14339 winners - 46642 horses
Win: 30.7% 0.82 ROI (per $1)
Place: 52.4% 0.87
Show: 67.4% 0.90

could expand the query to like 3 SR point difference in average minimum, but doubt there is much to it

SR's are Bris

[Myhorse1_X]

TG44:

Can you try:

Best 2 in last 3, plus best one from 4-10 for me???

Average the three numbers.

My Horse1

[Dave Schwartz]

Interesting concept... L2 vs L5

We are using something like that but Final Time will show nothing. It has to be based upon some analysis of energy use. Some races may have slower final times but were actually better races.

Regards,
Dave Schwartz

[kitts]

TG44-
I am curious about "repeaters"-horses that win again--

Like how likely is a horse that won to win again
How likely is a winner of 2 straight to win again
How likely is a winner of 3 staright to win again

I recall reading somewhere that the odds of a horse making three in a row are 12/1 and I would like to verify this.

I have no idea if this is easy or difficult but if you can do it (for free or fee) just let me know.

Thanks.

[andicap]

Can anyone test factors based on a time period. E.g. Best Beyer in last 45 days. As well as final odds.
That is, 2nd best Beyer in last 45 days at 6-1 or over.

[TG44]

all right, quite a to-do list developing.

for Myhorse1

Your query probably needs refiining as you don't tell me how to handle situations where a horse only has 3 starts or a race where 3/4 of the horses are first time starters and so on.

Here's the results though:

best 2 SR's of last 3 plus best SR from races 4-10
(if no 4-10 race counted as zero)
50,000+ races

highest rated (no ties)
24.7% winners 0.86 ROI

the results could be quite a bit better I would think if you eliminate the races with lots of question marks and/or lightly raced horses and come up with some rules for how to handle null data

[TG44]

okay, Kitts, here's the numbers for horses off a win (and debunking the myth of the 1 in 12 for four in a row):

won last race (and maybe more)

16.2% wins 0.77 ROI
30.7% place 0.77 ROI
44.1% shows 0.78 ROI

Won 1 race in a row (eg didn't win 2 back or only 1 start)
15.3% wins 0.78 ROI

Won 2 races in a row
19.4% wins 0.72 ROI

won 3 races in a row
24.6% wins 0.73 ROI

[TG44]

ok andicap:

horse rated #2 in "best SR in last 45 days" at 6-1+ Odds:

7.3% wins 0.82 ROI

it occurs to me that perhaps you meant to use the horse with the #1 "second best SR in 45 days"

if so, let me know and I'll run it like that

[TG44]

dave schwartz, I am familiar with energy distribution, so if you can clarify for me the specifics of the query I can probably run it for you

[kitts]

TG44
Thanks for your prompt and informative response.

[PMANN1]

Thanks for the test. I've been wanting to learn access and how to do queries (if that's what you're doing) for some time. There are so many things to test for. It's hard to fit pure research time into my life with work, handicapping and family taking up so much time. Maybe I'll tackle it this fall when things slow down. You confirmed what I've found in a small sample on the L2 vs L5.
One more and then I'll leave you alone for a while (again I'm happy to pay). The search should be for non-maiden races and the horses should have 4 or more races lifetime in the U.S.
1) last race speed rating is 6 points or higher than any other race in the horses PP.
Results all and didn't finish within1 length last race.
2) Last race or last race within 2 furlongs was the horses highest speed rating by 0-5 points.
If today is a turf race the qualifying race must be a turf race.
Results a) all b) didn't win last c) didn't win last and qualifying race is one of the top 3 speed ratings or within 2 points of 3rd highest speed rating of all horses last race.
3) Same thing as #2 except the last race was the horses 2nd highest speed rating in it's past performance or within 1 point of it's second highest rating. (seperate within 4 points highest?)
I'm trying to test one aspect of what I've found on Thorograph, The Sheets etc. I have found in small samples that #1 should bounce, #2 is good in certain situations (2nd -4th off lay-off...) and #3 has been erratic. Thanks for your help. It's very greatly appreciated and I hope the results save us all some time and give some insight.

[Dave Schwartz]

TG44,

It is a concept we call the "power rating" and is somewhat proprietery. I have no problem disclosing it, but it would take a multi-page explanation on my part, scads of coding for you and a good set of par times.

It is probably not feasible.


Regards,
Dave schwartz

[TG44]

okay PMANN1, here's part I

last race SR 6+ points higher than any other horse in race
no maiden races
all horses in race have 4+ rated races (eg US/Canada)

1612 wins in 5509 starts
29.2% wins
0.83 ROI (on $1 bet)

Beaten lengths last > 1
618 - 2364
26.1%
0.82 ROI

[Topcat]

Originally posted by TG44
I have a database of 50,000 odd races.

Could test for you depending on the factors you are looking at. If it's a simple enough query I'll do it for free!

OK TG44,

If we haven't exhausted your patience I've got one for you.
Let's call it the best of the worst. The idea is to rate each horse
off of a "bad" race to reward a non-obviosu but consistent horse. Hopefully picking up in ROI what is lost in %.

For each race I'd like to see each horse rated off its second worst race out of its own past performances and compared to the rest of the field. We'll throw out the worst race to give some slack to those flukes that can heappen to any horse. So if a horse is showing Beyers of 88,87,88, 95,65,78,81,82,83,86 the race rated 78 would be he opne for comparison and so on for each horse int he field. After going through this "best of the worst" comparison the horse with the highest rating would be the selection.

And if I'm not really pusing it I would like to see the top two results for this approach.

Thanks
TC

[TG44]

topcat, I changed the query to only look at last 5 races. This may not be as good as using last ten, but I can limit it to races where all horses have 5+ races and still have a big sample. If you use ten races, there are many races where some of the horses don't have ten and then what do you do? plus some horses the tenth race back is like 3 years ago

by top two I assume you mean the two highest rated "best of worst" horses

"best of worst"
8038 - 34640
23.2% wins
0.88 ROI

"second best of worst"
5884-34640
16.9%
0.82 ROI


(PMANN I'll get to parts II and III soon, but they are a little more complicated and will require more than the few seconds to kick off the query that some of these do)

(Dave Schwartz - sounds interesting, but you're right that scads of coding wouldn't be good for me right now, I already have that with the regular job!)

[pmann2]

TG44
I wanted to get to you before it was too late. The first part was supposed to be horses who achieved their best speed rating last race by 6 points or higher compared to their previous races. It looks like you ran the queery for having a 6 point advantage over the other horses in the field. If I'm wrong please correct me. In parts 2 and 3 I hope I was a little clearer in the way I presented it. I'm looking at speed ratings of the horses last race compared to its own speed ratings in the PP. That's what The Sheets and Thorograph do. You don't have to run the first one again unless its super easy. I can't thank you enough for your help and time.

[Myhorse1_X]

TG44:

Thanks, I would think the easier format would be not to include Maiden Races if you could.
Or in your terms, eliminate null races.

I appreciate your efforts.

MyHorse1

[TG44]

oops, sorry pmann, bad blunder on my part. Yes I did run it as 6pts higher than other horses...

will do parts 1,2,3 tonight

[TG44]

ok, trying to make amends for the blunder:

horse with last race SR that is 6+ points better than any other race in its PP's

no maiden races
must have 4 rated races

1837-12434
14.7% wins
0.72 ROI

Place:
28.9%
0.76 ROI

Show
41.7%
0.75 ROI


-->Horse won or finished within 1 length last race
1245-7536
16.5%
0.75 ROI

-->Horse finished 1.25+ lengths back in last race
592-4898
12.0%
0.67 ROI


...looks like the Bounce is real, particularly with the 1.25+ finish. At long odds these types of horses are really terrible

[andicap]

TG,

1. SR is which rating??? DRF SR? BRIS??? Beyer???

2. Can you limit the last query so it just uses 4 yr olds and up. In 2 and 3 yr old races, it can be a good thing to have that improvement. Also, can you include a "time off" element. In other words, horses that improve by 6 pts over their best with little time off (say, 3 weeks or less) are theoretcially worse bets than horses that get freshened from their "top."
Any way to test this???

[PMANN1]

TG44,

That's very helpful. I've found these horses to be overbet and the do react. They can be particularly vulnerable when they've run their first fraction or second fraction 4 or more points higher than any other in their PP (using Bris). If you look at this years triple crown you'll notice a few examples in the Belmont. (It helped me key Saravan in the 2nd and 3rd spot--I didn't win obviously and swore at myself a little). The only exceptions I usually make are for 3 and 4 year olds off long rests that have had time to mature but I'm still shaky on these.
I hope the results help you in your handicapping. It'll help me be more confident when I bet vs. them even though they beat me once in a while. Curious to see that those that didn't win did worse. Maybe it was because they were chasing the other horses and exerted themselves to a greater extent than if they ran on their own while the winners ran at a pace under their own control. Now I'm really interested to see if you find a difference in the horses who progressed slightly in their new high speed number.

[GR1@HTR]

TG,

What if we looked at 1st time geldings who were beaten between 5 to 10 lengths in thier last race, are now dropping in class, only by trainers who won better than 12% within the last 365 days, who also showed early speed 2 races back, now switching from turf to dirt but not slop, with also a positve jockey switch, run only on weekdays begining with the letter T? well, ok never mind...

[ranchwest]

Originally posted by GR1
TG,

What if we looked at 1st time geldings who were beaten between 5 to 10 lengths in thier last race, are now dropping in class, only by trainers who won better than 12% within the last 365 days, who also showed early speed 2 races back, now switching from turf to dirt but not slop, with also a positve jockey switch, run only on weekdays begining with the letter T? well, ok never mind...

I've already got the results of that query here somewhere. I ran it last week. Let me see if I can find it....

[TG44]

for andicap

sr numbers are BRIS

last race horse's best SR by 6+ points, no mdn, must have 4+ starts, must be 4yr old or 4U race

276-1894
14.5% wins
0.68 ROI

days off breakout:

1-10 21% winners 0.88 ROI
11-20 17% 0.73
21-30 13% 0.58
31-40 15% 0.81
41-60 5% 0.45
61-90 13% 0.68
91-180 11% 0.53
181+ 6% 0.21

small sample size of course in some categories

[andicap]

Interesting.
What's the ROI if you exclude the 0.88 ROI data? We're getting pretty close to some good elimination dope.

[andicap]

actually, looking at it again, also nice to know the ROI just for 21+ days off. This is certainly not what you would expect. You would think horses with LESS rest would bounce more, but that's not the case. Horses off cyclical tops who come right back actually do BETTER than the track take.
I don't know if they are pairing up.

Interesting to know -- though probably beyond your db's capabilities when a horse has two races in a row that are 5 or 6 points better than anything in last 10.

[hurrikane]

Interesting stat overall. But not sure you could use them cold. Some things would have be checked to quantify the data. For instance.

What was the age of the horse. A 7yo is more likely to be raced into shape than a 4yo.
What is the 'class' of the animal.
GR1 horses may only run every 30-40 days simply because there aren't any races or they can earn thier feed in one race. A 5k claimer can't afford to lay down for 45 days. If he does there is likely something wrong. Especially if his last race was a good one. If he is ready to fire the trainer better have him in a race. Imho you can even remove the SR qualifier here and run cheaper horse that finished within 1 last. My db has shown cheaper animals that run a good race and then lay off are bad bets but not true for higher priced animals.

What is the ability of the trainer. Some of these guys can't get a horse ready to take a leak much less run well off a layoff.

Curious what the sample size was on some of this?

IMHO you have some good info here. Just no way to apply it universally.

[Sam-I am]

TG44,

Thanks in advance,
Please try this... top SR out of last 3 with "Best of the worst from 3-10 races" with BRIS power rating. Average the 3 numbers

Thanks again,
David

[Topcat]

[QUOTE]Originally posted by TG44
[B]topcat, I changed the query to only look at last 5 races. This may not be as good as using last ten, but I can limit it to races where all horses have 5+ races and still have a big sample. If you use ten races, there are many races where some of the horses don't have ten and then what do you do? plus some horses the tenth race back is like 3 years ago

by top two I assume you mean the two highest rated "best of worst" horses

"best of worst"
8038 - 34640
23.2% wins
0.88 ROI

"second best of worst"
5884-34640
16.9%
0.82 ROI

Thanks TG,

Your results are very similar to mine, although mine were run
on a much smaller sample.

My results show a slightly lower % but a slightly higher ROI.

By going back 10 races even if it was 3 years ago it puts one on the less obvious contenders. (for those with less than 10 lines I merely go back the maximum amount of racing lines shown- which "rewards" lighly raced horses)

This is somewhat of a cntrarian approach that says let's not do what everyone else is doing which is looking at recent lines and the "best" races. It does put you on horses that tend to run more consistently and figure to be there.

Oh yes, by top 2, I meant the top two "best of worst horses"

Your results confirm what I have found that just starting with this
best fo the worst beats the track take. Filter for type of races and put in few qualifiers like trainer win % and you'll find profitable subsets.

Thanks again for your generosity in runnng the query.
TC

[PMANN1]

TG44,
I'm not trying to rush you but I got to thinkin which is where the trouble usually starts. Instead of doing a queery like I listed in the 2nd and 3rd parts of my post you could do something a lot simpler when you have the time if you're so inclined. Simply run last race best in the horses PP (compared to itself) by 0-5 points for part 2 and last was the 2nd best race in the horses PP for part 3. All horses have 5 or more races in the US. That way the results could be compared to part one (6 or more points higher) and all parts to each other to see if there's much difference. Breaking it down by days (0-27, 28-44 and 44+) was great to look at and age also. When you have the time it would be appreciated and this way would save you some time and work.
Thanks again.

[TG44]

hey PMann1,

I still plan to do the tests for you, just no time right now as I'm busy with this football site I work with. Maybe in a week or two it will all have calmed down and I can get back to it.